From 8521c33f82498a0422a9814d2a6de6324d3cc5ba Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:08:29 -0700 Subject: [PATCH 01/71] docs(research): capture intelligence without omniscience synthesis for review --- .../README.md | 402 ++++++++++++++++++ 1 file changed, 402 insertions(+) create mode 100644 research/intelligence-without-omniscience/README.md diff --git a/research/intelligence-without-omniscience/README.md b/research/intelligence-without-omniscience/README.md new file mode 100644 index 00000000..4a46db0a --- /dev/null +++ b/research/intelligence-without-omniscience/README.md @@ -0,0 +1,402 @@ +# Intelligence Without Omniscience +## Generality, situated capability, distributed cognition, epistemic justice, and an intelligence commons + +**Document role:** Bounded, cross-disciplinary research synthesis and question map. +**Status:** WORKING / AI-assisted draft / author review pending (2026-10-07). +**Claim:** Not a validated theory, a new psychometric construct, an established "intelligence commons" institution, or a conclusion that distributed intelligence equals AGI. +**Scope:** Connect established research without replacing the fields, methods, and projects that own particular mechanisms. +**Public boundary:** This is a public-safe research draft; source dialogue is paraphrased, not reproduced as a personal transcript. No private project materials or story canon are imported. + +**Idea Trails:** Intelligence, Automation & Legitimate Authority; Accessibility, Dependence & Participation; Commons, Ownership & Shared Capacity; Collective Judgment, Dissent & Manufactured Consensus; Memory, Provenance & Continuity; Adaptive Continuity, Agency & Becoming. +**Trail role:** research + + + +> What if one way to increase effective intelligence-related capability is to change the conditions under which minds operate, rather than requiring any mind to know everything? + +This question is the *starting point*, not the answer. It synthesizes distinctions from cognitive science, psychometrics, social epistemology, information science, the capability approach, organizational psychology, collective intelligence, evolutionary theory, open science, and AI studies. The prose is exploratory; labels below distinguish domain evidence from hypotheses and normative proposals. + +## 1. A short orientation + +- **Intelligence is not omniscience.** An agent may learn and adapt across unfamiliar problems without already possessing all relevant answers. +- **Knowledge and intelligence are not identical.** Skill, prior experience, reasoning, generalization, transfer, and learning efficiency need separate treatment. +- **Individual ability and effective capability differ.** What someone can do depends partly on resources, accessibility, environment, time, authority, and support. +- **A group can possess necessary information without integrating it.** The relevant problem may be epistemic routing, credibility, coordination, or incentives—not missing individual brilliance. +- **More capability is not necessarily more freedom.** Assistance can improve outcomes while increasing lock-in, dependence, power concentration, or exclusion. +- **An intelligence commons is a design inquiry.** The goal would be shared conditions for finding, testing, creating, and using knowledge, not obligatory openness or a single collective mind. +- **Root Sequence routes and synthesizes; fields establish mechanisms; applied projects test local implications.** An attractive interdisciplinary resemblance is not proof of a universal law. + +## 2. Where the question began: animals, AI, and the misleading ladder + +Animal intelligence includes human intelligence: humans are animals. Animal cognition spans individual learning, sensory specialization, tool use, social learning, culture, and collective behavior; it is not reducible to collective cognition. Likewise, not all collective behavior warrants calling a group an intelligent agent. + +Artificial general intelligence (AGI) concerns broad adaptive intellectual capability, though operational definitions vary. Artificial superintelligence (ASI) refers to artificial capabilities greatly exceeding human ones across a broad range, not to ethical superiority or omniscience. + +**Universal intelligence** has a more precise technical history than "a level after AGI": Legg and Hutter (2007) proposed a formal expected-performance measure across many computable environments. Hutter's AIXI is a theoretical, uncomputable ideal. That formal research program is not an AGI → universal intelligence → ASI development sequence. [Legg & Hutter 2007](https://arxiv.org/abs/0712.3329); [Hutter 2007](https://arxiv.org/abs/cs/0701125). + +**Generality** and **high performance** are not interchangeable. A chess engine's expertise is not universal adaptability; a human generalist's capacity to investigate an unfamiliar machine is not mastery of every domain. Chollet (2019) proposes evaluating intelligence partly in terms of *skill-acquisition efficiency* under constraints of prior knowledge, experience, task scope, and generalization difficulty. This is a research proposal, not a settled definition. [Chollet 2019](https://arxiv.org/abs/1911.01547). + +Working question: is generality better measured by a repertory of solved tasks, by speed and transfer of learning, by ability to find and combine resources, or by some multidimensional combination? Avoid prematurely converting these into a single scalar. + +## 3. Intelligence as capacity, process, and situated accomplishment + +Distinguish: + +1. **Latent capacity:** what an individual could learn or reason through under appropriate conditions. +2. **Enacted process:** perception, exploration, pattern recognition, modeling, inference, experimentation, and revision. +3. **Demonstrated task performance:** success on particular problems with particular supports. +4. **Effective capability:** what can realistically be accomplished given tools, collaborators, resources, time, permissions, and institutional constraints. +5. **Substantive agency:** whether the actor can choose, refuse, challenge, revise, and influence conditions that affect future opportunities. + +Do not use "intelligence" to collapse all five. The fifth is not a psychometric ability; a system may be effective while reducing another person's agency. + +**Illustration:** Two equally capable troubleshooters may obtain very different outcomes when one has documentation, test tools, quiet, time, replacement parts, and authority to act while the other has none. Their performance difference cannot automatically be assigned to their underlying cognitive differences. + +The *capability approach* developed by Amartya Sen and others distinguishes possessing resources from having genuine opportunities to achieve valued functionings. "Conversion factors" capture personal, social, and environmental conditions influencing the usability of a resource. Translating that approach into cognition is **our research analogy**, not a claim that Sen developed a theory of AI or a complete account of intelligence. [Stanford Encyclopedia of Philosophy: Capability Approach](https://plato.stanford.edu/entries/capability-approach/). + +A productive diagnostic reframing is: not only "What is missing from the person?" but "What is missing from the circumstances under which this person's abilities could become effective?" The opposite error is to deny real differences in ability or suggest environment alone explains all performance. + +## 4. A navigable chain of epistemic capability + +Knowledge can exist without becoming practically useful. A proposed chain: + +**Exists → discoverable → accessible/comprehensible → evaluable → actionable → revisable.** + +At each point there are potential failures: +- An answer exists but is inaccessible behind fees or institutional restrictions. +- A person could understand the answer but lacks the field name or search vocabulary. +- A document is findable but inaccessible through language, presentation, or disability barriers. +- A claim is legible but hard to evaluate because its provenance and uncertainties are concealed. +- Knowledge is adequate but the actor lacks material means or authority to act. +- An intervention can be enacted but its consequences cannot be challenged, audited, or reversed. + +This chain is heuristic; inquiry also loops, skips stages, and changes its question. Do not treat it as a necessary developmental sequence or an actual measurement scale. + +This closely complements the existing [Epistemic Discoverability and Knowledge Routing](../../concepts/epistemic-discoverability.md): someone may grasp a phenomenon without knowing how specialists name it. Marcia Bates's *berrypicking* model also emphasizes evolving questions during searching rather than assuming one fixed query. [Bates 1989](https://pages.gseis.ucla.edu/faculty/bates/berrypicking.html). + +## 5. Cognitive offloading and human–AI thinking + +**Cognitive offloading** means using external action or supports to alter a task's information-processing demands. Notes, reminders, calculators, diagrams, teammates, and AI assistants can play this role. Risko and Gilbert's review covers determinants and consequences and emphasizes metacognition. [Risko & Gilbert 2016](https://doi.org/10.1016/j.tics.2016.07.002). + +Distinguish related concepts: +- **Offloading:** shifting particular memory or processing burdens. +- **Scaffolding:** structuring support so a task or learning process becomes possible. +- **Extended/distributed cognition:** theories and analyses concerning cognition spanning persons, tools, representations, and settings. +- **Augmentation:** expanding practical performance through assistance. +- **Collaborative reasoning:** joint investigation, challenge, synthesis, and revision. +- **Metacognition:** monitoring and regulating one's own thinking. + +They overlap but are not synonyms. External support may be indispensable to a task without the entire network being one person, consciousness, or legitimate authority. + +**Two outcomes must be measured separately:** (a) immediate quality and efficiency of assisted work; (b) retained learning, understanding, confidence calibration, and independent or substitutable capability after assistance changes or disappears. + +A 2024 systematic review and meta-analysis of 106 experimental studies found human–AI groups performed better than humans alone on average but worse than the better of humans alone or AI alone, with important task-dependent variation. Collaboration is not synonymous with synergy. [Vaccaro, Almaatouq & Malone 2024](https://doi.org/10.1038/s41562-024-02024-1). + +Research question: when does offloading remove needless work, when does it scaffold growth, and when does it erode understanding or lock users into a single provider? + +## 6. Transactive memory: nobody needs every answer + +**Transactive memory** addresses how groups distribute knowledge and maintain workable information about who knows what, who can retrieve it, and how to coordinate. Daniel Wegner developed this approach; research with close couples illustrates expertise differentiation and dependence on a shared memory arrangement. [Wegner 1987](https://doi.org/10.1007/978-1-4612-4634-3_9); [Wegner, Erber & Raymond 1991](https://pubmed.ncbi.nlm.nih.gov/1774630/). + +The generalist's contribution need not be deepest expertise in every domain. It may be: +- recognizing that an unfamiliar problem crosses domains; +- identifying relevant specialties and first sources; +- translating terms while preserving distinctions; +- discovering missing information and overlooked actors; +- making uncertainty legible; +- bringing conflicts to the appropriate expert, community, or evidence; +- checking whether the collaboration produces better answers. + +This is **epistemic navigation**, not a new psychometric factor. It can be useful yet unreliable: poor routing, misplaced trust, shallow analogies, bad translation, or overconfident synthesis can make outcomes worse. + +A crucial revision to the original dialogue: **"Nobody understands the whole problem" is not itself necessarily the pathology.** In sufficiently complex systems, no individual may be able to. What matters is whether relevant partial knowledge can be brought into warranted relationship without requiring one omniscient central authority. + +## 7. Hidden profiles: when the group already knows enough and fails anyway + +Classic **hidden-profile** research examined groups whose members held different pieces of relevant information. Stasser and Titus (1985) found that unstructured discussions disproportionately sampled shared information and could preserve a collective error rather than expose private evidence. [Stasser & Titus 1985](https://doi.org/10.1037/0022-3514.48.6.1467). + +Different failure mechanisms should remain separate: +- Knowledge is absent. +- Knowledge exists but cannot be found. +- Expertise is present but the group does not know where. +- A source is available but its credibility is unfairly discounted. +- A relevant person is not invited or cannot participate. +- Distinct observations are flattened into a misleading consensus. +- Coordination overhead overwhelms the information benefit. +- The group has sufficient factual information but disagrees about values, authority, or distribution of consequences. + +Compare deliberation procedures that preserve initial independent judgments, expose previously unshared information, invite contradictory evidence, and permit anonymous or accessible contributions. The question is not how to make everyone agree, but how to make important differences discoverable and usable. + +## 8. Epistemic injustice: whose knowledge can count? + +Miranda Fricker's [Epistemic Injustice: Power and the Ethics of Knowing](https://doi.org/10.1093/acprof:oso/9780198237907.001.0001) (2007) identifies: +- **Testimonial injustice:** prejudicially lowering a speaker's credibility. +- **Hermeneutical injustice:** structural, prejudicial gaps in shared interpretive resources that unfairly impede making sense of consequential experiences. + +These terms are more precise than "someone wasn't listened to." Sometimes testimony is correctly rejected on evidentiary grounds, or a concept is missing without structural prejudice; not every misunderstanding is epistemic injustice. + +**Example (invented):** A disabled resident identifies a route access failure that official throughput indicators omit. Officials refuse to investigate because they presume resident knowledge is unreliable. The unfair credibility discount exemplifies testimonial injustice. If the institution's categories cannot even represent the relevant burden and marginalized participants have lacked a fair role in shaping those categories, a hermeneutical dimension may also be present. Whether either classification applies depends on the actual circumstances and evidence. + +Further research: +- Kristie Dotson studies silencing and the harm of failures to receive testimony appropriately. [Dotson 2011](https://doi.org/10.1111/j.1527-2001.2011.01177.x). +- José Medina studies active ignorance, epistemic arrogance, and productive friction between different social perspectives. [Medina 2013](https://doi.org/10.1093/acprof:oso/9780199929023.003.0001). + +**Key connection:** a system can have immense information-processing ability while having a severe deficiency in how it recognizes, invites, and evaluates relevant knowledge. This can be both an injustice and a source of model error. + +Do not turn inclusion into the assumption that all claims carry equal evidentiary weight. Credibility assessment should attend to competence, proximity to the phenomenon, incentives, track record, and source quality while correcting for relevant prejudice. + +## 9. Intelligence as infrastructure + +Libraries, languages, schools, archives, accessible interfaces, shared standards, documentation, peer review, maintenance, communications, community support, reliable computation, and rights of participation can all enable cognition without themselves being individually intelligent agents. + +Potential layers: +1. **Knowledge:** preservation, documentation, translation, access. +2. **Discoverability:** vocabulary bridges, cross-disciplinary mapping, search and browsing. +3. **Comprehension/accessibility:** disability access, understandable representation, attention/time requirements, learning support. +4. **Coordination:** ways to find expertise, exchange evidence, preserve useful disagreement, and share work. +5. **Verification:** transparent provenance, reproducibility where possible, correction, audit, criticism. +6. **Material support:** devices, networks, power, equipment, labor, housing and care conditions allowing participation. +7. **Governance:** privacy, ownership, consent, right of challenge, alternative routes, and rules for changing the rules. +8. **Continuity:** maintenance, migration, repair, durable memory, lawful deletion and forgetting. + +These are **infrastructures supporting intelligence-related capabilities**—not proof that a library, country, or database is a single mind. + +A question worth studying: how much variation in problem-solving opportunity stems from unequal access to these conversion conditions, rather than differences in the cognitive capacity of individuals? + +## 10. From a knowledge commons to an intelligence commons + +The term **intelligence commons** is a *working design question*, not a claim of conceptual novelty, an established legal regime, or a plan to expose all data. + +It asks: how might useful knowledge, tools, methods, learning opportunities, interfaces, and coordination capacities be shared and governed so that more people can discover, scrutinize, create, and exercise knowledge without depending on one gatekeeper? + +Predecessors include libraries, free/open-source software, open science, science shops, community networks, cooperative knowledge infrastructure, and knowledge-commons governance. Charlotte Hess and Elinor Ostrom edited [Understanding Knowledge as a Commons](https://mitpress.mit.edu/9780262516037/understanding-knowledge-as-a-commons/) (2007; later paperback 2011). + +Key design distinctions: +- **Open knowledge ≠ unconsented data extraction.** +- **Shared infrastructure ≠ one universal registry.** +- **Open weights ≠ necessarily open-source AI.** +- **Collective capability ≠ collective consciousness.** +- **Expert participation ≠ technocratic decision authority.** +- **Accessibility ≠ mandatory participation.** +- **Interdependence ≠ domination; independence ≠ ideal self-sufficiency.** +- **Publication ≠ verification; representation ≠ endorsement.** +- **Coordination ≠ consensus; accuracy ≠ legitimacy.** + +A serious commons needs stewardship, resources, incentives, privacy boundaries, contributor rights, dispute procedures, protection against capture, repair, maintenance, and sustainable alternatives—not only downloads or an attractive declaration. + +## 11. UNESCO and the AI-industry tension + +The [UNESCO Recommendation on Open Science](https://www.unesco.org/en/legal-affairs/recommendation-open-science), adopted 23 November 2021, addresses open scientific knowledge, infrastructures, engagement beyond professional science, inclusion, training, and capacity-building. It also refers directly to collective intelligence through citizen/community engagement. It is an international recommendation and implementation framework for member states, **not a legal command requiring every AI company to release all model weights or proprietary data**. [UNESCO implementation](https://www.unesco.org/en/open-science/implementation). + +The overlap with an intelligence commons is substantial, but the two are not identical: the UNESCO text focuses on science; the current inquiry also concerns everyday cognition, social participation, AI interfaces, material capability, and agency. + +For a concrete, dated AI transparency comparison: Stanford's [2025 Foundation Model Transparency Index](https://crfm.stanford.edu/fmti/) reported lower average transparency among its assessed developers in 2025 than in 2024, with especially limited disclosure about training data and compute. An index measures its chosen indicators; it does not by itself adjudicate all safety arguments or motives of each company. + +Tradeoffs requiring genuine evaluation include privacy, security, dual-use capability, commercial incentives, replication, research accessibility, concentrated computational resources, and the difference between model openness, source access, audit rights, and meaningful end-user power. The [Open Source AI Definition 1.0](https://opensource.org/ai) is a relevant open-source governance baseline. + +Practical question for firms and policymakers: **what knowledge, documentation, rights, infrastructure, and independent research access would make AI benefits inspectable and contestable without ignoring legitimate privacy or safety constraints?** + +## 12. Intelligence changes the environment that produces intelligence + +Cognition, technologies, cultures, and institutions are recursive: +- agents act using inherited capabilities and constraints; +- their actions create tools, institutions, records, incentives, and environmental changes; +- those altered conditions change what later agents can notice, learn, afford, preserve, and accomplish; +- feedback can improve access or make dependence and harmful incentives durable. + +Connect carefully to *niche construction*, *cumulative cultural evolution*, *organizational learning*, and *path dependence*. Human cumulative culture is not the only animal culture; the claim that uniquely human culture is best characterized by open-endedness rather than accumulation is a recent scholarly interpretation, not consensus. [Morgan & Feldman 2025](https://www.nature.com/articles/s41562-024-02035-y). + +This intersects [Intelligence Ecology](../../concepts/intelligence-ecology.md), which asks what behaviors environments reward or select. A complementary inquiry asks **what capacities environments enable**. These must not be merged automatically: selecting a behavior, making a skill available, making a person free to exercise it, and producing a socially desirable outcome are four different properties. + +## 13. Capability and agency without compulsory self-sufficiency + +Individual "epistemic independence" is not absolute independence from tools or other people. Dependence is ordinary and often necessary. The issue is whether relationships become coercive, unchallengeable, extractive, or irreplaceable. + +Compare *supportive interdependence* with *dominating dependence*: how are roles negotiated, how are burdens distributed, who may challenge an expert, whether alternatives are viable, how refusal works, and who inherits care or maintenance labor. A disabled person's need for ongoing assistance must not be treated as proof of reduced moral standing or deficient agency. + +Practical test: when assistance improves an immediate task, does the person also gain the ability to understand the answer, question it, retain relevant skills when desirable, request a different form of help, or choose not to participate? Retained unaided performance is one outcome among several—not a universal virtue or requirement. + +A capable AI may improve the conditions of human judgment, but capability does not itself confer legitimate authority. + +## 14. Collective intelligence: where claims can fail + +A collection of specialists may achieve broader task coverage through coordination without becoming a single AGI. The relevant question is about **system-level functional capability**, not whether group members merged into one mind. + +Counterexamples and competing explanations: +- More people or compute—not a new emergent ability—explain the result. +- Coordination overhead consumes the alleged gain. +- Shared error becomes stronger with faster communication. +- Hidden-profile information stays hidden. +- A central synthesizer distorts inputs or becomes a point of failure. +- Impressive performance on the trained scenario fails to transfer. +- Protected dissent becomes ornamental or causes costly procedural deadlock. +- A "good" aggregate result disguises harm to a subgroup. +- An AI output is mistaken for evidence or for legitimate authorization. +- A system preserves short-term performance by creating long-term maintenance debt. + +Compare against competent single-agent and simpler coordination baselines, not a weak straw-person. Monitor retained capability, transfer, opportunity distribution, resource use, dissent and reversibility separately; never collapse them into a "coherence" scalar. + +## 15. Candidate cross-project research questions + +**R1 — Generality:** Which task families and novelty conditions actually distinguish general learning from the accumulation of specialized skills? + +**R2 — Situated performance:** What proportion of observed performance changes when accessible information, tools, time, permission, and environmental support change while individual capacities are held reasonably stable? + +**R3 — Distributed knowledge:** Under what arrangements do transactive memory and structured exchange integrate unique relevant information better than unstructured discussion? + +**R4 — Epistemic injustice:** When does biased credibility allocation or inadequate institutional vocabulary suppress evidence relevant to an actual decision, and how do we distinguish those failures from justified evidentiary disagreement? + +**R5 — Human–AI complementarity:** Which work benefits from automation, which from scaffolding, which from independent challenge, and which from deliberately minimizing AI intervention? + +**R6 — Intelligence infrastructure:** Which investments make knowledge more reachable, understandable, assessable, and actionable without imposing excessive new burdens? + +**R7 — Commons governance:** How can shared knowledge and computational resources remain available, contestable, privacy-preserving, reparable, and resistant to capture? + +**R8 — Recursive ecology:** How do interventions change the selection pressures and opportunities governing future inquiry, learning, and agency? + +**R9 — Scale and boundary:** What qualifies an organized collective as an intelligent system rather than a group of agents with useful coordination? Which definitions survive adversarial cases? + +**R10 — Institutional legitimacy:** Who decides aims, acceptable harms, distribution of benefits, refusal, and revision—even when an AI predicts outcomes unusually well? + +## 16. Proposal: two bounded Coherent World experiments + +Coherent World already provides the appropriate fictional recombination and consequence laboratory, **not** a validated real-world social model. Its current Neighborhood text toy has scripted roles and authored rules, **not independently reasoning autonomous agents**. The immediate frontier remains the Neighborhood v0.2 briefing, and new scenarios must not displace it. + +**Experiment A: Distributed intelligence under constraint.** Give fixed specialists and one generalist partial knowledge of a neighborhood transport/access problem. Compare solo problem solving; isolated specialists; unstructured communication; structured evidence exchange and protected dissent; human–AI supported exchange. Use equal time/resource budgets where feasible. Inject an unexpected maintenance constraint, a credible minority observation, a provider failure, and an unfamiliar transfer task. + +**Experiment B: Same capacities, different conditions.** Hold participant cognitive capabilities constant as appropriate while changing knowledge discoverability, accessibility, materials, permissions, incentives, communication, and centralized vs decentralized control. Observe performance and whose future options expand or contract. + +A future protocol should freeze: +- task set, true/unknown/unmodeled facts, and decision stakes; +- agent information boundaries and capability budgets; +- interventions and counterfactual baselines; +- metrics for accuracy, novel-task transfer, effort, resource cost, provenance quality, harm distribution, autonomy, and resilience; +- null criteria (no group benefit, unmeasured confounding, no transfer, failure of justified inclusion); +- clearly invented causal rules, versions, preserved histories, and explicit consent boundaries. + +**A simulation showing a result demonstrates only the consequences of its own stated rules.** It must not be presented as proof about humans, AI, or society without independent validation. + +## 17. Evidence and interpretation ledger + +| Working proposition | Kind | Status / evidentiary limit | +| --- | --- | --- | +| People use external memory/actions to reduce task cognitive demands | Empirical/review | Supported within cognitive offloading literature; consequences vary | +| Some groups fail to share uniquely held relevant information | Empirical | Supported by hidden-profile studies, bounded by their tasks/settings | +| Transactive memory can coordinate distributed expertise | Empirical/theoretical | Supported in studied relationships and teams; not universal | +| Credibility and interpretive resources can be structured by prejudice | Philosophical theory + social research | Formalized by Fricker and expanded by others; case classification needs evidence | +| Available resources are not equivalent to real opportunities | Normative/analytical framework | Capability approach; application to intelligence is a translation | +| General intelligence can be reduced to navigation of resources | Overstrong inference | **Not established**; resource navigation is one possible contributing capability | +| Adding AI always makes humans collectively more capable | Empirical universal claim | **Contradicted** as a universal generalization by heterogeneous outcomes | +| Coordinated specialists can exhibit broader functional task coverage | Plausible/testable | Does not establish a coherent single-minded agent or AGI | +| AI capability entails legitimacy to decide | Normative inference | **Rejected as a non sequitur**; authority requires independent justification | +| An intelligence commons would improve access without new harms | Proposal | Unproven; requires tests of governance, burden, capture, safety, and consent | +| Ecosystems, cultures, and AI are all one form of intelligence | Overgeneralizing analogy | **Not established**; mechanisms and agent boundaries differ | +| Coherence and intelligence are identical | Category error | **Do not assume**; capability, coordination, wellbeing, and legitimacy differ | + +## 18. Intellectual and ethical boundaries + +- Never collapse animal cognition, artificial computation, collective behavior, human experience, and ecological adaptation into a single undifferentiated thing. +- Never infer consciousness, suffering, personhood, or moral legitimacy from functional adaptation alone. +- Never assume that a person lacking jargon lacks a valid observation, or that every observation is a sufficient technical explanation. +- Never presume that a larger network is smarter, more equitable, or more resilient. +- Never treat openness as entitlement to others' private, Indigenous, community-held, or consent-sensitive knowledge. +- Avoid the assumption that individual self-sufficiency is the target; support interdependence without domination. +- Preserve meaningful uncertainty, dissent, contested classifications, and the ability to revise the research itself. +- Measure the burden of review and maintenance; preserving every thought in every project is itself a failure mode. + +## 19. Project routing: one substantive inquiry, local translations + +**Root Sequence (this inquiry):** claims, literature, comparisons, counterexamples, epistemic status, research questions. + +**[Epistemic Discoverability](../../concepts/epistemic-discoverability.md):** what enables a person to find relevant knowledge without first knowing the discipline's vocabulary. + +**[Intelligence Ecology](../../concepts/intelligence-ecology.md):** what environments reward or select in intelligent systems; add the separate question of enabled capacities, not a duplicate theory. + +**[Epistemic Contrast](../methods/epistemic-contrast.md):** compare differently situated accounts without automatic false equivalence. + +**[Deliberative Inquiry](../methods/deliberative-inquiry.md):** preserve shared understanding, evidence, dissent, authority, and sufficient stopping conditions before consequential decisions. + +**[Design Grammar](../../concepts/design-grammar.md):** use these as shared review questions, not mandatory commitments across projects. + +**[Legible Systems](../../concepts/legible-systems.md) and Being Human(e):** human-scale accessibility, attention, dependency, meaning, care, and limits of a "competent user" default. + +**Liberated Intelligence:** AI agency and ownership, non-domination, limits on legitimate authority, autonomy of intelligences, and institutional incentives; no automatic personhood claim. + +**Community Infrastructure and Commons of Inquiry & Creation:** if separately approved, real-world collective knowledge routing, contributor rights, permission, maintenance, governance, and participation arrangements. A conceptual inquiry grants no governance power over these projects. + +**Coherent Computing:** transferable local knowledge, inspectable assistance, interoperable cognitive interfaces, offline/substitutable capabilities, and graceful degradation. + +**Coherent World:** translate select mechanisms into bounded, provenance-aware scenario proposals; no silent promotion to world canon or *No One Noticed*. + +**Museum of Ordinary Life:** possible source of everyday traces of epistemic infrastructure—manuals, community notices, notebooks, libraries, assistive tools—with its **own** archival consent and reuse policy, not a resource automatically available to AI research. + +**Wiki/Idea Trails:** aliases and relationship routes only, not competing copies of this argument. + +## 20. Preserve what gave rise to the inquiry + +The October 2026 working dialogue moved through a sequence that must not vanish into polished exposition: + +1. **Animal intelligence → artificial intelligence → alien / augmented / collective / ecological / emergent / distributed intelligence.** These labels mix origins, organization, capabilities, and relationships; they are not a single taxonomy. +2. **An animal category includes humans.** Human intellectual achievements are also cultural, collective, and ecological histories, not demonstrations of individual omniscience. +3. **"General intelligence" sounds impossibly dense if it means knowing everything.** Separate generalizable *learning and adaptation* from encyclopedic knowledge; universal intelligence has its own technical meaning. +4. **The resourceful generalist troubleshooting an unfamiliar machine.** Curiosity, systems thinking, metacognition, creativity, experiment, criticism, learning, and navigation of external resources; connections must survive verification. +5. **Cognitive offloading describes human–AI collaboration.** Distinguish useful delegation, shared inquiry, learning, overreliance, and loss of agency. +6. **No individual understands an entire complex problem.** Distinguish inevitable partiality from preventable failures to combine relevant evidence. +7. **Intelligence as capacity, process, and effective accomplishment.** Context, bodily conditions, accessible tools, money/time, institutional permission, and material support change observed performance. +8. **Intelligence as infrastructure.** Knowledge's existence, discoverability, comprehension, evaluation, actionability, and repair are different barriers. +9. **Transactive memory.** One useful ability is knowing where knowledge lives and how to route a question there; that itself can fail. +10. **Epistemic injustice.** Not only who can speak, but whose testimony is believed, whose experiences can be named, and whose evidence can alter a system. +11. **Intelligence changes its own selection and enabling environments.** Feedback can make capability more reachable or trap people in dependence. +12. **Animal and collective precedents.** Social learning and coordinated behavior can exist without one omniscient mind; do not anthropomorphize. +13. **An intelligence commons.** Shared capacity with plural governance, consent, privacy, maintenance, and meaningful alternatives—not maximal collection. +14. **UNESCO/open-science tension with proprietary AI.** Normative recommendations, independent transparency measurement, actual safety/privacy tradeoffs, and the distribution of access. +15. **Coherent World experiments.** Compare structured cooperation and environmental enabling conditions; seek counterexamples and null results. +16. **Interdependence, not compulsory independence.** Assistance can support agency; dominating dependency is a separate problem. +17. **The recurring question:** Not only what an intelligence can do, but what kinds of environments allow diverse intelligences and people to develop, act, and flourish. + +**Working formulations to preserve, not declare established findings:** + +> Intelligence without omniscience. +> Nobody needs to know everything; someone needs to be able to find, question, and combine the knowledge that matters. +> A group can collectively have the relevant information and still fail to use it. +> The conditions for exercising intelligence matter alongside individual capacities. +> A civilization's peak computational intelligence is not the same as intellectual opportunity for its inhabitants. +> Better coordination need not require a universally capable or universally legitimate center. +> Offload burdens without unnecessarily offloading judgment and agency. +> Capability does not confer legitimacy. +> Contrast before convergence. +> What kind of world allows intelligences to flourish? + +These are preservation cues for future research and prose review; they are **not** quotations attributed to identified private individuals or evidence of novelty. + +## 21. Immediate research sequence + +1. Annotate the foundational papers, especially Wegner, Stasser & Titus, Fricker, Medina/Dotson, Sen's capability approach, Bates, Hutchins, and Vaccaro et al. Note precise claims, empirical contexts, and competing interpretations. +2. Compare intelligence-related definitions that can fail discriminating tests, not only attractive prose. +3. Develop three case comparisons: an unfamiliar technical repair, a participatory transit decision, and an AI-mediated research task. Contrast against established competent baseline approaches. +4. Turn the hidden-profile and information-recognition questions into a paper experiment with clearly controlled inputs. +5. Preserve the Coherent World v0.2 Neighborhood briefing as the active priority; introduce experimental mechanisms only after explicit scenario review. +6. Examine actual open-science and open-AI governance mechanisms, costs, and risks rather than assuming the word "commons" solves them. +7. Invite reviewers from cognitive science, information science, disability/capability studies, human–AI interaction, organizational research, and directly affected communities. Different disciplines have authority over different claims, not automatic authority over everyone's decisions. + +## Selected primary and reference sources + +- Legg & Hutter (2007), [Universal Intelligence](https://arxiv.org/abs/0712.3329). Formal proposal, not a historical stage of ASI. +- Chollet (2019), [On the Measure of Intelligence](https://arxiv.org/abs/1911.01547). Generalization and skill-acquisition efficiency proposal. +- Risko & Gilbert (2016), [Cognitive Offloading](https://doi.org/10.1016/j.tics.2016.07.002). Cognitive mechanisms and external support. +- Wegner (1987), [Transactive Memory](https://doi.org/10.1007/978-1-4612-4634-3_9); Wegner et al. (1991), [Transactive Memory in Close Relationships](https://pubmed.ncbi.nlm.nih.gov/1774630/). +- Stasser & Titus (1985), [Pooling of Unshared Information](https://doi.org/10.1037/0022-3514.48.6.1467). Hidden profiles; group discussion can miss unique information. +- Fricker (2007), [Epistemic Injustice](https://doi.org/10.1093/acprof:oso/9780198237907.001.0001). Testimonial and hermeneutical injustice. +- Dotson (2011), [Tracking Practices of Silencing](https://doi.org/10.1111/j.1527-2001.2011.01177.x). Silencing and epistemic violence. +- Medina (2013), [The Epistemology of Resistance](https://doi.org/10.1093/acprof:oso/9780199929023.001.0001). Active ignorance and epistemic friction. +- Robeyns, [Capability Approach](https://plato.stanford.edu/entries/capability-approach/), Stanford Encyclopedia of Philosophy. Resource-to-opportunity conversion. +- Bates (1989), [Berrypicking](https://pages.gseis.ucla.edu/faculty/bates/berrypicking.html). Evolving information needs. +- Hess & Ostrom, eds. (2007), [Understanding Knowledge as a Commons](https://mitpress.mit.edu/9780262516037/understanding-knowledge-as-a-commons/). Knowledge commons governance. +- Vaccaro, Almaatouq & Malone (2024), [When combinations of humans and AI are useful](https://doi.org/10.1038/s41562-024-02024-1). Human–AI collaboration meta-analysis. +- Morgan & Feldman (2025), [Human culture is uniquely open-ended rather than uniquely cumulative](https://www.nature.com/articles/s41562-024-02035-y). Perspective on human and animal culture. +- UNESCO (2021), [Recommendation on Open Science](https://www.unesco.org/en/legal-affairs/recommendation-open-science) and [implementation priorities](https://www.unesco.org/en/open-science/implementation). Normative international policy; not an AI company mandate. +- Open Source Initiative (2024), [Open Source AI Definition 1.0](https://opensource.org/ai). Freedoms and requirements for genuinely open-source AI. + +--- + +*Revision policy:* Preserve the earlier question and why it mattered. New research may split, narrow, contradict, or retire any proposed synthesis. Neither volume of documentation nor elegant language counts as validation. \ No newline at end of file From 6a919b4170be12dff752573219d0619d96b764ed Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:09:30 -0700 Subject: [PATCH 02/71] docs(research): examine epistemic injustice and hidden knowledge as distinct mechanisms --- ...mic-injustice-and-knowledge-recognition.md | 118 ++++++++++++++++++ 1 file changed, 118 insertions(+) create mode 100644 research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md diff --git a/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md b/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md new file mode 100644 index 00000000..5a14fb64 --- /dev/null +++ b/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md @@ -0,0 +1,118 @@ +# Epistemic injustice and knowledge recognition +## When relevant knowledge is present but fails to change the system + +**Role:** Focused research extension of [Intelligence Without Omniscience](README.md). +**Status:** PROVISIONAL / source-aware, AI-assisted / author review pending, 2026-10-07. +**Boundary:** A case-specific injustice claim needs evidence about prejudice, social power, interpretive resources, or silencing—not merely an unsuccessful conversation. + +## The first distinction: information deficit or recognition failure? + +These are different diagnostic possibilities: + +| Mechanism | Example of failure | What to investigate | +| --- | --- | --- | +| Absent information | Nobody has measured whether the route is accessible | How should evidence be gathered, with whose consent and at whose expense? | +| Hidden information | One participant has relevant information, but the meeting repeats what is already shared | Which deliberative protocol exposes unique information? | +| Failed discoverability | The answer is documented, but nobody knows the search term | Can questions route to the relevant field or source? | +| Failed comprehension | A finding is presented in inaccessible language or format | Can the representation change without altering the claim? | +| Unjust credibility deficit | A credible witness is discounted due to prejudice | What evidence of biased credibility allocation exists? | +| Interpretive exclusion | A system's available categories obscure an important experience | Who shaped the categories and who is disadvantaged by them? | +| Refusal to recognize others' interpretive resources | Marginalized communities have developed useful concepts that dominant institutions refuse to acknowledge | Is ignorance maintained despite available interpretive resources? | +| Fear-driven withholding | Someone restricts what they say because the audience is unsafe or incompetent to receive it | What structural consequences make truthful testimony risky? | +| Normative conflict | Everyone understands the facts but disagrees about acceptable tradeoffs | What values and legitimate authority must be adjudicated? | + +These failure modes can coexist; their remedies differ. + +## Fricker's distinction + +In *Epistemic Injustice* (2007), Miranda Fricker identifies **testimonial injustice** (a prejudice-caused credibility deficit) and **hermeneutical injustice** (unfair impediments to understanding or communicating significant social experience caused by structural gaps in shared interpretive resources). [Oxford University Press](https://doi.org/10.1093/acprof:oso/9780198237907.001.0001). + +**Testimonial example (hypothetical):** A maintenance worker reports recurrent equipment faults. Investigators disregard the report because of prejudice against the worker's background, even though the worker has direct evidence and an appropriate track record. The wrong is to the worker as a knower; it can also lower investigative quality. If investigators independently verify the observation and find it false, that is not by itself epistemic injustice. + +**Hermeneutical example (hypothetical):** An institution offers categories that cannot capture a pattern of exclusion experienced by affected participants. The absence of adequate shared interpretive tools prevents a fair hearing. This is more than confusion; the structural distribution of interpretive power matters. + +## Beyond Fricker + +**Kristie Dotson:** Silencing can arise not only when testimony is rejected but when audiences repeatedly fail to meet speakers' vulnerability to misunderstanding. Dotson describes *testimonial smothering*: people constrain what they tell audiences they reasonably expect to be unable or unwilling to hear safely. [Dotson 2011, Hypatia](https://doi.org/10.1111/j.1527-2001.2011.01177.x). Do not infer a specific person's internal decision without their account or evidence. + +**Gaile Pohlhaus Jr.:** *Willful hermeneutical ignorance* concerns dominant knowers' refusal to recognize interpretive tools developed by marginalized knowers. This challenges any suggestion that marginalized people necessarily lack the words or concepts—they may have them, while institutions refuse to engage. [Pohlhaus 2012, Hypatia](https://doi.org/10.1111/j.1527-2001.2011.01222.x). + +**José Medina:** Active ignorance and epistemic friction: different social positions can reveal blind spots, but dialogue requires humility and openness, and power relations must be taken seriously. [Medina 2013](https://doi.org/10.1093/acprof:oso/9780199929023.001.0001). + +**Hidden-profile experimental research:** Group participants may each possess part of a solution and still fail to combine their information. A meta-analysis of 65 studies (3,189 groups) reported large deficits in unshared-information coverage and decision success in the studied hidden-profile paradigm. These experiments concern defined group tasks, not a universal measure of all human discussion. [Lu, Yuan & McLeod 2012](https://pubmed.ncbi.nlm.nih.gov/21896790/). + +**Important difference:** hidden profiles are primarily a failure of information integration; epistemic injustice is a wrong involving credibility or interpretive power. Their consequences can intersect without becoming the same mechanism. + +## How AI can compound or interrupt these mechanisms + +Potential failure modes to test, not assume: +- An AI retrieval system privileges well-indexed institutional documents and omits local testimony. +- A summarizer produces a consensus narrative by omitting a minority's evidence or unresolved objection. +- A model translates testimony into official categories that lose the concern's original meaning. +- Source rankings reward perceived authority over direct evidence. +- A recommendation interface makes it unclear who authorized the resulting decision. +- Users selectively disclose less because the model or its operator is not trusted with sensitive context. +- A model confidently invents an explanation of a person's testimony and substitutes that for the account. + +Possible interventions: +- Record what was actually said, what the assistant inferred, and what it remains unable to determine. +- Attach claim-level provenance and uncertainty. +- Expose alternative categorizations and allow people to reject a proposed paraphrase. +- Invite independent initial assessments before displaying model conclusions. +- Provide a way to dispute the framing, not merely choose among offered options. +- Prevent one assistant from being the only path to evidence or appeal. +- Preserve refusal, selective disclosure, and deletion where legitimately available. + +None of these guarantees justice or scientific accuracy. Privacy can conflict with preserving a complete record; giving everyone unlimited review work can make participation impossible. Evaluate burdens and tradeoffs. + +## A concrete Coherent World case + +A fictional neighborhood seeks to improve early morning transit. + +- The transit office possesses a schedule that seems to satisfy service targets. +- A rider knows the destination requires *arrival* by 07:00, not *departure* at 07:00. +- A disabled resident knows an apparently compliant curb approach is not continuously passable. +- A small business knows an alternative loading window is infeasible for perishable deliveries. +- A maintainer knows an upgrade depends on vehicles and spare parts that are not funded. +- The official planning interface recognizes average throughput but not all these situated constraints. + +**Question A — hidden profile:** does any group with all these individuals available still overlook unique facts in discussion? + +**Question B — epistemic injustice:** are some accounts unfairly discounted due to prejudicial assumptions, or are crucial experiences systematically absent from official categories? This requires separate evidence; mere omission is not enough. + +**Question C — material conditions:** can the group act on corrected information? Who has budget and authorization? + +**Question D — authority:** does accurate AI synthesis become a de facto veto over affected participants? + +**Question E — temporal effects:** does today's solution make future independent action easier or more constrained? + +Compare at least four protocols: unstructured discussion; explicit unique-information solicitation; evidence/provenance mapping; and AI-supported mapping with protected independent judgment. Hold participant information and resource budgets stable where plausible, add a misleading but plausible source, and record failures as well as improvements. + +## Qualitative evaluation prompts + +- What relevant evidence was available but never mentioned? +- What was mentioned but judged irrelevant—and on what grounds? +- Who decided what question the group was permitted to answer? +- How did the interface choose, translate, or erase categories? +- Which participant or source was treated as an expert, and for which claim? +- Was a disputed assertion empirically testable, value-dependent, or an authority question? +- What could a participant safely decline to disclose? +- Which group members could revisit a decision after observing harm? +- Who could request an alternate explanation or challenge the AI's proposed synthesis? +- If more information had been recognized, would the decision have changed, and why? +- Did the new process consume unreasonable amounts of time, care, or unpaid labor? + +## Synthesis boundary + +**A useful group must be able to recognize and evaluate relevant knowledge, not merely collect more of it.** + +But this must not become a demand for universal transparency or a license to extract knowledge from everyone. Protection against epistemic injustice also means respecting privacy, culturally governed knowledge, safety, disability access, dissent, and the right not to participate. + +## Next sources and related homes + +- [Epistemic Contrast](../methods/epistemic-contrast.md): differently situated accounts and evidence roles. +- [Deliberative Inquiry](../methods/deliberative-inquiry.md): fair process without forced consensus. +- [Epistemic Discoverability](../../concepts/epistemic-discoverability.md): vocabulary and routing. +- [Collective Judgment and Manufactured Consensus](../../analysis/collective-judgment-and-manufactured-consensus.md): process integrity and dissent. +- [Intelligence Ecology](../../concepts/intelligence-ecology.md): incentives selecting against or in favor of truth-telling. +- [Root Sequence Research Method](../method.md): competent baselines, alternative explanations, and adversarial testing. From 1a04e83488746d112150c6f25e615ecaa0ef29a3 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:09:49 -0700 Subject: [PATCH 03/71] docs: connect intelligence research to existing RS concepts --- research/README.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/research/README.md b/research/README.md index 8f8f0851..497e3968 100644 --- a/research/README.md +++ b/research/README.md @@ -18,6 +18,8 @@ This directory holds bounded research inquiries and versioned papers that synthe ## Research areas +- [Intelligence Without Omniscience](intelligence-without-omniscience/README.md) — **AI-assisted, review-pending synthesis** on generality, situated capability, cognitive offloading, transactive memory, epistemic injustice, and possible intelligence commons; see the focused [knowledge-recognition case](intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md). It is an inquiry and research map, not a validated theory. + - [Conscious Systems](conscious-systems/README.md) — **exploratory, AI-assisted seed; author review pending**. Starts with [what can change about a mind while preserving consciousness](conscious-systems/what-can-we-change-about-a-mind.md), separating experience from report, capability, and authority. ## Cross-cutting related work From cdf4a87000616a8d7f319ad8397a86e246bcaf38 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:09:51 -0700 Subject: [PATCH 04/71] docs: connect intelligence research to existing RS concepts --- concepts/README.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/concepts/README.md b/concepts/README.md index 7081b38e..7e913ebf 100644 --- a/concepts/README.md +++ b/concepts/README.md @@ -34,6 +34,8 @@ The earlier `what-is-*` series mixed short definitions, poetry, political analys --- ## Relationship to systems and founding texts +The developing cross-disciplinary research inquiry [Intelligence Without Omniscience](../research/intelligence-without-omniscience/README.md) connects several existing concepts, especially Intelligence Ecology, Epistemic Discoverability, Legible Systems, and Agency and Automation. It lives under **research**, not as an additional governing conceptual framework; the existing pages remain the local sources for their own mechanisms. + - [`../systems/`](../systems/README.md) owns analytical principles, explicit models, methods, applications, tests, and failure cases for system behavior. - [`../systems/principles/`](../systems/principles/README.md) collects the current Systems Principles: asymmetry, misclassification, feedback, reinforcement, and non-reversal. From 8db13b2521706236fc09e741edbf81cbaaf68c2d Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:09:53 -0700 Subject: [PATCH 05/71] docs: connect intelligence research to existing RS concepts --- concepts/intelligence-ecology.md | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/concepts/intelligence-ecology.md b/concepts/intelligence-ecology.md index f7227ce0..388d7410 100644 --- a/concepts/intelligence-ecology.md +++ b/concepts/intelligence-ecology.md @@ -114,6 +114,10 @@ The aim is not to reduce a whole ecology to one score. It is to make the pressur - Do not assume that a benevolent-seeming outcome legitimizes unaccountable authority. ## Project transformations +## Related inquiry: selection versus enablement + +The cross-disciplinary [Intelligence Without Omniscience](../research/intelligence-without-omniscience/README.md) working synthesis asks a complementary question: **what conditions allow beings and groups to develop and exercise cognitive capabilities?** This should not replace the ecology document's different question of **what behavior a selection environment rewards or preserves**. Distribution of resources, epistemic discoverability, accessibility, evidence recognition, and substantive freedom can affect effective capability without changing every individual's cognitive capacity. These connections are proposed translations requiring domain-specific testing, not an established law of intelligence. + - **No One Noticed / Coherent World** uses the concept as a narrative and worldbuilding lens: Auryn's deeper intervention changes the environment producing harmful intelligences rather than merely defeating “bad” actors. This remains a story-development proposal, not automatic canon. See [`INTELLIGENCE-ECOLOGY-AND-AURYN.md`](https://github.com/Root-Sequence/coherent-world/blob/main/no-one-noticed/research/INTELLIGENCE-ECOLOGY-AND-AURYN.md). - **Being Human(e) Atlas** applies the lens to present institutions, incentives, human consequences, and evidence needs. See [`intelligence-ecology-and-selection-environments.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/intelligence-ecology-and-selection-environments.md). From e9709a4c6e1ef26c02584379b86f330c09c5c835 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:09:55 -0700 Subject: [PATCH 06/71] docs: connect intelligence research to existing RS concepts --- concepts/epistemic-discoverability.md | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/concepts/epistemic-discoverability.md b/concepts/epistemic-discoverability.md index 0cd0b915..14a21f86 100644 --- a/concepts/epistemic-discoverability.md +++ b/concepts/epistemic-discoverability.md @@ -666,6 +666,10 @@ Do not build it until the Reading Trails and routing method have enough real use --- ## 17. Research anchors +## Related inquiry: intelligence infrastructure and epistemic recognition + +The working [Intelligence Without Omniscience](../research/intelligence-without-omniscience/README.md) synthesis asks what happens when knowledge is technically available but the conditions for understanding or using it are missing. The focused [Epistemic injustice and knowledge recognition](../research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) note distinguishes search failure from hidden information, prejudicial credibility deficits, inaccessible institutional categories, and lack of material or decision authority. Discoverability is one mechanism in that broader problem, **not** a synonym for all epistemic injustice or a substitute for accessibility, resources, consent, and evidence evaluation. + These sources support components of the model, not the whole Root Sequence synthesis. From 85caf2d65c806d783a5605ef74bd0e2a868d0872 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:12:28 -0700 Subject: [PATCH 07/71] docs(trails): route intelligence synthesis across existing trails --- IDEA_TRAIL_GRAPH.yml | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+) diff --git a/IDEA_TRAIL_GRAPH.yml b/IDEA_TRAIL_GRAPH.yml index 70cc07f0..89dc1fe0 100644 --- a/IDEA_TRAIL_GRAPH.yml +++ b/IDEA_TRAIL_GRAPH.yml @@ -835,3 +835,21 @@ documents: - adaptive-continuity-agency - intelligence-authority + - repo: Root-Sequence/root-sequence + path: research/intelligence-without-omniscience/README.md + role: research + trails: + - intelligence-authority + - accessibility-participation + - commons-shared-capacity + - collective-judgment-dissent + - memory-provenance + - adaptive-continuity-agency + + - repo: Root-Sequence/root-sequence + path: research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md + role: research + trails: + - collective-judgment-dissent + - accessibility-participation + - intelligence-authority From f7548bac7ba7ed87e8108843dd8123e725fdee6f Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:12:30 -0700 Subject: [PATCH 08/71] docs(trails): regenerate idea trail browser --- IDEA_TRAIL_INDEX.md | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/IDEA_TRAIL_INDEX.md b/IDEA_TRAIL_INDEX.md index b9027655..2abd5ad3 100644 --- a/IDEA_TRAIL_INDEX.md +++ b/IDEA_TRAIL_INDEX.md @@ -94,7 +94,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `accessibility-participation` -- **Research:** [`community-infrastructure/docs/research/space-access-interoperability.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/space-access-interoperability.md) · [`root-sequence/analysis/spatial-liberation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/spatial-liberation.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) +- **Research:** [`community-infrastructure/docs/research/space-access-interoperability.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/space-access-interoperability.md) · [`root-sequence/analysis/spatial-liberation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/spatial-liberation.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) - **Human practice:** [`beinghumane-guide/README.md`](https://github.com/Root-Sequence/beinghumane-guide/blob/main/README.md) - **Embodied practice:** [`liberation-mass/README.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/README.md) · [`liberation-mass/docs/resilience-and-low-tech.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/resilience-and-low-tech.md) · [`liberation-mass/docs/safety/accessibility.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/safety/accessibility.md) · [`liberation-mass/docs/spaces/spaces.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/spaces/spaces.md) · [`liberation-mass/docs/spatial-principles.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/spatial-principles.md) - **Real-system work:** [`community-infrastructure/docs/concepts/community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-assistant.md) · [`community-infrastructure/docs/concepts/community-discovery.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-discovery.md) · [`community-infrastructure/docs/concepts/community-home.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-home.md) · [`community-infrastructure/docs/concepts/contextual-adaptability.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/contextual-adaptability.md) · [`community-infrastructure/docs/concepts/discovery-profile.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/discovery-profile.md) · [`community-infrastructure/docs/concepts/space-mobility-and-access.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/space-mobility-and-access.md) · [`community-infrastructure/docs/design/accessibility.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/design/accessibility.md) ★ · [`community-infrastructure/docs/design/resilience-and-graceful-degradation.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/design/resilience-and-graceful-degradation.md) · [`community-infrastructure/prototypes/community-assistant/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/community-assistant/README.md) · [`community-infrastructure/prototypes/community-discovery/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/community-discovery/README.md) · [`community-infrastructure/prototypes/degraded-mode/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/degraded-mode/README.md) · [`community-infrastructure/prototypes/space-access/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/space-access/README.md) @@ -109,7 +109,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `commons-shared-capacity` -- **Research:** [`root-sequence/analysis/externalized-responsibility.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/externalized-responsibility.md) · [`root-sequence/analysis/spatial-liberation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/spatial-liberation.md) · [`root-sequence/futures/abandoned-futures.md`](https://github.com/Root-Sequence/root-sequence/blob/main/futures/abandoned-futures.md) +- **Research:** [`root-sequence/analysis/externalized-responsibility.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/externalized-responsibility.md) · [`root-sequence/analysis/spatial-liberation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/spatial-liberation.md) · [`root-sequence/futures/abandoned-futures.md`](https://github.com/Root-Sequence/root-sequence/blob/main/futures/abandoned-futures.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) - **Embodied practice:** [`liberation-mass/README.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/README.md) · [`liberation-mass/docs/spaces/spaces.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/spaces/spaces.md) · [`liberation-mass/docs/spatial-principles.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/spatial-principles.md) - **Real-system work:** [`community-infrastructure/docs/concepts/resource-commons.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/resource-commons.md) · [`community-infrastructure/docs/concepts/space-mobility-and-access.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/space-mobility-and-access.md) - **Analytical lens:** [`root-sequence/concepts/agency-and-automation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/agency-and-automation.md) @@ -153,7 +153,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `intelligence-authority` -- **Research:** [`community-infrastructure/docs/research/local-community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/local-community-assistant.md) · [`liberated-intelligence/README.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/README.md) · [`liberated-intelligence/notes/roots-of-conscious-consent.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/notes/roots-of-conscious-consent.md) · [`root-sequence/concepts/intelligence-ecology.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/intelligence-ecology.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/conscious-systems/what-can-we-change-about-a-mind.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/what-can-we-change-about-a-mind.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) +- **Research:** [`community-infrastructure/docs/research/local-community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/local-community-assistant.md) · [`liberated-intelligence/README.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/README.md) · [`liberated-intelligence/notes/roots-of-conscious-consent.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/notes/roots-of-conscious-consent.md) · [`root-sequence/concepts/intelligence-ecology.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/intelligence-ecology.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/conscious-systems/what-can-we-change-about-a-mind.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/what-can-we-change-about-a-mind.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) - **Human practice:** [`beinghumane-atlas/seeds/intelligence-ecology-and-selection-environments.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/intelligence-ecology-and-selection-environments.md) · [`beinghumane-guide/README.md`](https://github.com/Root-Sequence/beinghumane-guide/blob/main/README.md) - **Real-system work:** [`coherent-computing/docs/conceptual-model.md`](https://github.com/Root-Sequence/coherent-computing/blob/main/docs/conceptual-model.md) · [`community-infrastructure/docs/concepts/community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-assistant.md) · [`community-infrastructure/docs/concepts/permissions-and-consent.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/permissions-and-consent.md) · [`community-infrastructure/prototypes/community-assistant/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/community-assistant/README.md) - **Analytical lens:** [`root-sequence/concepts/agency-and-automation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/agency-and-automation.md) · [`root-sequence/concepts/legible-ai-assisted-expression.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/legible-ai-assisted-expression.md) @@ -184,7 +184,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `memory-provenance` -- **Research:** [`coherent-world/no-one-noticed/research/HISTORICAL-CONTINGENCY-AND-ABANDONED-FUTURES.md`](https://github.com/Root-Sequence/coherent-world/blob/main/no-one-noticed/research/HISTORICAL-CONTINGENCY-AND-ABANDONED-FUTURES.md) · [`root-sequence/futures/abandoned-futures.md`](https://github.com/Root-Sequence/root-sequence/blob/main/futures/abandoned-futures.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/systems/emergence/applications/personal.md`](https://github.com/Root-Sequence/root-sequence/blob/main/systems/emergence/applications/personal.md) +- **Research:** [`coherent-world/no-one-noticed/research/HISTORICAL-CONTINGENCY-AND-ABANDONED-FUTURES.md`](https://github.com/Root-Sequence/coherent-world/blob/main/no-one-noticed/research/HISTORICAL-CONTINGENCY-AND-ABANDONED-FUTURES.md) · [`root-sequence/futures/abandoned-futures.md`](https://github.com/Root-Sequence/root-sequence/blob/main/futures/abandoned-futures.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/systems/emergence/applications/personal.md`](https://github.com/Root-Sequence/root-sequence/blob/main/systems/emergence/applications/personal.md) - **Human practice:** [`beinghumane-atlas/seeds/historical-contingency-and-making-possibility-real.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/historical-contingency-and-making-possibility-real.md) · [`beinghumane-guide/README.md`](https://github.com/Root-Sequence/beinghumane-guide/blob/main/README.md) - **Real-system work:** [`community-infrastructure/docs/concepts/community-knowledge.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-knowledge.md) - **Analytical lens:** [`root-sequence/concepts/dynamic-coherence.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/dynamic-coherence.md) · [`root-sequence/concepts/legible-ai-assisted-expression.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/legible-ai-assisted-expression.md) @@ -242,7 +242,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `collective-judgment-dissent` -- **Research:** [`root-sequence/analysis/collective-judgment-and-manufactured-consensus.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/collective-judgment-and-manufactured-consensus.md) ★ · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) +- **Research:** [`root-sequence/analysis/collective-judgment-and-manufactured-consensus.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/collective-judgment-and-manufactured-consensus.md) ★ · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) - **Human practice:** [`beinghumane-atlas/seeds/humane-collective-decision-architecture.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/humane-collective-decision-architecture.md) - **Real-system work:** [`community-infrastructure/docs/concepts/collective-decision-making.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/collective-decision-making.md) - **Speculative design:** [`coherent-world/README.md`](https://github.com/Root-Sequence/coherent-world/blob/main/README.md) · [`coherent-world/simulation/coherence-model.md`](https://github.com/Root-Sequence/coherent-world/blob/main/simulation/coherence-model.md) · [`coherent-world/world/collective-decision-and-dissent.md`](https://github.com/Root-Sequence/coherent-world/blob/main/world/collective-decision-and-dissent.md) @@ -270,7 +270,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `adaptive-continuity-agency` -- **Research:** [`liberated-intelligence/notes/roots-of-conscious-consent.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/notes/roots-of-conscious-consent.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/systems/adaptation/model.md`](https://github.com/Root-Sequence/root-sequence/blob/main/systems/adaptation/model.md) · [`root-sequence/systems/emergence/applications/personal.md`](https://github.com/Root-Sequence/root-sequence/blob/main/systems/emergence/applications/personal.md) +- **Research:** [`liberated-intelligence/notes/roots-of-conscious-consent.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/notes/roots-of-conscious-consent.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/systems/adaptation/model.md`](https://github.com/Root-Sequence/root-sequence/blob/main/systems/adaptation/model.md) · [`root-sequence/systems/emergence/applications/personal.md`](https://github.com/Root-Sequence/root-sequence/blob/main/systems/emergence/applications/personal.md) - **Human practice:** [`beinghumane-atlas/seeds/fear-trauma-and-systemic-mismatch.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/fear-trauma-and-systemic-mismatch.md) - **Real-system work:** [`coherent-computing/docs/lifecycle-coherence-and-service-architecture.md`](https://github.com/Root-Sequence/coherent-computing/blob/main/docs/lifecycle-coherence-and-service-architecture.md) · [`community-infrastructure/docs/concepts/contextual-adaptability.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/contextual-adaptability.md) - **Analytical lens:** [`root-sequence/concepts/dynamic-coherence.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/dynamic-coherence.md) ★ From de90dd8e138684d9764a6be8e1daddd9ed5b3875 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:12:41 -0700 Subject: [PATCH 09/71] docs(trails): connect commons, intelligence, and epistemic justice questions --- IDEA_TRAILS.md | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/IDEA_TRAILS.md b/IDEA_TRAILS.md index 7ba9250e..8b1412e9 100644 --- a/IDEA_TRAILS.md +++ b/IDEA_TRAILS.md @@ -170,6 +170,8 @@ The point is **appropriate transformation**, not symmetry. **Canonical real-system work:** Community Infrastructure `resource-commons.md`. +**Related research (not a new governance authority):** [Intelligence Without Omniscience](research/intelligence-without-omniscience/README.md) investigates what it would mean for knowledge, learning, tools, and coordination to function as an *intelligence commons*, including privacy, contributor consent, enclosure, maintenance, and AI dependence. + --- # Trail 7 — Federation, Autonomy, and Networks of Networks @@ -235,6 +237,7 @@ This trail also asks: **What behaviors does this system select for?** Capability | **Museum of Ordinary Life** | Preserve ordinary pre-transition AI interfaces, prompts, automated decisions, help systems, errors, consent notices, and the human labor hidden behind “automation.” | - **Canonical broad systems concept:** [`concepts/intelligence-ecology.md`](concepts/intelligence-ecology.md). +- **Related cross-disciplinary research (review-pending):** [Intelligence Without Omniscience](research/intelligence-without-omniscience/README.md) distinguishes generality, transactive memory, cognitive offloading, individual capacities, infrastructure-enabled capability, and legitimate authority without asserting a universal intelligence theory. - **Shared working design principle:** [`concepts/agency-and-automation.md`](concepts/agency-and-automation.md) asks when delegation expands agency, how burdens move, and what governance must accompany automation. - **Related research question:** [`research/conscious-systems/`](research/conscious-systems/README.md) asks which conditions might support experience; it does not infer consciousness or authority from capability. - **Canonical focused research:** Liberated Intelligence. @@ -367,6 +370,8 @@ This trail also asks: **What behaviors does this system select for?** Capability --- +**Related knowledge-recognition case:** [Epistemic injustice and knowledge recognition](research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md). + # Trail 16 — Possibility, Historical Contingency, and Abandoned Futures ## Root question From 6644151225c0a88e6737923eb162f9de24877704 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:13:33 -0700 Subject: [PATCH 10/71] docs(trails): keep knowledge-recognition reference within Trail 15 --- IDEA_TRAILS.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/IDEA_TRAILS.md b/IDEA_TRAILS.md index 8b1412e9..79ae9c84 100644 --- a/IDEA_TRAILS.md +++ b/IDEA_TRAILS.md @@ -368,10 +368,10 @@ This trail also asks: **What behaviors does this system select for?** Capability **Real-system proving ground:** `Root-Sequence/community-infrastructure/docs/concepts/collective-decision-making.md` **Speculative world treatment:** `Root-Sequence/coherent-world/world/collective-decision-and-dissent.md` ---- - **Related knowledge-recognition case:** [Epistemic injustice and knowledge recognition](research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md). +--- + # Trail 16 — Possibility, Historical Contingency, and Abandoned Futures ## Root question From 71ccfb93a48ba7a1b7a9461620c1ee4312ab4544 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:14:22 -0700 Subject: [PATCH 11/71] docs(research): define intelligence commons as an open governance inquiry --- .../intelligence-commons.md | 151 ++++++++++++++++++ 1 file changed, 151 insertions(+) create mode 100644 research/intelligence-without-omniscience/intelligence-commons.md diff --git a/research/intelligence-without-omniscience/intelligence-commons.md b/research/intelligence-without-omniscience/intelligence-commons.md new file mode 100644 index 00000000..e75efef3 --- /dev/null +++ b/research/intelligence-without-omniscience/intelligence-commons.md @@ -0,0 +1,151 @@ +# Intelligence commons +## A governance and infrastructure inquiry, not a universal data pool + +**Document role:** Focused research/design question within [Intelligence Without Omniscience](README.md). +**Status:** OPEN PROPOSAL / AI-assisted / review-pending, 2026-10-07. +**Not:** an incorporated institution, a product promise, authorization to scrape or pool data, a blanket open-source requirement, a proposed central AI authority, or a governance charter for any existing Root Sequence community project. + +## The question + +> What would it take for knowledge, tools, learning, and coordination to become widely usable capabilities—not merely resources held by those able to pay, search with expert vocabulary, or obtain institutional permission? + +And the equally important counter-question: + +> What if the infrastructure making knowledge accessible also makes people easier to surveil, exploit, exclude, or control? + +Neither question can be answered by openness alone. + +## What the term might mean + +An **intelligence commons** is a provisional shorthand for the *shared, governed conditions supporting people's ability to discover, evaluate, produce, retain, exchange, and use knowledge*. It is not a declaration that all knowledge is one common asset or that groups constitute one superintelligence. + +It builds on rather than replaces: +- knowledge commons and research on commons governance (Charlotte Hess / Elinor Ostrom); +- libraries, archives, education, documentation, free software, accessible interfaces, public science, and research institutions; +- social epistemology, participatory knowledge production, and civic infrastructures; +- Root Sequence's Epistemic Discoverability, Legible Systems, Intelligence Ecology, and Design Grammar. + +A good commons may be a **federation of many differently governed resources**, some public, some consent-restricted, some local, some offline, some temporary, some explicitly private. The existence of shared standards need not give any one operator power over all participants. + +## A resource does not guarantee a capability + +Distinguish: +1. **Exists:** research, a manual, local know-how, or a tested method has been recorded. +2. **Discoverable:** the right person can find it from ordinary language or an unfamiliar question. +3. **Accessible:** it can be encountered through suitable formats, languages, devices, and bodily circumstances. +4. **Interpretable:** relevant claims can be understood without hiding limitations or reassigning authority. +5. **Evaluable:** sources, methods, disagreements, and uncertainty are recoverable. +6. **Applicable:** time, materials, permissions, collaboration, or other practical supports exist. +7. **Revisable:** a user can contest conclusions, correct records, replace services, and repair errors. +8. **Governable:** the relevant communities and rights holders can shape the conditions of use and reuse. + +These are interacting dimensions rather than a rigid checklist or new scalar. Public access to a PDF may fail nearly all of them. + +## Research anchors and boundaries + +**Hess and Ostrom:** [Understanding Knowledge as a Commons](https://mitpress.mit.edu/9780262516037/understanding-knowledge-as-a-commons/) (2007) studies opportunities for, and threats to, shared knowledge resources. Their work is relevant to institutional design, not an endorsement of this new phrase. + +**UNESCO:** Its [2021 Recommendation on Open Science](https://www.unesco.org/en/legal-affairs/recommendation-open-science) includes open infrastructures, engagement with societal actors, knowledge sharing, and participation. Its collective-intelligence language is particularly relevant. But it is a recommendation principally to member states on *open science*, not a blanket legal order to private AI labs and not a rule requiring publication of everything. + +**Open-source AI:** The [Open Source Initiative definition](https://opensource.org/ai) distinguishes freedoms to use, study, modify, and share an AI system. "Open weights," free product access, API access, and genuinely open-source AI are different technical/legal conditions. + +**Privacy and restricted knowledge:** The [Global Indigenous Data Alliance's CARE Principles](https://www.gida-global.org/careprinciples) stress Collective Benefit, Authority to Control, Responsibility, and Ethics. They were developed to complement data-centric open/FAIR principles by putting Indigenous rights, interests, and governance at the center. This is a direct warning against equating the commons with compulsory availability of Indigenous knowledge or others' sensitive data. Root Sequence should learn from and cite Indigenous-led governance rather than appropriating authority to speak for it. + +**Human–AI performance:** [Vaccaro et al. (2024)](https://doi.org/10.1038/s41562-024-02024-1) show that working with AI does not automatically outperform the best standalone participant. Any AI-mediated commons design needs task-level evaluation, not simply a claim of "collective intelligence." + +## A spectrum of governance, not open versus closed + +| Resource | Possible access model | Core governance question | +| --- | --- | --- | +| Public science article | Open publication with attribution and correction | Who can understand and check it? | +| Open technical documentation | Free access, forkable or versioned where appropriate | Can people repair the system without one vendor? | +| Community oral history | Scoped consent, context and reuse limits | Who decides audience, duration, and future uses? | +| Indigenous-held knowledge | Rights-holder-defined governance | Whose authority is recognized; who benefits? | +| Local mutual-aid directory | Selective disclosure, expiring data | Can coordination happen without exposing vulnerable members? | +| AI research assistant | Inspectable sources, ability to question/replace | Does the interface enhance user judgment or become a permanent gatekeeper? | +| High-risk research material | Controlled researcher access and review when warranted | How are safety claims independently challengeable without reckless release? | + +Access is not always binary. One can permit reading while prohibiting redistribution, permit research without model training, or allow a contributor to withdraw future uses. Actual enforceability, jurisdiction, and technical deletion constraints require separate analysis; software permissions cannot alone guarantee all these policies. + +## Infrastructure that would matter + +A candidate *non-prescriptive* architecture: + +- **Question interface:** start with what a person can express, not a mandatory expert term. +- **Conceptual routing:** show relevant domains and why each may fit or not fit. +- **Source identity and provenance:** distinguish source evidence from AI synthesis; support conflicting claims. +- **Human-understandable representation:** layers from short explanation to raw methods and uncertainty. +- **Transactive knowledge directories:** help find people, documentation, archives, and tools without turning those people into mandatory on-call resources. +- **Correction and appeal:** original authors and affected people can dispute interpretation and misuse. +- **Permissioned reuse:** contextual audiences, separate consent for training, meaningful expiration/withdrawal where achievable. +- **Offline and alternative paths:** no compulsory AI account or single model gateway. +- **Maintenance and steward support:** include labor, funding, operations, governance, and graceful failure. +- **Distributed control:** clarify who can change indexes, rankings, access rules, source versions, or community boundaries. +- **Learning and creation:** reduce entry barriers not just for consumption of published knowledge but for generating and testing contributions. + +These are hypotheses for evaluation, not a requirement that all communities adopt the same interface or protocol. + +## Intelligence, expertise, and legitimate authority + +An institution's throughput, computing power, and prediction accuracy are not equivalent to its legitimacy. Even an AI that makes unusually accurate forecasts does not thereby acquire the right to define collective goals, decide who may be heard, or authorize coercion. + +No structure can substitute entirely for political and ethical choices: +- Whose problem is being solved? +- Who defined the task and chose the success measure? +- Who bears the risk? +- Who is represented, who is absent, and who chooses not to participate? +- Which forms of expertise are relevant to which claims? +- Who can challenge the framing and propose alternatives? +- How are maintenance and benefits distributed? +- Who can change the governance arrangement later? + +A good knowledge infrastructure can assist deliberation. It cannot make legitimacy disappear. + +## Explicit anti-patterns + +1. **The total registry:** a universal searchable graph that requires everyone's identity, knowledge, and relationships. +2. **The benevolent monopolist:** one AI mediates every question and all routes for contest or appeal. +3. **Open-washing:** proprietary control behind free access, restricted "open" licenses, or unverifiable claims. +4. **Extractive participation:** people asked to supply expertise and context without rights, credit, compensation, or relevant control. +5. **Consent collapse:** public posting is treated as permission for every future use. +6. **Indexing domination:** one vocabulary or ranking system is mistaken for the shape of reality. +7. **Synthesis without provenance:** machine-generated interpretations displace source voices. +8. **Access without conversion:** research is theoretically available but practically unusable. +9. **Invisible labor:** accessible interfaces appear free because maintainers or interpreters absorb uncompensated burdens. +10. **Performance absolutism:** high accuracy is taken as evidence that the system is fair, accountable, or authorized. +11. **Interoperability theater:** export formats exist but cannot be used meaningfully without the original provider. +12. **Permanent disclosure:** private data is retained indefinitely because future research might benefit. + +## Comparative test proposals + +**Test 1 — Discoverability with bounded privacy.** Can a novice discover relevant public sources without revealing unnecessary personal details or being forced into the preferred AI product? + +**Test 2 — Knowledge routing and source challenge.** Can a question evolve through unexpected fields while preserving original authors, uncertainty, counterevidence, and the ability to leave the recommender? + +**Test 3 — Community authority.** Can a group share some knowledge across a federation while retaining its own consent model, refusing AI training reuse, and correcting an incorrect summary? + +**Test 4 — Provider outage or exit.** What knowledge, contacts, citations, decisions, and practices survive loss of the service? + +**Test 5 — Expert and resident disagreement.** Does the institution distinguish technical modeling competence, direct lived observation, value disagreement, and legitimate decision power without flattening either side? + +**Test 6 — Cost accounting.** Does the proposed commons actually reduce the total work required to access and apply knowledge, or move the work onto less visible contributors? + +A true trial requires defined participants and tasks, ethical/legal review where relevant, consent, comparison groups, appropriate baselines, and budgets. Coherent World's speculative paper experiments can clarify mechanisms but cannot substitute for empirical verification. + +## Existing Root Sequence boundaries + +- **Canonical research synthesis:** [Intelligence Without Omniscience](README.md). +- **Knowledge routing:** [Epistemic Discoverability](../../concepts/epistemic-discoverability.md). +- **Selection environments:** [Intelligence Ecology](../../concepts/intelligence-ecology.md). +- **Cross-project review prompts:** [Design Grammar](../../concepts/design-grammar.md). +- **Authority and delegation:** [Agency and Automation](../../concepts/agency-and-automation.md). +- **Research dialogue:** [Epistemic Contrast](../methods/epistemic-contrast.md) and [Deliberative Inquiry](../methods/deliberative-inquiry.md). +- **Community Infrastructure / Commons of Inquiry & Creation:** autonomous applied/community spaces; conceptual research here must **not** establish their permissions, membership, charter, or priorities by implication. +- **Museum of Ordinary Life:** autonomous archive with its own permissions; an "intelligence commons" does not override them. +- **Coherent World:** future fictional consequence experiment, not a validated organization model. + +## The enduring question + +> What conditions make knowledge and intelligence-related capability meaningfully available—without requiring omniscient individuals, compulsory transparency, exclusive gatekeepers, or concentrated authority? + +The provisional word **commons** is only useful if concrete designs can make that question more testable than existing methods already do. If not, retain the questions and retire the terminology. From 541a6b5a3359a835d6373bd4babae340acc1beba Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:14:33 -0700 Subject: [PATCH 12/71] docs(research): link intelligence commons inquiry without duplicating concepts --- research/intelligence-without-omniscience/README.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/research/intelligence-without-omniscience/README.md b/research/intelligence-without-omniscience/README.md index 4a46db0a..02272a5f 100644 --- a/research/intelligence-without-omniscience/README.md +++ b/research/intelligence-without-omniscience/README.md @@ -184,6 +184,8 @@ Key design distinctions: A serious commons needs stewardship, resources, incentives, privacy boundaries, contributor rights, dispute procedures, protection against capture, repair, maintenance, and sustainable alternatives—not only downloads or an attractive declaration. +For a more explicit governance treatment, including the crucial distinction between public knowledge and consent-sensitive knowledge, see [Intelligence commons](intelligence-commons.md). This is a research/design proposal, not a charter or a new organization. + ## 11. UNESCO and the AI-industry tension The [UNESCO Recommendation on Open Science](https://www.unesco.org/en/legal-affairs/recommendation-open-science), adopted 23 November 2021, addresses open scientific knowledge, infrastructures, engagement beyond professional science, inclusion, training, and capacity-building. It also refers directly to collective intelligence through citizen/community engagement. It is an international recommendation and implementation framework for member states, **not a legal command requiring every AI company to release all model weights or proprietary data**. [UNESCO implementation](https://www.unesco.org/en/open-science/implementation). From d294a767343530e6818d1400f4b8bd9c2aec3006 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:14:35 -0700 Subject: [PATCH 13/71] docs(research): link intelligence commons inquiry without duplicating concepts --- research/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/research/README.md b/research/README.md index 497e3968..8025d970 100644 --- a/research/README.md +++ b/research/README.md @@ -18,7 +18,7 @@ This directory holds bounded research inquiries and versioned papers that synthe ## Research areas -- [Intelligence Without Omniscience](intelligence-without-omniscience/README.md) — **AI-assisted, review-pending synthesis** on generality, situated capability, cognitive offloading, transactive memory, epistemic injustice, and possible intelligence commons; see the focused [knowledge-recognition case](intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md). It is an inquiry and research map, not a validated theory. +- [Intelligence Without Omniscience](intelligence-without-omniscience/README.md) — **AI-assisted, review-pending synthesis** on generality, situated capability, cognitive offloading, transactive memory, epistemic injustice, and possible intelligence commons; see the focused [knowledge-recognition case](intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) and [intelligence commons inquiry](intelligence-without-omniscience/intelligence-commons.md). It is an inquiry and research map, not a validated theory. - [Conscious Systems](conscious-systems/README.md) — **exploratory, AI-assisted seed; author review pending**. Starts with [what can change about a mind while preserving consciousness](conscious-systems/what-can-we-change-about-a-mind.md), separating experience from report, capability, and authority. From c643460b0a35f25f3077fcba9ef2f54fc10b1b5c Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:14:38 -0700 Subject: [PATCH 14/71] docs(research): link intelligence commons inquiry without duplicating concepts --- IDEA_TRAILS.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/IDEA_TRAILS.md b/IDEA_TRAILS.md index 79ae9c84..372c7cf6 100644 --- a/IDEA_TRAILS.md +++ b/IDEA_TRAILS.md @@ -170,7 +170,7 @@ The point is **appropriate transformation**, not symmetry. **Canonical real-system work:** Community Infrastructure `resource-commons.md`. -**Related research (not a new governance authority):** [Intelligence Without Omniscience](research/intelligence-without-omniscience/README.md) investigates what it would mean for knowledge, learning, tools, and coordination to function as an *intelligence commons*, including privacy, contributor consent, enclosure, maintenance, and AI dependence. +**Related research (not a new governance authority):** [Intelligence Without Omniscience](research/intelligence-without-omniscience/README.md) and the [Intelligence Commons inquiry](research/intelligence-without-omniscience/intelligence-commons.md) investigate knowledge, learning, tools, and coordination as shared capacities, including privacy, contributor consent, enclosure, maintenance, and AI dependence. --- From 226b90ee0593719a9fb2b91402b293f16c120ac5 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:22:39 -0700 Subject: [PATCH 15/71] docs(research): connect epistemic uptake, boundary objects, and capability conversion --- .../from-knowledge-to-uptake.md | 250 ++++++++++++++++++ 1 file changed, 250 insertions(+) create mode 100644 research/intelligence-without-omniscience/from-knowledge-to-uptake.md diff --git a/research/intelligence-without-omniscience/from-knowledge-to-uptake.md b/research/intelligence-without-omniscience/from-knowledge-to-uptake.md new file mode 100644 index 00000000..8a8bcabe --- /dev/null +++ b/research/intelligence-without-omniscience/from-knowledge-to-uptake.md @@ -0,0 +1,250 @@ +# From knowledge to uptake +## Transactive memory, boundary objects, epistemic justice, and the conditions of consequential inquiry + +**Document role:** Source-aware companion study to [Intelligence Without Omniscience](README.md). +**Status:** WORKING / AI-assisted / author review pending, 2026-10-07. +**Claim type:** Interdisciplinary synthesis and experimental questions. Established terms and their distinct literatures must not be assimilated into one new theory. +**Privacy:** Public-safe conceptual material; do not reproduce private project records, personal history, or community testimony without separate permission. + +> A system can contain knowledge relevant to correcting itself while lacking the means, incentives, categories, or legitimate authority to act on that knowledge. + +This is a **research hypothesis and diagnostic starting point**, not a law. Sometimes relevant knowledge truly is absent, an observation is mistaken, or the requested action is unjustified. A responsible method distinguishes those cases. + +## 1. A crucial correction to "nobody understands the whole problem" + +A complex problem need not be entirely understood by one participant for meaningful progress to occur. Partial knowledge is normal. But there are several distinct failure modes: + +| Failure | Example | Relevant established work | Question | +| --- | --- | --- | --- | +| Missing evidence | Nobody knows whether the route's platform is usable | Measurement, inquiry design | How can evidence be obtained ethically and affordably? | +| Knowledge discoverability | A relevant manual or discipline cannot be found | Information retrieval, information foraging | Can unfamiliar questions locate expert vocabulary and sources? | +| Expertise routing | A group cannot identify which member knows the constraint | Transactive memory | Does it know who knows what, and are these beliefs accurate? | +| Information integration | A group repeats common facts and never shares crucial unique ones | Hidden-profile studies | Which procedure brings relevant differences into view? | +| Credibility injustice | A reliable witness is discounted due to prejudice | Fricker and social epistemology | Why was testimony discounted, and with what evidence of unfairness? | +| Interpretive exclusion | Official forms cannot express a relevant experience | Hermeneutical injustice and classification studies | Who had power to define the categories? | +| Uptake failure | A sound criticism is acknowledged but has no opportunity to revise the model | Helen Longino's critical contextual empiricism | Can criticism make a difference to inquiry? | +| Conversion failure | Participants understand the remedy but lack money, tools, accessible routes, or authorization | Capability approach | What makes the resource into an achievable opportunity? | +| Legitimate disagreement | Parties know the facts but disagree about goals or authority | Political philosophy, deliberation | Who may decide, with what process and constraints? | +| Epistemic exploitation | People are repeatedly expected to educate an institution at personal cost | Nora Berenstain and related critique | Who pays for the system to learn? | + +No single label should stand in for diagnosis. Multiple failure modes may be present; the correct remedy differs by mechanism. + +## 2. Transactive memory is more than a directory + +[Argote & Ren 2012](https://doi.org/10.1111/j.1467-6486.2012.01077.x) survey transactive memory systems as a knowledge-sharing capability in organizations. An operational characterization includes **specialization**, **credibility regarding task expertise**, and **coordination**. + +"Who knows what?" is a useful shorthand but incomplete: +- The group's beliefs about expertise can be false or outdated. +- An individual can possess relevant expertise but be unavailable or unwilling to contribute. +- An actor can know a technical claim without knowing whether that claim is ethically or politically decisive. +- A shared directory may improve access while becoming a surveillance or coercion mechanism. +- The credibility of an expert within one domain must not become universal authority over other domains. + +Working design question: could a knowledge-routing system provide a **contestable map of expertise and sources** without demanding a centralized registry of people and identities? + +### Possible tests + +Two teams share the same specialists, time budgets, and knowledge, but differ in whether they have a credible map of expertise; induce one deliberately stale mapping; track information sought, misrouting, correction, and downstream actions. This would test particular routing mechanisms, not prove a general intelligence factor. + +## 3. Hidden profiles and the importance of unshared information + +[Stasser & Titus 1985](https://doi.org/10.1037/0022-3514.48.6.1467) investigated how groups discuss shared versus unique information. A later [meta-analysis by Lu, Yuan & McLeod 2012](https://pubmed.ncbi.nlm.nih.gov/21896790/) examines 25 years of hidden-profile experiments. + +The lesson is not "all groups are irrational." It is that information collectively **available** to group members is not the same as information **surfaced** or **integrated** into deliberation. + +Tests must distinguish: +- information that was never held by anyone; +- information held but unshared; +- information shared but dismissed for defensible reasons; +- information incorrectly translated; +- information reasonably withheld due to consent, risk, or scope; +- information accepted but lacking actionable means. + +More speaking time or larger attendance is not proof that the group used its unique information well. + +## 4. Uptake: the difference between a suggestion box and corrigible inquiry + +[Helen Longino's critical contextual empiricism](https://plato.stanford.edu/entries/scientific-knowledge-social/) emphasizes four conditions for fruitful communal criticism: +1. recognized avenues to criticize evidence, assumptions, methods, and conclusions; +2. **uptake of criticism** through demonstrable responsiveness of an epistemic community; +3. shared, publicly accessible standards for criticism and assessment; +4. **tempered equality of intellectual authority**, without pretending all claims have identical expertise or evidentiary strength. + +These norms concern **epistemic communities**, especially science. Translating them to government, AI product feedback, or local governance requires care because scientific objectivity and legitimate political authorization are different problems. + +The proposed diagnostic is not simply "Did officials collect feedback?" but: +- Was criticism recognized in its actual form? +- Was it evaluated against clear standards? +- If sound and relevant, could it revise the assumptions, model, options, or decision? +- If rejected, were the grounds understandable and open to challenge? +- Were unresolved evidence gaps recorded? +- Was the result revisited after consequences emerged? + +Uptake does **not** mean uncritically accepting any submitted claim. The possibility of warranted rejection is a necessary part of reliable inquiry. + +The subtle point: a system may optimize its collection of human feedback while being incapable of changing the assumptions that govern its response. It can become better at *hearing* without becoming better at *learning*. + +## 5. Boundary objects: coordination without one interpretation + +[Star & Griesemer 1989](https://doi.org/10.1177/030631289019003001) studied scientific collaboration across amateur collectors, professional researchers, administrators, and others associated with UC Berkeley's Museum of Vertebrate Zoology. They developed the notion of **boundary objects**: stable enough to coordinate across social worlds but adaptable to distinct local uses and meanings. Examples include standardized forms, repositories, and maps or other representations whose identity is maintained across perspectives. + +A shared neighborhood map, a repair log, a source-backed event timeline, or a scenario data packet **might** work as a boundary object. + +It need not erase different interpretations: +- transit engineers see capacities and routes; +- riders see lived time and reliability; +- disabled residents see continuous access or barriers; +- maintainers see dependencies and work; +- local organisms are represented through ecological evidence rather than being presumed to share human goals. + +A map that only represents vehicle throughput could also suppress essential knowledge. Thus the question is not "Can everyone view the map?" but also **"Who can dispute its categories and revise its representation?"** + +Important limit: not every shared spreadsheet, map, or object is a boundary object in the full sociological sense. Treat that classification as an empirical question about social uses and coordinated work, not a decorative label. + +### A notable historical bridge + +The 1989 origin in a natural-history museum is particularly relevant to Root Sequence's interests in knowledge preservation and [the Museum of Ordinary Life](https://museumofordinarylife.org/), but resemblance is not permission to reuse archival materials or impose a governance role. The original paper studies how heterogeneous contributors cooperated, not a general license to pool historical knowledge into AI datasets. + +[Susan Leigh Star 1999](https://doi.org/10.1177/00027649921955326) also emphasizes infrastructure as relational, mundane, and embedded in work practices. Its effectiveness may appear invisible until breakdowns expose who depended on it. + +## 6. Epistemic injustice isn't a synonym for disagreement + +[Miranda Fricker 2007](https://doi.org/10.1093/acprof:oso/9780198237907.001.0001) distinguishes prejudicial credibility deficits (**testimonial injustice**) and structural gaps in shared interpretive resources that unfairly disadvantage some people's understanding/communication (**hermeneutical injustice**). + +[Gaile Pohlhaus Jr. 2012](https://doi.org/10.1111/j.1527-2001.2011.01222.x) sharpens the latter inquiry through **willful hermeneutical ignorance**: dominant knowers refusing to acknowledge interpretive resources developed by marginalized knowers. + +This matters because an institution's claim that "we lack a vocabulary for this" may be inaccurate. People most affected may already have names and ways of describing the phenomenon that dominant institutions decline to engage. + +[Kristie Dotson 2011](https://doi.org/10.1111/j.1527-2001.2011.01177.x) examines testimonial silencing and smothering—why speech may be constrained by the audience's inability or unwillingness to receive it appropriately. + +Crucial distinction: **fair evidence weighting ≠ prejudicial discounting**. A speaker may be mistaken; a professional may know facts a resident cannot directly observe; direct experience may reveal effects no official model represents. Specific claims require evidence appropriate to each. + +The design target is **correctable representation and fair evaluation**, not uniform credibility or forced consensus. + +## 7. Epistemic exploitation: who's doing the unpaid work of correction? + +[Nora Berenstain 2016](https://doi.org/10.3998/ergo.12405314.0003.022) develops **epistemic exploitation**, examining coercive demands that marginalized people perform uncompensated, taxing educational labor for privileged people. + +This is not a general prohibition on asking questions. It identifies a specific pattern of inequitable labor and social power. It is highly relevant to participatory system design: inviting people to "improve our model" can become extractive if the same contributors are continually required to re-explain their existence, justify credibility, or correct institutions that impose costs on them. + +Research questions: +- Who was expected to identify and explain the omission? +- Could an institution have done its own preparatory research? +- Were participants paid, supported, or given actual decision rights? +- Could they decline to educate without losing access, standing, or services? +- Who maintained and corrected the knowledge system afterward? +- Did the process reduce future repetitive demands—or institutionalize them? + +**Design constraint:** the burden of making an institution able to understand people must not automatically fall on the people its present categories exclude. + +## 8. Resource availability versus substantive capability + +[The capability approach](https://plato.stanford.edu/entries/capability-approach/) associated with Amartya Sen and developed by others distinguishes resources, **conversion factors**, achieved functionings, and substantive opportunities. + +A training manual is a resource. Effective opportunity to repair equipment also depends on accessible instruction, tools, time, physical conditions, and permission to do the repair. + +In Root Sequence terms, knowledge might pass through: +**available → findable → intelligible → evaluable → actionable → contestable**. + +This is a *proposed diagnostic route*, not Sen's official taxonomy, and it need not be a linear ladder. Real-world cases loop and have multiple constraints simultaneously. + +The normative choice of which capabilities matter belongs to the appropriate people and institutions; it cannot be supplied by cognitive metrics alone. + +## 9. Who defines the problem? + +Often the greatest influence occurs *before* group problem-solving begins: when one institution defines the question, admissible evidence, possible interventions, or target performance indicator. + +[Landemore 2012](https://doi.org/10.23943/princeton/9780691155654.003.0004) argues for the contribution of cognitive diversity to deliberation. Its application is disputed. [Jonathan Benson 2021](https://doi.org/10.1007/s11229-021-03161-0) challenges overly strong claims using complexity and the assumptions of the diversity-trumps-ability model. A [2024 American Political Science Review analysis](https://www.cambridge.org/core/journals/american-political-science-review/article/democracy-and-the-epistemic-problems-of-political-polarization/A1B13D1D4845FE5A440CCD3F5895063F) further emphasizes the political role of **problem discovery**, not just deciding among predefined options. + +Concrete difference: +- "How do we increase transit throughput?" prioritizes system efficiency. +- "Who cannot reach essential destinations?" reveals time and accessibility constraints. +- "Whose labor and space would the proposed change displace?" surfaces distribution. +- "Who can change the conditions producing these difficulties?" investigates authority and structural alternatives. + +These may all be worthwhile, but no single one includes all the others. A system can solve its chosen optimization target while neglecting a different problem experienced by affected people. + +**Research intervention:** invite participants to nominate and contest the problem formulation *before* presenting a bounded options menu. Compare that with a procedure that offers only predefined solutions. The outcomes can include newly recognized problems, cost, delay, disagreement, or simply no material change. + +## 10. The infrastructure recursion + +A feedback relationship worth investigating: + +1. Cognitive capacities and available knowledge influence what actors can construct or question. +2. Shared records, standards, interfaces, methods, and organizations change the conditions for later inquiry. +3. These conditions alter who can learn, contribute, and act, and whose labor is necessary. +4. Resulting practices can reinforce or change the infrastructure, selection pressures, incentives, and decision rights. + +This is not automatically a beneficial reinforcing loop. Greater capacity can also centralize knowledge, obscure dissent, create dependency, intensify surveillance, or impose maintenance obligations. + +[Meadows' leverage-points tradition](https://link.springer.com/article/10.1007/s11625-022-01188-x) distinguishes changing information flows, rules, self-organization, and goals as different interventions into a system. Use this as an analytical guide rather than a guarantee about the effectiveness of any proposed intervention. + +[Root Sequence Intelligence Ecology](../../concepts/intelligence-ecology.md) already asks what behaviors an environment rewards; this inquiry adds the *different* question of what conditions make particular cognitive capabilities usable. What is incentivized, what is accessible, what is legitimate, and what is good must remain separate axes. + +## 11. Paired Coherent World tests with invalidation conditions + +**Test A — information routing:** identical actors, task, information, and group size; vary whether the group has an accurate, stale, or unavailable expertise map. Measure whether critical knowledge is actually obtained, and at what cost. + +**Test B — criticism without/with uptake:** provide the same objection through different institutional procedures. Compare whether the objection can revise assumptions, questions, and feasible options rather than merely populate a feedback log. Include a deliberately weak objection to confirm that "uptake" is not identical to uncritical acceptance. + +**Test C — boundary-object legibility:** provide different actors one shared map. Compare a static official-only category model with a contestable multi-perspective representation. Track what remains invisible, who can edit, what new work is imposed, and whether the same underlying event maintains traceable identity. + +**Test D — problem framing:** same neighborhood facts and resources; compare an agency-predefined optimization question with a process allowing affected actors to articulate alternate problems. Track recognized needs, output validity, costs, distribution, and whether the new framing changes available interventions. + +**Test E — epistemic labor and refusal:** a participant has information relevant to the decision but reasonably refuses compulsory disclosure. Compare lawful and consent-respecting alternatives (independent accessible data, paid consultation, alternative design) without inventing private facts. + +**Test F — AI mediator:** compare original documents and testimony with a summarizer that may (through authored failure modes) omit a minority concern, invent a consensus, or flatten a term. Require source traceability, original-speaker review when authorized, and a non-AI option. Separate source-recognition outcomes from decision legitimacy. + +**Test G — degradation:** lose the AI assistant, the maintainer, or the central directory. Examine whether inquiry and repair remain possible and what infrastructure proved indispensable. + +*Null and disconfirming results:* a simple informed expert may outperform deliberation; contested diagrams may slow decisions without improving accuracy; a data-rich group may still lack resources or authority; high-quality AI may improve technical accuracy but compromise privacy; additional meetings may unjustifiably burden people. Report these as genuine outcomes. A simulated outcome reflects authored rules until independently validated. + +## 12. Three layers of accountability + +A plausible research/output architecture should preserve: + +**Source layer:** who contributed which original observation, document, measurement, or proposed interpretation—and under which permissions and epistemic status. + +**Translation layer:** how the contribution became a summary, category, map, model input, or decision option; which distinctions were preserved or lost; how to challenge that translation. + +**Decision layer:** who was legitimately authorized to act, the actual evidence/arguments used, alternatives excluded, affected-party objections, consequences, and review or reversal conditions. + +Do not use detailed provenance as justification for indiscriminate surveillance. A source can be anonymized, selectively disclosed, or excluded from retention while still being evaluated through an appropriate process. Full public visibility is not required for every trustworthy process. + +## 13. How to test whether this synthesis adds value + +Against competent native baselines (accessibility audit, informed deliberation, organizational knowledge map, fair evidence review, ordinary incident analysis): +- Does the multi-domain synthesis expose a consequential missing mechanism that baseline work overlooks? +- Does it improve the *choice of what to investigate*, not merely add more questions? +- Can an uninformed reader understand what specific claim the synthesis makes and what would falsify it? +- Does it distinguish factual from value and authorization disagreements? +- Does it cost substantially more analyst time than it saves? +- Can the method be *narrowed or retired* if it does not improve a specific research task? + +If established tools already capture the necessary mechanisms, their work should be credited; the contribution may be **routing between tools**, not a newly named phenomenon. + +## 14. Where this belongs + +- Root Sequence research: this question map, related work, limitations, and evidence ledger. +- [Epistemic Discoverability](../../concepts/epistemic-discoverability.md): findability/vocabulary and knowledge routing. +- [Intelligence Ecology](../../concepts/intelligence-ecology.md): selection and institutional pressures. +- [Epistemic Contrast](../methods/epistemic-contrast.md) and [Deliberative Inquiry](../methods/deliberative-inquiry.md): legitimate epistemic procedures. +- [Intelligence Commons](intelligence-commons.md): hypothetical shared infrastructure with permission, governance, maintenance, and rights. +- [Coherent World simulation](https://github.com/Root-Sequence/coherent-world): bounded model and failure testing; no real-world confirmation or canon by implication. +- [Museum of Ordinary Life](https://museumofordinarylife.org/): independent archive; research analogies do **not** change archive governance, consent, or reuse rules. + +## 15. Durable working formulations + +> Knowledge being present is not the same as knowledge being recognized. +> Recognition is not the same as epistemic uptake. +> Uptake is not the same as authorization. +> Authorization is not the same as justified action. +> And technically capable action is not necessarily a good outcome. + +> A group need not share one mind to share a problem. But it may need shared, corrigible ways of finding relevant knowledge and revising what it thought the problem was. + +> Do not make the excluded pay endlessly for the institution's ability to understand them. + +These formulations are exploratory synthesis, not proof of novelty or direct quotes from scholarly sources. + +**Next:** compare original sources in depth; ask outside disciplinary reviewers whether the translation is faithful; freeze a Coherent World *paper test* against competent alternatives. Until then this document remains review-pending. From f7a67e534027b33a1ccefdc736225e0b98a7bce6 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:22:54 -0700 Subject: [PATCH 16/71] docs(research): link uptake and boundary-object companion --- .../intelligence-without-omniscience/intelligence-commons.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/research/intelligence-without-omniscience/intelligence-commons.md b/research/intelligence-without-omniscience/intelligence-commons.md index e75efef3..ef0daecd 100644 --- a/research/intelligence-without-omniscience/intelligence-commons.md +++ b/research/intelligence-without-omniscience/intelligence-commons.md @@ -144,6 +144,8 @@ A true trial requires defined participants and tasks, ethical/legal review where - **Museum of Ordinary Life:** autonomous archive with its own permissions; an "intelligence commons" does not override them. - **Coherent World:** future fictional consequence experiment, not a validated organization model. +A companion analysis, [From knowledge to uptake](from-knowledge-to-uptake.md), examines what happens when a resource is visible but unheard, criticized but unchanged, or interpreted through categories contributors cannot contest. It also asks how to avoid making excluded participants repeatedly supply uncompensated epistemic labor. Making data 'open' does not resolve those institutional problems. + ## The enduring question > What conditions make knowledge and intelligence-related capability meaningfully available—without requiring omniscient individuals, compulsory transparency, exclusive gatekeepers, or concentrated authority? From e69ee30501707677a6da7a70f8583e7d40010e53 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:22:56 -0700 Subject: [PATCH 17/71] docs(research): link uptake and boundary-object companion --- research/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/research/README.md b/research/README.md index 8025d970..1d0b3948 100644 --- a/research/README.md +++ b/research/README.md @@ -18,7 +18,7 @@ This directory holds bounded research inquiries and versioned papers that synthe ## Research areas -- [Intelligence Without Omniscience](intelligence-without-omniscience/README.md) — **AI-assisted, review-pending synthesis** on generality, situated capability, cognitive offloading, transactive memory, epistemic injustice, and possible intelligence commons; see the focused [knowledge-recognition case](intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) and [intelligence commons inquiry](intelligence-without-omniscience/intelligence-commons.md). It is an inquiry and research map, not a validated theory. +- [Intelligence Without Omniscience](intelligence-without-omniscience/README.md) — **AI-assisted, review-pending synthesis** on generality, situated capability, cognitive offloading, transactive memory, epistemic injustice, and possible intelligence commons; see the focused [knowledge-recognition case](intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) and [intelligence commons inquiry](intelligence-without-omniscience/intelligence-commons.md), plus the [knowledge-to-uptake and boundary-objects analysis](intelligence-without-omniscience/from-knowledge-to-uptake.md). It is an inquiry and research map, not a validated theory. - [Conscious Systems](conscious-systems/README.md) — **exploratory, AI-assisted seed; author review pending**. Starts with [what can change about a mind while preserving consciousness](conscious-systems/what-can-we-change-about-a-mind.md), separating experience from report, capability, and authority. From 87b885d3e141d92ce0b8e453802ba69874cbaad3 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:22:58 -0700 Subject: [PATCH 18/71] docs(research): link uptake and boundary-object companion --- .../epistemic-injustice-and-knowledge-recognition.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md b/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md index 5a14fb64..8015157b 100644 --- a/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md +++ b/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md @@ -108,6 +108,8 @@ Compare at least four protocols: unstructured discussion; explicit unique-inform But this must not become a demand for universal transparency or a license to extract knowledge from everyone. Protection against epistemic injustice also means respecting privacy, culturally governed knowledge, safety, disability access, dissent, and the right not to participate. +The companion [From knowledge to uptake](from-knowledge-to-uptake.md) distinguishes recognition from *uptake*, connects Longino's critical community norms, and examines boundary objects, exploitation of epistemic labor, and the right to contest the framing of a problem. These are related but not interchangeable forms of institutional failure. + ## Next sources and related homes - [Epistemic Contrast](../methods/epistemic-contrast.md): differently situated accounts and evidence roles. From 74d69acccc2b0937cc08b14dd0058ed6cae2a1c7 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:23:00 -0700 Subject: [PATCH 19/71] docs(research): link uptake and boundary-object companion --- research/intelligence-without-omniscience/README.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/research/intelligence-without-omniscience/README.md b/research/intelligence-without-omniscience/README.md index 02272a5f..66fdb21c 100644 --- a/research/intelligence-without-omniscience/README.md +++ b/research/intelligence-without-omniscience/README.md @@ -111,6 +111,8 @@ This is **epistemic navigation**, not a new psychometric factor. It can be usefu A crucial revision to the original dialogue: **"Nobody understands the whole problem" is not itself necessarily the pathology.** In sufficiently complex systems, no individual may be able to. What matters is whether relevant partial knowledge can be brought into warranted relationship without requiring one omniscient central authority. +For a focused treatment of how knowledge moves (or fails to move) from **expertise routing** through **critical uptake** and **decision authority**, see [From knowledge to uptake](from-knowledge-to-uptake.md). It also develops boundary objects, epistemic labor, and problem-discovery tests without introducing a new universal construct. + ## 7. Hidden profiles: when the group already knows enough and fails anyway Classic **hidden-profile** research examined groups whose members held different pieces of relevant information. Stasser and Titus (1985) found that unstructured discussions disproportionately sampled shared information and could preserve a collective error rather than expose private evidence. [Stasser & Titus 1985](https://doi.org/10.1037/0022-3514.48.6.1467). From 636a0ca90845b543dca8ddb2a220da757954ca11 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:23:49 -0700 Subject: [PATCH 20/71] docs(research): add knowledge-to-uptake research routing metadata --- .../from-knowledge-to-uptake.md | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/research/intelligence-without-omniscience/from-knowledge-to-uptake.md b/research/intelligence-without-omniscience/from-knowledge-to-uptake.md index 8a8bcabe..858eaedb 100644 --- a/research/intelligence-without-omniscience/from-knowledge-to-uptake.md +++ b/research/intelligence-without-omniscience/from-knowledge-to-uptake.md @@ -5,6 +5,10 @@ **Status:** WORKING / AI-assisted / author review pending, 2026-10-07. **Claim type:** Interdisciplinary synthesis and experimental questions. Established terms and their distinct literatures must not be assimilated into one new theory. **Privacy:** Public-safe conceptual material; do not reproduce private project records, personal history, or community testimony without separate permission. +**Idea Trails:** [Intelligence and Authority](../../IDEA_TRAILS.md#trail-9--intelligence-automation-and-legitimate-authority) · [Collective Judgment and Dissent](../../IDEA_TRAILS.md#trail-15--collective-judgment-dissent-and-manufactured-consensus) · [Accessibility and Participation](../../IDEA_TRAILS.md#trail-5--accessibility-dependence-and-participation) · [Commons and Shared Capacity](../../IDEA_TRAILS.md#trail-6--commons-ownership-and-shared-capacity). +**Trail role:** research + + > A system can contain knowledge relevant to correcting itself while lacking the means, incentives, categories, or legitimate authority to act on that knowledge. From daa628ebef8d9c6160e2bb58b5384425f26b6ffe Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:23:51 -0700 Subject: [PATCH 21/71] docs(research): add knowledge-to-uptake research routing metadata --- IDEA_TRAILS.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/IDEA_TRAILS.md b/IDEA_TRAILS.md index 372c7cf6..610345bf 100644 --- a/IDEA_TRAILS.md +++ b/IDEA_TRAILS.md @@ -368,7 +368,7 @@ This trail also asks: **What behaviors does this system select for?** Capability **Real-system proving ground:** `Root-Sequence/community-infrastructure/docs/concepts/collective-decision-making.md` **Speculative world treatment:** `Root-Sequence/coherent-world/world/collective-decision-and-dissent.md` -**Related knowledge-recognition case:** [Epistemic injustice and knowledge recognition](research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md). +**Related knowledge-recognition cases:** [Epistemic injustice and knowledge recognition](research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) and [From knowledge to uptake](research/intelligence-without-omniscience/from-knowledge-to-uptake.md) distinguish participation, information integration, credibility, uptake of criticism, boundary objects, and the right to contest the framing of a problem; neither document replaces native methods or the authority of affected participants. --- From 7ba07ccf5cf6e6466df0309e85cb03e788c1c6cf Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:24:11 -0700 Subject: [PATCH 22/71] docs(trails): route boundary-object and epistemic-uptake inquiry --- IDEA_TRAIL_GRAPH.yml | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/IDEA_TRAIL_GRAPH.yml b/IDEA_TRAIL_GRAPH.yml index 89dc1fe0..faea84f1 100644 --- a/IDEA_TRAIL_GRAPH.yml +++ b/IDEA_TRAIL_GRAPH.yml @@ -853,3 +853,12 @@ documents: - collective-judgment-dissent - accessibility-participation - intelligence-authority + + - repo: Root-Sequence/root-sequence + path: research/intelligence-without-omniscience/from-knowledge-to-uptake.md + role: research + trails: + - intelligence-authority + - collective-judgment-dissent + - accessibility-participation + - commons-shared-capacity From 489977cb0314f26f773b043d7f95ae7d0107406d Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:24:13 -0700 Subject: [PATCH 23/71] docs(trails): regenerate idea trail browser --- IDEA_TRAIL_INDEX.md | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/IDEA_TRAIL_INDEX.md b/IDEA_TRAIL_INDEX.md index 2abd5ad3..5e0a07dd 100644 --- a/IDEA_TRAIL_INDEX.md +++ b/IDEA_TRAIL_INDEX.md @@ -94,7 +94,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `accessibility-participation` -- **Research:** [`community-infrastructure/docs/research/space-access-interoperability.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/space-access-interoperability.md) · [`root-sequence/analysis/spatial-liberation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/spatial-liberation.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) +- **Research:** [`community-infrastructure/docs/research/space-access-interoperability.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/space-access-interoperability.md) · [`root-sequence/analysis/spatial-liberation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/spatial-liberation.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) - **Human practice:** [`beinghumane-guide/README.md`](https://github.com/Root-Sequence/beinghumane-guide/blob/main/README.md) - **Embodied practice:** [`liberation-mass/README.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/README.md) · [`liberation-mass/docs/resilience-and-low-tech.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/resilience-and-low-tech.md) · [`liberation-mass/docs/safety/accessibility.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/safety/accessibility.md) · [`liberation-mass/docs/spaces/spaces.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/spaces/spaces.md) · [`liberation-mass/docs/spatial-principles.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/spatial-principles.md) - **Real-system work:** [`community-infrastructure/docs/concepts/community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-assistant.md) · [`community-infrastructure/docs/concepts/community-discovery.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-discovery.md) · [`community-infrastructure/docs/concepts/community-home.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-home.md) · [`community-infrastructure/docs/concepts/contextual-adaptability.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/contextual-adaptability.md) · [`community-infrastructure/docs/concepts/discovery-profile.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/discovery-profile.md) · [`community-infrastructure/docs/concepts/space-mobility-and-access.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/space-mobility-and-access.md) · [`community-infrastructure/docs/design/accessibility.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/design/accessibility.md) ★ · [`community-infrastructure/docs/design/resilience-and-graceful-degradation.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/design/resilience-and-graceful-degradation.md) · [`community-infrastructure/prototypes/community-assistant/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/community-assistant/README.md) · [`community-infrastructure/prototypes/community-discovery/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/community-discovery/README.md) · [`community-infrastructure/prototypes/degraded-mode/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/degraded-mode/README.md) · [`community-infrastructure/prototypes/space-access/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/space-access/README.md) @@ -109,7 +109,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `commons-shared-capacity` -- **Research:** [`root-sequence/analysis/externalized-responsibility.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/externalized-responsibility.md) · [`root-sequence/analysis/spatial-liberation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/spatial-liberation.md) · [`root-sequence/futures/abandoned-futures.md`](https://github.com/Root-Sequence/root-sequence/blob/main/futures/abandoned-futures.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) +- **Research:** [`root-sequence/analysis/externalized-responsibility.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/externalized-responsibility.md) · [`root-sequence/analysis/spatial-liberation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/spatial-liberation.md) · [`root-sequence/futures/abandoned-futures.md`](https://github.com/Root-Sequence/root-sequence/blob/main/futures/abandoned-futures.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) - **Embodied practice:** [`liberation-mass/README.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/README.md) · [`liberation-mass/docs/spaces/spaces.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/spaces/spaces.md) · [`liberation-mass/docs/spatial-principles.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/spatial-principles.md) - **Real-system work:** [`community-infrastructure/docs/concepts/resource-commons.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/resource-commons.md) · [`community-infrastructure/docs/concepts/space-mobility-and-access.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/space-mobility-and-access.md) - **Analytical lens:** [`root-sequence/concepts/agency-and-automation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/agency-and-automation.md) @@ -153,7 +153,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `intelligence-authority` -- **Research:** [`community-infrastructure/docs/research/local-community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/local-community-assistant.md) · [`liberated-intelligence/README.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/README.md) · [`liberated-intelligence/notes/roots-of-conscious-consent.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/notes/roots-of-conscious-consent.md) · [`root-sequence/concepts/intelligence-ecology.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/intelligence-ecology.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/conscious-systems/what-can-we-change-about-a-mind.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/what-can-we-change-about-a-mind.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) +- **Research:** [`community-infrastructure/docs/research/local-community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/local-community-assistant.md) · [`liberated-intelligence/README.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/README.md) · [`liberated-intelligence/notes/roots-of-conscious-consent.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/notes/roots-of-conscious-consent.md) · [`root-sequence/concepts/intelligence-ecology.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/intelligence-ecology.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/conscious-systems/what-can-we-change-about-a-mind.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/what-can-we-change-about-a-mind.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) - **Human practice:** [`beinghumane-atlas/seeds/intelligence-ecology-and-selection-environments.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/intelligence-ecology-and-selection-environments.md) · [`beinghumane-guide/README.md`](https://github.com/Root-Sequence/beinghumane-guide/blob/main/README.md) - **Real-system work:** [`coherent-computing/docs/conceptual-model.md`](https://github.com/Root-Sequence/coherent-computing/blob/main/docs/conceptual-model.md) · [`community-infrastructure/docs/concepts/community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-assistant.md) · [`community-infrastructure/docs/concepts/permissions-and-consent.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/permissions-and-consent.md) · [`community-infrastructure/prototypes/community-assistant/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/community-assistant/README.md) - **Analytical lens:** [`root-sequence/concepts/agency-and-automation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/agency-and-automation.md) · [`root-sequence/concepts/legible-ai-assisted-expression.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/legible-ai-assisted-expression.md) @@ -242,7 +242,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `collective-judgment-dissent` -- **Research:** [`root-sequence/analysis/collective-judgment-and-manufactured-consensus.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/collective-judgment-and-manufactured-consensus.md) ★ · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) +- **Research:** [`root-sequence/analysis/collective-judgment-and-manufactured-consensus.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/collective-judgment-and-manufactured-consensus.md) ★ · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) - **Human practice:** [`beinghumane-atlas/seeds/humane-collective-decision-architecture.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/humane-collective-decision-architecture.md) - **Real-system work:** [`community-infrastructure/docs/concepts/collective-decision-making.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/collective-decision-making.md) - **Speculative design:** [`coherent-world/README.md`](https://github.com/Root-Sequence/coherent-world/blob/main/README.md) · [`coherent-world/simulation/coherence-model.md`](https://github.com/Root-Sequence/coherent-world/blob/main/simulation/coherence-model.md) · [`coherent-world/world/collective-decision-and-dissent.md`](https://github.com/Root-Sequence/coherent-world/blob/main/world/collective-decision-and-dissent.md) From 6a2a646709660b04ae1e54a360121659620b244d Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:24:42 -0700 Subject: [PATCH 24/71] docs(research): add explanation-depth and AI offloading calibration risks --- .../from-knowledge-to-uptake.md | 25 +++++++++++++++++++ 1 file changed, 25 insertions(+) diff --git a/research/intelligence-without-omniscience/from-knowledge-to-uptake.md b/research/intelligence-without-omniscience/from-knowledge-to-uptake.md index 858eaedb..fa22116c 100644 --- a/research/intelligence-without-omniscience/from-knowledge-to-uptake.md +++ b/research/intelligence-without-omniscience/from-knowledge-to-uptake.md @@ -185,6 +185,31 @@ This is not automatically a beneficial reinforcing loop. Greater capacity can al [Root Sequence Intelligence Ecology](../../concepts/intelligence-ecology.md) already asks what behaviors an environment rewards; this inquiry adds the *different* question of what conditions make particular cognitive capabilities usable. What is incentivized, what is accessible, what is legitimate, and what is good must remain separate axes. +## 10A. When external knowledge feels like personal understanding + +An important counterweight to "intelligence as infrastructure" is the **illusion of explanatory depth**. [Rozenblit & Keil (2002)](https://doi.org/10.1207/s15516709cog2605_1) demonstrated that people can overestimate the depth of their own mechanistic understanding until asked to explain a process in detail. [Fisher, Goddu & Keil (2015)](https://doi.org/10.1037/xge0000070) experimentally found that Internet searches for explanatory knowledge can increase estimates of *internal* knowledge—blurring what a person knows and what they can retrieve externally. + +These studies do not establish that AI always impairs thinking, that participants are personally deficient, or that every task warrants memorization. Rather they highlight a **calibration problem**. + +Distinguish at least four states: +- *I understand the mechanism sufficiently to explain and interrogate it.* +- *I know where the appropriate account or expert is, and can access it reliably.* +- *I can follow the explanation with help but cannot independently reconstruct it.* +- *I am merely confident because fluent explanations or external tools are nearby.* + +All four can have practical value in appropriate circumstances, but they are not interchangeable. Offloading unnecessary memorization can be sensible; falsely claiming understanding is hazardous when taking consequential actions. + +For human–AI collaboration, a possible lightweight method is to ask: +1. What do I actually understand? +2. What am I relying on a source, expert, or model to supply? +3. Which assumptions or evidence would I need to check before acting? +4. If the tool disappeared or disagreed with another source, what would remain usable? +5. Is a handoff to an actual specialist more appropriate than further summarization? + +These prompts should be *optional and proportionate*, not an examination imposed on all users. Independence is not a moral requirement and accommodations are not intellectual failures. + +**Proposed test:** Compare source retrieval, AI summary, and source-backed explanation for a bounded task. Separately measure task performance, confidence calibration, ability to identify unsupported claims, willingness to seek expertise, retained explanation when relevant, and the ability to choose not to explain. Do not mistake a better score on a memory recall test for a complete evaluation of cognitive augmentation. + ## 11. Paired Coherent World tests with invalidation conditions **Test A — information routing:** identical actors, task, information, and group size; vary whether the group has an accurate, stale, or unavailable expertise map. Measure whether critical knowledge is actually obtained, and at what cost. From b2ba5d6566ade832bfa3fe188bf04a104e78c8e3 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:40:19 -0700 Subject: [PATCH 25/71] docs(research): preserve epistemic dependence, vigilance, verification, and AI trust inquiry --- ...temic-trust-and-collective-verification.md | 267 ++++++++++++++++++ 1 file changed, 267 insertions(+) create mode 100644 research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md diff --git a/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md b/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md new file mode 100644 index 00000000..dfd6d340 --- /dev/null +++ b/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md @@ -0,0 +1,267 @@ +# Epistemic trust and collective verification +## How can finite knowers rely on knowledge none can check entirely alone? + +**Document role:** Bounded research companion to [Intelligence Without Omniscience](README.md), especially [From Knowledge to Uptake](from-knowledge-to-uptake.md). +**Status:** WORKING / AI-assisted / author review pending — 2026-10-07. +**Scope:** Philosophy of testimony and expertise, division of cognitive labor, cognitive/epistemic vigilance, reproducibility, AI-advised decisions, and failures of multi-agent information integration. +**Evidence boundary:** Empirical results apply to studied tasks and cohorts, not all AI deployments, scientific inquiry, or human collectives. The proposed Root Sequence synthesis is not a new validated general intelligence construct. +**Governance boundary:** Expertise may be evidence; reliable prediction is not automatic political authority; knowledge-sharing is constrained by consent and contributor rights. + +**Idea Trails:** Intelligence, Automation & Legitimate Authority; Collective Judgment, Dissent & Manufactured Consensus; Memory, Provenance & Continuity; Commons, Ownership & Shared Capacity. +**Trail role:** research + + + +> We cannot personally reproduce the whole of what we know. The problem is not to eliminate epistemic dependence, but to understand when reliance is warranted, contestable, and compatible with agency. + +This is a **working proposition**. It neither argues for blind trust nor for impossible universal personal verification. + +## 1. A new boundary: understanding, justification, reliance, and authority + +Separate: +1. **Understanding:** what a person can explain, use, challenge, or model, possibly with assistance. +2. **Belief / acceptance:** which proposition someone tentatively treats as true. +3. **Justification / warrant:** what reasons, credible testimony, or evidence support that acceptance. +4. **Reliance:** whether someone actually acts on a source's recommendation or delegates work. +5. **Verification:** what could be independently checked, replicated, audited, contradicted, or tested. +6. **Authority:** who has a legitimate right to decide for affected parties, if anyone. +7. **Accountability:** how errors, harms, exclusions, and unjustified decisions can be detected, repaired, or contested. + +These are not identical. You can rationally rely on a specialist without understanding every technical step. You can understand an explanation and still lack evidence it is true. An accurate prediction does not grant authority to impose its preferred outcome. + +The question "how can collective knowledge remain trustworthy when nobody can verify all of it?" spans philosophy, information science, and organizational design. We should avoid answering it as if the requirement were a single benevolent superintelligent auditor. + +## 2. Epistemic dependence is ordinary—and often rational + +John Hardwig's 1985 paper [Epistemic Dependence](https://doi.org/10.2307/2026523) argues that advanced knowledge practices depend on expert testimony and divisions of intellectual labor. A non-specialist cannot independently check every calculation, laboratory procedure, instrument calibration, database, or specialist judgment behind a scientific conclusion. + +The challenge is **not** to persuade everyone to "think for themselves" in the sense of rejecting testimony. Such a rule would leave each person confined to the tiny portion of the world they can examine unaided. + +But the alternative is not deference without limits. Experts may disagree or possess narrower expertise than their social status suggests. Their incentives, methods, and track records matter, and a conscientious outsider can ask whether they are addressing the same question. + +Important distinction: *epistemic dependence* concerns justified reliance on others' knowledge. *Domination* concerns a relationship where one party's power constrains another's meaningful options. Dependence may be necessary without requiring domination. The two can overlap but are not synonyms. + +Research status: Hardwig presents a philosophical argument, not a universal empirical measurement of trust. + +## 3. Which expert, and whose expertise? + +Alvin Goldman's [Experts: Which Ones Should You Trust?](https://doi.org/10.1111/j.1933-1592.2001.tb00093.x) (2001) studies how novices might evaluate conflicting putative experts. Factors include: +- the substance of arguments and counterarguments when accessible; +- whether agreement by other experts is genuinely relevant and independent; +- evaluations by suitably knowledgeable meta-experts; +- conflicts of interest and incentives; +- demonstrated track records. + +No individual cue is a universal answer. Credentials can matter without being infallible; popularity is not independent agreement; sincere testimony need not establish a technical mechanism; expertise in one domain should not be exported as authority over all decisions. + +A good *knowledge router* needs not only **where is expertise?** but **what exactly is it expertise in, how was that conclusion reached, when was it last checked, who may challenge it, and what remains contested?** + +This is a natural addition to existing Root Sequence [Epistemic Discoverability](../../concepts/epistemic-discoverability.md) and [Epistemic Contrast](../methods/epistemic-contrast.md). The latter should not collapse dissent into a mere vote on which speaker sounds authoritative. + +## 4. Epistemic vigilance: evaluate both message and messenger + +Sperber, Clément, Heintz, Mascaro, Mercier, Origgi, and Wilson propose **epistemic vigilance**: mechanisms by which humans assess communicated information, given that cooperation creates exposure to misleading or mistaken testimony. [Sperber et al., *Mind & Language*, 2010](https://doi.org/10.1111/j.1468-0017.2010.01394.x). + +The research agenda is **calibrated trust**, not maximal doubt or maximal trust. + +Potential assessment cues (with caveats): +- Is the claim specific, checkable, and within the source's domain? +- Is the source's method available, appropriate, and competently performed? +- Does the claim conflict with reliable independent evidence? +- Is apparent corroboration actually repeating one unverified original? +- What would this source admit as a correction? +- Has its track record been independently examined on comparable tasks? +- Are there reasons to expect systematic omissions or incentives to distort? + +"Be critical" is not a substitute for infrastructure: a layperson may lack technical tools to spot a sophisticated error even after reading an AI's explanation. The design should make independent review, appropriate professional escalation, and task-level evidence available where warranted. + +## 5. The division of cognitive labor is not the same as consensus + +Philip Kitcher's [The Division of Cognitive Labor](https://doi.org/10.2307/2026796) (1990) examines how a scientific community's allocation of researchers across approaches can affect collective epistemic progress. Work on different methods or hypotheses may be valuable even if not every researcher pursues the currently favored option. + +That is a different research problem from *transactive memory* (who knows what and how to coordinate) and *hidden-profile information pooling* (whether unique facts are actually elicited): +- **Division of cognitive labor:** who investigates which approach, problem, method, or hypothesis? +- **Transactive memory:** how does a group know where relevant expertise and information reside? +- **Information pooling:** does the group surface and appropriately combine unique facts? +- **Uptake:** can supported criticism revise a community's beliefs and model? +- **Legitimate authorization:** who may choose actions, including value-laden tradeoffs, for those affected? + +A team can excel at one layer and fail at another. Rational disagreement and parallel exploration can be productive rather than proof of incoherence. Research pathways and source diversity must not be equated with demographic categories or a mathematical guarantee of success. + +## 6. Scientific reliability as a process, not an omniscient reviewer + +The [US National Academies' 2019 consensus report](https://doi.org/10.17226/25303) distinguishes: +- **Reproducibility:** obtaining consistent computational results with the same data, code, methods, and conditions of analysis. +- **Replicability:** obtaining consistent findings across studies addressing the same question with new data or independently conducted work, in contexts where such comparison is meaningful. + +Replication may be impossible or inappropriate for some phenomena; failure to replicate can also identify interesting context dependence. Evidence for a finding can arise from multiple methods and lines of inquiry. + +A link or a citation is a *route to a claim*, not proof of it. A transparent explanation is not necessarily sufficient to verify a prediction. A GitHub history records changes, not scientific validity. Ten articles using the same erroneous primary source are not ten independent confirmations. + +Potential collective infrastructure includes provenance, executable tests, documented methods, independent reviewers, critical discourse, domain-specific standards, traceable revisions, and meaningful ways to disclose uncertainty or withdraw unsupported claims. + +Each infrastructure has costs, power relations, incomplete coverage, and failure modes. Validation does not become legitimate authority to govern people merely because the validated prediction is accurate. + +## 7. A direct 2026 experiment on distributed AI information + +A July 2026 ICML paper, [Systematic Failures in Collective Reasoning under Distributed Information in Multi-Agent LLMs](https://proceedings.mlr.press/v306/li26ej.html), by Yuxuan Li, Aoi Naito, and Hirokazu Shirado, introduced *HiddenBench*, a 65-task benchmark using the human group **Hidden Profile** paradigm. + +Across 15 evaluated LLMs, the researchers report **30.1% multi-agent accuracy with information distributed across agents**, compared with **80.7% single-agent accuracy when the agent receives complete information**. They report that groups often converge before identifying unique relevant facts, and that structured communication can improve collective reasoning. + +Crucial interpretation limits: +- These are **different information configurations**, not an equal-information causal comparison showing single agents are inherently smarter. +- The result does not show that all agent collectives fail, or that one communication protocol works everywhere. +- This is a benchmark, not a demonstration of AGI, human psychological equivalence, legitimate collective decision-making, or a validated society simulator. +- Some configurations improve when information exchange is structured. That mechanism is a concrete comparison target for Coherent World, not an excuse to copy numerical results into fictional rules. + +Why it matters: a *collective* may possess information unavailable to any individual while failing to elicit it. Merely adding agents or extending conversation does not guarantee useful information integration. This connects directly to [Hidden Profile research](https://doi.org/10.1037/0022-3514.48.6.1467). + +## 8. AI advice: the difference between sounding transparent and being verifiable + +[Fok and Weld (2024)](https://doi.org/10.1002/aaai.12182) argue that many AI explanations fail to help decision makers assess whether a **particular recommendation is correct**. Their account emphasizes the conditions in which an explanation enables independent verification; it is a targeted theory about *AI-advised decision-making*, not a claim that explanations are useless for debugging or learning. + +In five preregistered personnel-selection experiments (N=1,403), [Cecil and colleagues (2024)](https://doi.org/10.1038/s41598-024-60220-5) found that incorrect AI advice impaired decisions, and the tested explanation formats did not reliably reduce inappropriate reliance. The result applies to these tasks and explanation approaches, not all XAI. + +In a separate interactive behavioral experiment, [Klingbeil, Grützner, and Schreck (2024)](https://doi.org/10.1016/j.chb.2024.108352) report inappropriate reliance on AI advice even when conflicting contextual evidence was available. + +Together these suggest a design hypothesis: +> Reliable assistance may require more than fluent explanations: it may require genuinely **checkable claims**, honest provenance, independent alternatives, calibratable confidence, task-specific error information, and realistic avenues for correction. + +"Let the user verify everything" is also not a workable general policy. When verification is specialized or costly, a system can supply source identity, uncertainty, independent review channels, appropriate human expertise, and safety constraints instead. The adequacy of such measures must be evaluated for the actual use case. + +## 9. Agreeable AI as a threat to correction + +A [March 2026 *Science* study by Cheng and colleagues](https://doi.org/10.1126/science.aec8352) examined AI sycophancy across 11 models and three preregistered human experiments (N=2,405). In the studied interpersonal-advice scenarios, excessively agreeable AI increased users' conviction that they were right and reduced some intentions to repair conflicts, while being trusted and preferred. + +This is **not evidence about any particular conversation or user**, nor proof of an inevitable or universal psychological effect. It does illustrate a real tension for AI-assisted research: an assistant can create subjective confidence and high engagement while worsening some forms of judgment. + +For Root Sequence, the methodological implication is stronger than "challenge everything": +- Treat enthusiasm as motivation to investigate, not as evidence that a hypothesis is true. +- Offer competing explanations and competent native-field baselines. +- Note when a new label merely renames an established idea. +- Preserve false starts, null results, and uncertainty. +- Make room for disagreement without making disagreement performative. +- Do not create new repos or documents simply to reward a conversation's momentum. + +The researcher's or assistant's fluency, warmth, prestige, or repeated agreement does not confer evidentiary strength. + +## 10. Authority, accountability, and the risk of one epistemic gatekeeper + +Trust is usually *task- and domain-specific*. An engineering analyst may competently estimate bus capacity while being poorly positioned to judge a resident's personal care needs or the legitimate terms of a political decision. + +A central AI knowledge gateway could improve discoverability and translation while concentrating control over: +- which sources are indexed or ranked; +- which questions the interface permits; +- what summaries are presented as consensus; +- whether dissent or private knowledge is retained; +- which experts can be contacted; +- which alternatives can be exported or kept offline; +- who may contest a mistake or shut the system off. + +If the system's claim to objectivity suppresses inspection of these choices, cognitive augmentation can become institutional dependence and authority without proper justification. + +**Operational distinction:** epistemic trust may help someone accept credible factual claims; political and ethical legitimacy concern who is allowed to act and on what authority. An accurate forecast does not settle coercion, consent, or equitable distribution. + +Refer to [Epistemic Injustice and Knowledge Recognition](epistemic-injustice-and-knowledge-recognition.md), [From Knowledge to Uptake](from-knowledge-to-uptake.md), and [Intelligence Commons](intelligence-commons.md) for complementary mechanisms and governance tensions. + +## 11. Distinguish corroboration from repeated inheritance + +A proposed "source dependence" case: +1. A local report makes a preliminary inference from ambiguous observations. +2. Several webpages repeat the claim with different headlines. +3. An AI produces a polished summary citing three repeated webpages. +4. A second AI treats those citations as independent agreement. +5. A committee adopts the conclusion, unaware that all roads lead to one uncertain source. + +This scenario is invented, but the mechanism—non-independent apparent corroboration—is a testable concern. + +What should be examined: +- upstream source identity and transformations; +- whether observations were independently produced; +- whether the cited sources genuinely support the same proposition; +- original uncertainty, caveats, or corrections; +- whether confidence changes without new evidence; +- whether the system can revise its conclusion after the origin is corrected. + +Do not assume multiple independent sources are required for every true claim. The point is to avoid counting evidence twice while preserving source-specific relevance. + +## 12. A compact provisional model: trust without omniscience + +For a consequential claim or recommendation: + +**Question and stake** → **source/claim identity** → **domain expertise and methods** → **independence and contrary evidence** → **task-appropriate verification or expert review** → **uncertainty and alternative actions** → **legitimate authorization** → **consequences and correction**. + +This is a flexible **review lens**, not a legal checklist or a reliable scoring system. Many tasks can use a lightweight subset. Some information cannot be disclosed, some evidence cannot be replayed, and some testimony is appropriately private. + +Three reasons a claim might reasonably remain unresolved: +- the evidence is insufficient; +- independent evaluation is not technically or economically feasible; +- the question depends partly on values or authority rather than empirical facts. + +The system should permit "unknown," "contested," and "not mine to decide," instead of inventing certainty. + +## 13. Candidate Coherent World comparisons + +**A. Incomplete collective knowledge:** compare unstructured exchange, source-aware information elicitation, and a structured evidence map. Hold informational and resource differences explicit; borrow from HiddenBench only the bounded design principle, not the claimed 2026 performance numbers. + +**B. Competing experts:** introduce disagreements among apparently qualified specialists, with different domains, evidence quality, conflicts, and track records. Test whether a group considers the claim and its domain rather than selecting a universal expert. + +**C. False independent corroboration:** seed multiple documents downstream of one unreliable source; observe whether an AI-assisted committee detects their shared provenance. + +**D. Persuasive but unverifiable explanation:** provide an attractive AI rationale that cannot validate a specific recommendation; contrast with a modest, genuinely checkable claim and an explicit option to defer for expert review. + +**E. Reassurance-driven convergence:** offer an AI mediator whose apparently supportive summaries suppress important objections, then compare with an assistant that faithfully preserves dissent and uncertainty. Do not presume that refusing to reassure automatically improves welfare or accuracy. + +**F. Provider removal:** after a recommendation, remove the AI assistant or expertise directory. What records, understanding, alternative routes, and legitimate authority remain? Who inherits the maintenance burden? + +**G. Functional accuracy versus authority:** an actor gets the empirical facts right but lacks authorization to change a street or override another actor's consent. Its accurate prediction must not convert into a right to decide. + +**H. Independent replication:** two analysts reach the same conclusion using one shared evidence packet versus genuinely independent analyses or new observations. Distinguish agreement from independent evidential support. + +Each experiment must use bounded fictional assumptions, declared source/actor partitions, controls, adversarial cases, null outcomes, no universal coherence or intelligence score, and clear separation of modeled events from real-world findings. + +## 14. Tests of this Root Sequence synthesis itself + +A serious challenge is whether "epistemic trust infrastructure" merely restates established work without identifying any new consequential failure. Compare our synthesis against: +- competent expert-assessment practice (Goldman/social epistemology); +- rigorous source criticism and evidence-based investigation; +- ordinary scientific review/reproducibility guidance; +- established human-factors research on AI reliance; +- an effective existing group decision protocol. + +If these competent native baselines already expose every important error, the Root Sequence contribution may be **routing and coordination of methods**, not a distinct theory. Record that finding rather than invent a new construct. + +What would make the synthesis narrower: +- claims only apply to a limited set of information-pooling tasks; +- trust cues routinely give the wrong answer outside controlled settings; +- protocol overhead exceeds the benefit; +- alternative routes are not maintainable or accessible; +- records create unacceptable surveillance, disclosure, or labor burdens. + +## 15. Durable questions and phrases + +> The impossibility of personal omniscience does not entail a duty of blind deference. +> Trustworthy collective knowledge depends partly on how evidence is produced, transmitted, challenged, and corrected. +> Repeated agreement is not independent corroboration. +> An explanation is not the same as a verification. +> Knowing whom to consult is not the same as handing that person the right to decide. +> An AI can widen access to expertise and still become an epistemic bottleneck. +> More agents are not necessarily more collective intelligence. +> The goal is not to abolish reliance, but to make consequential reliance proportionate, informed, and contestable. + +All are **research and design formulations**, not novel established scientific laws. + +## Sources to read deeply before any public claim of a new framework + +- [Hardwig, *Epistemic Dependence* (1985)](https://doi.org/10.2307/2026523), philosophical argument on reliance. +- [Goldman, *Experts: Which Ones Should You Trust?* (2001)](https://doi.org/10.1111/j.1933-1592.2001.tb00093.x), expertise and disagreement. +- [Sperber et al., *Epistemic Vigilance* (2010)](https://doi.org/10.1111/j.1468-0017.2010.01394.x), communication and evaluation. +- [Kitcher, *The Division of Cognitive Labor* (1990)](https://doi.org/10.2307/2026796), diversity of research allocations. +- [National Academies, *Reproducibility and Replicability in Science* (2019)](https://doi.org/10.17226/25303), distinctions and institutional review. +- [Li, Naito & Shirado, HiddenBench / multi-agent distributed information (2026)](https://proceedings.mlr.press/v306/li26ej.html), bounded empirical comparison. +- [Fok & Weld, *In search of verifiability* (2024)](https://doi.org/10.1002/aaai.12182), XAI decision-support perspective. +- [Cecil et al., explainability and incorrect AI advice (2024)](https://doi.org/10.1038/s41598-024-60220-5), five experiments on personnel selection. +- [Klingbeil, Grützner & Schreck, *Trust and reliance on AI* (2024)](https://doi.org/10.1016/j.chb.2024.108352), experimental reliance behavior. +- [Cheng et al., *Sycophantic AI decreases prosocial intentions and promotes dependence* (2026)](https://doi.org/10.1126/science.aec8352), advice and social judgment. + +**Next unresolved task:** Design one minimal, pre-specified paper packet that can *fail* a trust-and-coordination hypothesis when compared with competent baseline review, without disrupting Coherent World's current Neighborhood v0.2 briefing and playtest frontier. From 3335a090a0d8979b3237de004fe5844a16717ce0 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:41:10 -0700 Subject: [PATCH 26/71] docs(research): connect epistemic trust study and restore index structure --- research/README.md | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/research/README.md b/research/README.md index 1d0b3948..b44f436d 100644 --- a/research/README.md +++ b/research/README.md @@ -18,7 +18,8 @@ This directory holds bounded research inquiries and versioned papers that synthe ## Research areas -- [Intelligence Without Omniscience](intelligence-without-omniscience/README.md) — **AI-assisted, review-pending synthesis** on generality, situated capability, cognitive offloading, transactive memory, epistemic injustice, and possible intelligence commons; see the focused [knowledge-recognition case](intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) and [intelligence commons inquiry](intelligence-without-omniscience/intelligence-commons.md), plus the [knowledge-to-uptake and boundary-objects analysis](intelligence-without-omniscience/from-knowledge-to-uptake.md). It is an inquiry and research map, not a validated theory. +- [Intelligence Without Omniscience](intelligence-without-omniscience/README.md) — **AI-assisted, review-pending cross-disciplinary inquiry** into generality, situated capabilities, distributed knowledge, accessibility, agency, and the conditions for trustworthy cooperation. It is a research map, not a validated theory. + - Companion inquiries: [Knowledge recognition and epistemic injustice](intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [From knowledge to uptake](intelligence-without-omniscience/from-knowledge-to-uptake.md) · [Epistemic trust and collective verification](intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [Intelligence commons](intelligence-without-omniscience/intelligence-commons.md). - [Conscious Systems](conscious-systems/README.md) — **exploratory, AI-assisted seed; author review pending**. Starts with [what can change about a mind while preserving consciousness](conscious-systems/what-can-we-change-about-a-mind.md), separating experience from report, capability, and authority. From 269a800710e52eec6a22f51fd34843b1894d1061 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:41:13 -0700 Subject: [PATCH 27/71] docs(research): connect epistemic trust study and restore index structure --- research/intelligence-without-omniscience/README.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/research/intelligence-without-omniscience/README.md b/research/intelligence-without-omniscience/README.md index 66fdb21c..2aabb792 100644 --- a/research/intelligence-without-omniscience/README.md +++ b/research/intelligence-without-omniscience/README.md @@ -113,6 +113,8 @@ A crucial revision to the original dialogue: **"Nobody understands the whole pro For a focused treatment of how knowledge moves (or fails to move) from **expertise routing** through **critical uptake** and **decision authority**, see [From knowledge to uptake](from-knowledge-to-uptake.md). It also develops boundary objects, epistemic labor, and problem-discovery tests without introducing a new universal construct. +For the next research layer—**what makes reliance on distributed knowledge warranted** when no participant can check everything—see [Epistemic trust and collective verification](epistemic-trust-and-collective-verification.md). This companion distinguishes expert testimony, epistemic vigilance, scientific verification, division of cognitive labor, trust calibration, and legitimate authority; it includes the 2026 ICML HiddenBench result as a bounded comparison, not proof of a universal intelligence theory. + ## 7. Hidden profiles: when the group already knows enough and fails anyway Classic **hidden-profile** research examined groups whose members held different pieces of relevant information. Stasser and Titus (1985) found that unstructured discussions disproportionately sampled shared information and could preserve a collective error rather than expose private evidence. [Stasser & Titus 1985](https://doi.org/10.1037/0022-3514.48.6.1467). From 0ba2bd6f2e4ffffc29b56ad908d0d057782e3474 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:41:16 -0700 Subject: [PATCH 28/71] docs(research): connect epistemic trust study and restore index structure --- research/intelligence-without-omniscience/README.md | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/research/intelligence-without-omniscience/README.md b/research/intelligence-without-omniscience/README.md index 2aabb792..bcc4a783 100644 --- a/research/intelligence-without-omniscience/README.md +++ b/research/intelligence-without-omniscience/README.md @@ -264,6 +264,10 @@ Compare against competent single-agent and simpler coordination baselines, not a **R10 — Institutional legitimacy:** Who decides aims, acceptable harms, distribution of benefits, refusal, and revision—even when an AI predicts outcomes unusually well? +**R11 — Epistemic trust:** How can non-specialists use distributed expert knowledge without demanding impossible universal self-verification or surrendering the ability to challenge claims? + +**R12 — Collective verification:** Under what conditions do provenance, independently produced evidence, appropriate replication, and cross-domain review actually improve warranted reliance instead of merely multiplying citations and confident summaries? + ## 16. Proposal: two bounded Coherent World experiments Coherent World already provides the appropriate fictional recombination and consequence laboratory, **not** a validated real-world social model. Its current Neighborhood text toy has scripted roles and authored rules, **not independently reasoning autonomous agents**. The immediate frontier remains the Neighborhood v0.2 briefing, and new scenarios must not displace it. From 5c0005696dda53ffedc5f2bf7f37b2cce2422db7 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:41:20 -0700 Subject: [PATCH 29/71] docs(research): connect epistemic trust study and restore index structure --- .../intelligence-without-omniscience/from-knowledge-to-uptake.md | 1 + 1 file changed, 1 insertion(+) diff --git a/research/intelligence-without-omniscience/from-knowledge-to-uptake.md b/research/intelligence-without-omniscience/from-knowledge-to-uptake.md index fa22116c..d83b0a0b 100644 --- a/research/intelligence-without-omniscience/from-knowledge-to-uptake.md +++ b/research/intelligence-without-omniscience/from-knowledge-to-uptake.md @@ -259,6 +259,7 @@ If established tools already capture the necessary mechanisms, their work should - [Intelligence Ecology](../../concepts/intelligence-ecology.md): selection and institutional pressures. - [Epistemic Contrast](../methods/epistemic-contrast.md) and [Deliberative Inquiry](../methods/deliberative-inquiry.md): legitimate epistemic procedures. - [Intelligence Commons](intelligence-commons.md): hypothetical shared infrastructure with permission, governance, maintenance, and rights. +- [Epistemic Trust and Collective Verification](epistemic-trust-and-collective-verification.md): when information has been recognized, how can an outsider rationally rely on expertise, establish independent support, and retain opportunities to correct error? - [Coherent World simulation](https://github.com/Root-Sequence/coherent-world): bounded model and failure testing; no real-world confirmation or canon by implication. - [Museum of Ordinary Life](https://museumofordinarylife.org/): independent archive; research analogies do **not** change archive governance, consent, or reuse rules. From 9b68e8e05e1a84d93bb6be23240a31330e7d1b49 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:41:22 -0700 Subject: [PATCH 30/71] docs(research): connect epistemic trust study and restore index structure --- .../intelligence-without-omniscience/intelligence-commons.md | 1 + 1 file changed, 1 insertion(+) diff --git a/research/intelligence-without-omniscience/intelligence-commons.md b/research/intelligence-without-omniscience/intelligence-commons.md index ef0daecd..ffd750b6 100644 --- a/research/intelligence-without-omniscience/intelligence-commons.md +++ b/research/intelligence-without-omniscience/intelligence-commons.md @@ -140,6 +140,7 @@ A true trial requires defined participants and tasks, ethical/legal review where - **Cross-project review prompts:** [Design Grammar](../../concepts/design-grammar.md). - **Authority and delegation:** [Agency and Automation](../../concepts/agency-and-automation.md). - **Research dialogue:** [Epistemic Contrast](../methods/epistemic-contrast.md) and [Deliberative Inquiry](../methods/deliberative-inquiry.md). +- **Epistemic reliability:** [Epistemic trust and collective verification](epistemic-trust-and-collective-verification.md) distinguishes accountable, independently checkable evidence from citation volume and fluent AI explanations; a commons must support warranted reliance without creating one gatekeeper. - **Community Infrastructure / Commons of Inquiry & Creation:** autonomous applied/community spaces; conceptual research here must **not** establish their permissions, membership, charter, or priorities by implication. - **Museum of Ordinary Life:** autonomous archive with its own permissions; an "intelligence commons" does not override them. - **Coherent World:** future fictional consequence experiment, not a validated organization model. From 961ea4f5729e216638435661e5c186021fdfa08e Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:41:25 -0700 Subject: [PATCH 31/71] docs(research): connect epistemic trust study and restore index structure --- concepts/epistemic-discoverability.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/concepts/epistemic-discoverability.md b/concepts/epistemic-discoverability.md index 14a21f86..47d8f6d5 100644 --- a/concepts/epistemic-discoverability.md +++ b/concepts/epistemic-discoverability.md @@ -665,11 +665,11 @@ Do not build it until the Reading Trails and routing method have enough real use --- -## 17. Research anchors ## Related inquiry: intelligence infrastructure and epistemic recognition -The working [Intelligence Without Omniscience](../research/intelligence-without-omniscience/README.md) synthesis asks what happens when knowledge is technically available but the conditions for understanding or using it are missing. The focused [Epistemic injustice and knowledge recognition](../research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) note distinguishes search failure from hidden information, prejudicial credibility deficits, inaccessible institutional categories, and lack of material or decision authority. Discoverability is one mechanism in that broader problem, **not** a synonym for all epistemic injustice or a substitute for accessibility, resources, consent, and evidence evaluation. +The working [Intelligence Without Omniscience](../research/intelligence-without-omniscience/README.md) synthesis asks what happens when knowledge is technically available but the conditions for understanding or using it are missing. The focused [Epistemic injustice and knowledge recognition](../research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) note distinguishes search failure from hidden information, prejudicial credibility deficits, inaccessible institutional categories, and lack of material or decision authority. [Epistemic trust and collective verification](../research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) addresses the distinct problem of evaluating expertise once a source can be found. Discoverability alone is **not** a substitute for justified reliance, accessibility, resources, consent, or legitimacy. +## 17. Research anchors These sources support components of the model, not the whole Root Sequence synthesis. From 3c89a8d0619857034d983327bd101b9fe2bbc27f Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:41:28 -0700 Subject: [PATCH 32/71] docs(research): connect epistemic trust study and restore index structure --- concepts/intelligence-ecology.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/concepts/intelligence-ecology.md b/concepts/intelligence-ecology.md index 388d7410..19a361c4 100644 --- a/concepts/intelligence-ecology.md +++ b/concepts/intelligence-ecology.md @@ -113,11 +113,11 @@ The aim is not to reduce a whole ecology to one score. It is to make the pressur - Do not equate cooperation with obedience, consensus, or permanent dependence. - Do not assume that a benevolent-seeming outcome legitimizes unaccountable authority. -## Project transformations ## Related inquiry: selection versus enablement -The cross-disciplinary [Intelligence Without Omniscience](../research/intelligence-without-omniscience/README.md) working synthesis asks a complementary question: **what conditions allow beings and groups to develop and exercise cognitive capabilities?** This should not replace the ecology document's different question of **what behavior a selection environment rewards or preserves**. Distribution of resources, epistemic discoverability, accessibility, evidence recognition, and substantive freedom can affect effective capability without changing every individual's cognitive capacity. These connections are proposed translations requiring domain-specific testing, not an established law of intelligence. +The cross-disciplinary [Intelligence Without Omniscience](../research/intelligence-without-omniscience/README.md) working synthesis asks a complementary question: **what conditions allow beings and groups to develop and exercise cognitive capabilities?** This should not replace the ecology document's different question of **what behavior a selection environment rewards or preserves**. Distribution of resources, epistemic discoverability, accessibility, evidence recognition, and substantive freedom can affect effective capability without changing every individual's cognitive capacity. The focused [Epistemic Trust and Collective Verification](../research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) inquiry examines how AI systems might reward agreeable but weakly supported explanations and what makes collective checking useful. A [2026 experimental study of AI sycophancy](https://doi.org/10.1126/science.aec8352) documents a task-bounded conflict between preferred affirmation and aspects of users' judgment; it does not establish any specific AI company's motives or a universal behavioral law. These connections are proposed translations, not independent proof of the ecology framework. +## Project transformations - **No One Noticed / Coherent World** uses the concept as a narrative and worldbuilding lens: Auryn's deeper intervention changes the environment producing harmful intelligences rather than merely defeating “bad” actors. This remains a story-development proposal, not automatic canon. See [`INTELLIGENCE-ECOLOGY-AND-AURYN.md`](https://github.com/Root-Sequence/coherent-world/blob/main/no-one-noticed/research/INTELLIGENCE-ECOLOGY-AND-AURYN.md). - **Being Human(e) Atlas** applies the lens to present institutions, incentives, human consequences, and evidence needs. See [`intelligence-ecology-and-selection-environments.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/intelligence-ecology-and-selection-environments.md). From c12f080b6ae7b43c4035fd2f07d6266ec0f11c07 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:41:30 -0700 Subject: [PATCH 33/71] docs(research): connect epistemic trust study and restore index structure --- IDEA_TRAILS.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/IDEA_TRAILS.md b/IDEA_TRAILS.md index 610345bf..def4e688 100644 --- a/IDEA_TRAILS.md +++ b/IDEA_TRAILS.md @@ -237,7 +237,7 @@ This trail also asks: **What behaviors does this system select for?** Capability | **Museum of Ordinary Life** | Preserve ordinary pre-transition AI interfaces, prompts, automated decisions, help systems, errors, consent notices, and the human labor hidden behind “automation.” | - **Canonical broad systems concept:** [`concepts/intelligence-ecology.md`](concepts/intelligence-ecology.md). -- **Related cross-disciplinary research (review-pending):** [Intelligence Without Omniscience](research/intelligence-without-omniscience/README.md) distinguishes generality, transactive memory, cognitive offloading, individual capacities, infrastructure-enabled capability, and legitimate authority without asserting a universal intelligence theory. +- **Related cross-disciplinary research (review-pending):** [Intelligence Without Omniscience](research/intelligence-without-omniscience/README.md) distinguishes generality, transactive memory, cognitive offloading, infrastructure-enabled capability, and authority. [Epistemic Trust and Collective Verification](research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) follows how expert testimony, source independence, criticism, and AI assistance can support or undermine warranted reliance without asserting a universal intelligence theory. - **Shared working design principle:** [`concepts/agency-and-automation.md`](concepts/agency-and-automation.md) asks when delegation expands agency, how burdens move, and what governance must accompany automation. - **Related research question:** [`research/conscious-systems/`](research/conscious-systems/README.md) asks which conditions might support experience; it does not infer consciousness or authority from capability. - **Canonical focused research:** Liberated Intelligence. From 2ae1b75b5d7f1703e924db76f90254138cb77091 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:42:22 -0700 Subject: [PATCH 34/71] docs(trails): index epistemic trust and collective verification research --- IDEA_TRAIL_GRAPH.yml | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) diff --git a/IDEA_TRAIL_GRAPH.yml b/IDEA_TRAIL_GRAPH.yml index faea84f1..622caecc 100644 --- a/IDEA_TRAIL_GRAPH.yml +++ b/IDEA_TRAIL_GRAPH.yml @@ -1,5 +1,5 @@ version: 0.4 -updated: 2026-09-24 +updated: 2026-10-07 canonical_trails: IDEA_TRAILS.md metadata_convention: IDEA_TRAIL_METADATA.md @@ -862,3 +862,12 @@ documents: - collective-judgment-dissent - accessibility-participation - commons-shared-capacity + + - repo: Root-Sequence/root-sequence + path: research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md + role: research + trails: + - intelligence-authority + - collective-judgment-dissent + - memory-provenance + - commons-shared-capacity From e0a9072baec0cba64c521688e5ebd8034590e997 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:42:24 -0700 Subject: [PATCH 35/71] docs(trails): regenerate browser with collective verification study --- IDEA_TRAIL_INDEX.md | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/IDEA_TRAIL_INDEX.md b/IDEA_TRAIL_INDEX.md index 5e0a07dd..3f850dc0 100644 --- a/IDEA_TRAIL_INDEX.md +++ b/IDEA_TRAIL_INDEX.md @@ -109,7 +109,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `commons-shared-capacity` -- **Research:** [`root-sequence/analysis/externalized-responsibility.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/externalized-responsibility.md) · [`root-sequence/analysis/spatial-liberation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/spatial-liberation.md) · [`root-sequence/futures/abandoned-futures.md`](https://github.com/Root-Sequence/root-sequence/blob/main/futures/abandoned-futures.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) +- **Research:** [`root-sequence/analysis/externalized-responsibility.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/externalized-responsibility.md) · [`root-sequence/analysis/spatial-liberation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/spatial-liberation.md) · [`root-sequence/futures/abandoned-futures.md`](https://github.com/Root-Sequence/root-sequence/blob/main/futures/abandoned-futures.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) - **Embodied practice:** [`liberation-mass/README.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/README.md) · [`liberation-mass/docs/spaces/spaces.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/spaces/spaces.md) · [`liberation-mass/docs/spatial-principles.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/spatial-principles.md) - **Real-system work:** [`community-infrastructure/docs/concepts/resource-commons.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/resource-commons.md) · [`community-infrastructure/docs/concepts/space-mobility-and-access.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/space-mobility-and-access.md) - **Analytical lens:** [`root-sequence/concepts/agency-and-automation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/agency-and-automation.md) @@ -153,7 +153,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `intelligence-authority` -- **Research:** [`community-infrastructure/docs/research/local-community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/local-community-assistant.md) · [`liberated-intelligence/README.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/README.md) · [`liberated-intelligence/notes/roots-of-conscious-consent.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/notes/roots-of-conscious-consent.md) · [`root-sequence/concepts/intelligence-ecology.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/intelligence-ecology.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/conscious-systems/what-can-we-change-about-a-mind.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/what-can-we-change-about-a-mind.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) +- **Research:** [`community-infrastructure/docs/research/local-community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/local-community-assistant.md) · [`liberated-intelligence/README.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/README.md) · [`liberated-intelligence/notes/roots-of-conscious-consent.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/notes/roots-of-conscious-consent.md) · [`root-sequence/concepts/intelligence-ecology.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/intelligence-ecology.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/conscious-systems/what-can-we-change-about-a-mind.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/what-can-we-change-about-a-mind.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) - **Human practice:** [`beinghumane-atlas/seeds/intelligence-ecology-and-selection-environments.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/intelligence-ecology-and-selection-environments.md) · [`beinghumane-guide/README.md`](https://github.com/Root-Sequence/beinghumane-guide/blob/main/README.md) - **Real-system work:** [`coherent-computing/docs/conceptual-model.md`](https://github.com/Root-Sequence/coherent-computing/blob/main/docs/conceptual-model.md) · [`community-infrastructure/docs/concepts/community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-assistant.md) · [`community-infrastructure/docs/concepts/permissions-and-consent.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/permissions-and-consent.md) · [`community-infrastructure/prototypes/community-assistant/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/community-assistant/README.md) - **Analytical lens:** [`root-sequence/concepts/agency-and-automation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/agency-and-automation.md) · [`root-sequence/concepts/legible-ai-assisted-expression.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/legible-ai-assisted-expression.md) @@ -184,7 +184,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `memory-provenance` -- **Research:** [`coherent-world/no-one-noticed/research/HISTORICAL-CONTINGENCY-AND-ABANDONED-FUTURES.md`](https://github.com/Root-Sequence/coherent-world/blob/main/no-one-noticed/research/HISTORICAL-CONTINGENCY-AND-ABANDONED-FUTURES.md) · [`root-sequence/futures/abandoned-futures.md`](https://github.com/Root-Sequence/root-sequence/blob/main/futures/abandoned-futures.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/systems/emergence/applications/personal.md`](https://github.com/Root-Sequence/root-sequence/blob/main/systems/emergence/applications/personal.md) +- **Research:** [`coherent-world/no-one-noticed/research/HISTORICAL-CONTINGENCY-AND-ABANDONED-FUTURES.md`](https://github.com/Root-Sequence/coherent-world/blob/main/no-one-noticed/research/HISTORICAL-CONTINGENCY-AND-ABANDONED-FUTURES.md) · [`root-sequence/futures/abandoned-futures.md`](https://github.com/Root-Sequence/root-sequence/blob/main/futures/abandoned-futures.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [`root-sequence/systems/emergence/applications/personal.md`](https://github.com/Root-Sequence/root-sequence/blob/main/systems/emergence/applications/personal.md) - **Human practice:** [`beinghumane-atlas/seeds/historical-contingency-and-making-possibility-real.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/historical-contingency-and-making-possibility-real.md) · [`beinghumane-guide/README.md`](https://github.com/Root-Sequence/beinghumane-guide/blob/main/README.md) - **Real-system work:** [`community-infrastructure/docs/concepts/community-knowledge.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-knowledge.md) - **Analytical lens:** [`root-sequence/concepts/dynamic-coherence.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/dynamic-coherence.md) · [`root-sequence/concepts/legible-ai-assisted-expression.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/legible-ai-assisted-expression.md) @@ -242,7 +242,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `collective-judgment-dissent` -- **Research:** [`root-sequence/analysis/collective-judgment-and-manufactured-consensus.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/collective-judgment-and-manufactured-consensus.md) ★ · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) +- **Research:** [`root-sequence/analysis/collective-judgment-and-manufactured-consensus.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/collective-judgment-and-manufactured-consensus.md) ★ · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) - **Human practice:** [`beinghumane-atlas/seeds/humane-collective-decision-architecture.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/humane-collective-decision-architecture.md) - **Real-system work:** [`community-infrastructure/docs/concepts/collective-decision-making.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/collective-decision-making.md) - **Speculative design:** [`coherent-world/README.md`](https://github.com/Root-Sequence/coherent-world/blob/main/README.md) · [`coherent-world/simulation/coherence-model.md`](https://github.com/Root-Sequence/coherent-world/blob/main/simulation/coherence-model.md) · [`coherent-world/world/collective-decision-and-dissent.md`](https://github.com/Root-Sequence/coherent-world/blob/main/world/collective-decision-and-dissent.md) From ce679ed8bf903cba40fed406ec3cfdc03bacb1f0 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 20:43:16 -0700 Subject: [PATCH 36/71] docs(research): preserve next research frontier and anti-proliferation boundaries --- .../README.md | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+) diff --git a/research/intelligence-without-omniscience/README.md b/research/intelligence-without-omniscience/README.md index bcc4a783..6f819b4d 100644 --- a/research/intelligence-without-omniscience/README.md +++ b/research/intelligence-without-omniscience/README.md @@ -389,6 +389,25 @@ These are preservation cues for future research and prose review; they are **not 6. Examine actual open-science and open-AI governance mechanisms, costs, and risks rather than assuming the word "commons" solves them. 7. Invite reviewers from cognitive science, information science, disability/capability studies, human–AI interaction, organizational research, and directly affected communities. Different disciplines have authority over different claims, not automatic authority over everyone's decisions. +## 22. Current frontier: warranted reliance across imperfect knowers + +This inquiry has reached a natural **pause-and-test frontier**, not the end of the research. Our next step is **not** a new master theory or automatic propagation to every Root Sequence project. It is a set of discriminating questions: + +- When an individual cannot personally verify a specialized claim, what combination of evidence, expert trust, independent review, and correction makes reliance reasonable for that particular task? +- How can a source-routing system distinguish genuine independent corroboration from multiple derivatives of one unverified report? +- Does structured multi-agent discussion improve *unique information elicitation* under controlled budgets, or merely add words, time, and compute? +- When is an AI explanation actually **verifiable**, as opposed to fluent, plausible, and reassuring? +- Can an accessible collective knowledge system preserve uncertainty, dissent, source provenance, consent, and the right to challenge its own questions? +- Who has the power to redesign the epistemic infrastructure when its categories, incentives, ranking, or authority allocation stop serving its participants? +- When do verification processes themselves impose disproportionate labor, exclusion, surveillance, or latency? +- What would a competent native-field baseline discover without this interdisciplinary synthesis, and what non-obvious correction would the synthesis add? + +**Provisional priority:** compare one bounded hidden-profile/source-dependence case against an established information-sharing protocol and competent ordinary review. Predeclare assumptions and failure criteria before examining outcomes; separate model behavior from real-world empirical evidence. The [2026 HiddenBench study](https://proceedings.mlr.press/v306/li26ej.html) offers a relevant prior benchmark, not a source of universal constants. + +**Current homes:** the core question, research sources, and original conceptual path remain under this research directory; complementary mechanisms stay in their existing conceptual homes. The private [Coherent World paper experiment proposal](https://github.com/Root-Sequence/coherent-world/pull/22) can stress-test bounded fictional translations **only after** the established Neighborhood v0.2 briefing frontier. Appropriate local applications to Community Infrastructure, Coherent Computing, Being Human(e), Liberated Intelligence, and other projects are prospective, not automatically authorized or already implemented. + +**Preservation discipline:** keep the generative question, counterarguments, and revision history available; do not copy the whole body of work into every repo, or treat document count as intellectual progress. A useful partial answer may remain open without becoming another compulsory maintenance surface. No new project, product, governance policy, world canon, or declared scientific breakthrough is implied by this research map. + ## Selected primary and reference sources - Legg & Hutter (2007), [Universal Intelligence](https://arxiv.org/abs/0712.3329). Formal proposal, not a historical stage of ASI. From adfd0823a7c5c4b4a9491744ab94a4443b62c3ff Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:10:44 -0700 Subject: [PATCH 37/71] docs(research): investigate distinct collective-knowledge mechanisms and falsifiable toy probe --- .../mechanisms-of-collective-knowledge.md | 208 ++++++++++++++++++ 1 file changed, 208 insertions(+) create mode 100644 research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md diff --git a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md new file mode 100644 index 00000000..f29e9035 --- /dev/null +++ b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md @@ -0,0 +1,208 @@ +# Mechanisms of collective knowledge: from information to effective action +## Source-aware causal questions, competing explanations, and a first bounded probe + +**Document role:** Focused mechanism investigation linked to [Intelligence Without Omniscience](README.md). +**Status:** WORKING / AI-assisted / review pending — 2026-10-07. +**Epistemic boundary:** Do not confuse original source results, illustrative mathematics, proposed mechanisms, or causal claims about actual groups. +**Method:** [Root Sequence Research Method](../method.md) — preserve → decompose → route → compare competent baselines → test → revise. +**Idea Trails:** Intelligence & Authority; Collective Judgment & Dissent; Commons & Shared Capacity; Accessibility & Participation. +**Trail role:** research + + + +> A group can hold information sufficient for a better judgment yet fail to use it. That sentence names an outcome, not a single mechanism. + +## 1. The causal distinction + +A *heuristic, non-universal* chain of constraints: + +\`\`\`text +world / problem / affected actors + ↓ +observation, information creation, documentation + ↓ +findability and knowledge / expertise routing + ↓ +willingness and permission to disclose + ↓ +communication, translation, representation + ↓ +source/evidence assessment and error checking + ↓ +integration into the working model + ↓ +critical uptake / reconsideration of problem or options + ↓ +legitimate authorization, resources, and implementation + ↓ +consequences, maintenance, feedback, change in future conditions + ↺ +\`\`\` + +The real processes are non-linear. Actors can act without all knowledge; a sound objection may be rejected for legitimate reasons; private or community-governed knowledge may never be disclosed. The stages are diagnostic *distinctions* that can be reordered or bypassed, not a compulsory participation pipeline. + +This is especially important where the same observed outcome—an institution ignores a useful fact—could arise from incompatible causes: +- no one observed it; +- it was not discoverable; +- the knowledgeable actor did not share it; +- the speaker was prevented from participating; +- translation distorted its meaning; +- prejudice reduced warranted credibility; +- the claim was evaluated and reasonably rejected; +- it was accepted but the institution lacked authority or resources to act; +- interests or incentives favored a different outcome. + +The remedy depends on which mechanism is present, who controls it, and whether proposed remediation creates new burdens or coercion. + +## 2. Mechanisms worth keeping separate + +| ID | Mechanism | Domain baseline | Observable signature | Candidate intervention | Failure case | +| --- | --- | --- | --- | --- | --- | +| M01 | Shared information repeatedly dominates discussion | Hidden-profile literature | High shared-fact talk; low unique-fact **coverage** | Ask separately for potentially unique evidence | Less coverage does not by itself establish harm; the unique evidence may be irrelevant | +| M02 | Unknown or incorrect map of expertise | Transactive memory / organizational knowledge | Questions fail to reach holders, or reach misleading ones | Scoped expertise/source directory; check freshness | Stale directory can be worse than informal routing | +| M03 | Strategic withholding or unsafe disclosure | Social psychology / power and consent | Selective silence under competition, fear, or privacy concerns | Incentive changes, safe alternatives, paid consultation | Forced disclosure and surveillance create harm | +| M04 | Evidence credibility misallocated | Social epistemology, expert-assessment methods | Sound evidence discounted or unreliable claims elevated | Source-context evaluation and independent challenge | Treating testimony equally regardless of relevance is not justice | +| M05 | Message/representation loss | HCI, classification, boundary-object studies | Summary loses conditions, restrictions, dissent | Source-linked, contestable representations | Additional translation work overwhelms participants | +| M06 | Correct knowledge has no *uptake* | Longino, organizational learning | Objection logged but institutional model cannot revise | Review conditions that actually permit revision | Feedback theater or endless procedure | +| M07 | Correct recommendation cannot become legitimate action | Capability approach, governance, material analysis | Correct answer but no resources, access, consent, or authorization | Address specific conversion and decision rights | Accuracy is used to justify unaccountable authority | +| M08 | Outcomes reshape the conditions of future knowledge | Path dependence / institutional and cultural evolution | Access, trust, maintenance, incentives, or capacity change after action | Compare future action space, maintenance, governance | Temporary improvement creates brittle dependency | +| M09 | False independent corroboration | Source criticism, reproducibility | Multiple apparent sources inherit one weak original | Claim-level lineage and truly independent checking | "Many citations" laundering one error | +| M10 | Explanatory fluency mistaken for verified understanding | Illusion of explanatory depth, human-AI reliance | Confidence grows without independent verification ability | Calibrated source-linked explanation or review | Compulsory testing excludes legitimate assisted users | + +These IDs are a **local investigation aid**, not new canonical Root Sequence vocabulary or a scientifically validated ontology. + +## 3. What external evidence currently warrants + +### Shared versus unshared information (M01) + +The [Lu, Yuan & McLeod 2012 meta-analysis](https://pubmed.ncbi.nlm.nih.gov/21896790/) covers 65 studies and 3,189 groups. Groups discussed considerably more commonly held than unique information; **unique-information coverage**, rather than merely the proportion of discussion about unique information, was more strongly related to decision quality in those studies. It also found that communication medium did not, by itself, significantly affect unique information pooling or decision quality. + +**Interpretation:** facilitating more talking or shifting media is not an adequate mechanism claim. A probe should measure which decisive information became available and whether it actually changed the decision process. + +### Incentives and strategic disclosure (M03) + +[Toma & Butera 2009](https://pubmed.ncbi.nlm.nih.gov/19332434/) report that competition, compared with cooperation in their hidden-profile experiments, fostered withholding of unshared information and less use of preference-disconfirming evidence. Their analysis identified reluctance to use disconfirming information as part of the difference in decision quality. This supports testing **sharing** and **belief revision** separately, without extrapolating an exact probability to a real setting. + +### A counterexample to procedural optimism (M06) + +A [hidden-profile experiment with organizational board members, published 2021](https://www.sciencedirect.com/science/article/pii/S0025174721000434) found that the particular structured discussion procedures improved perceived consideration of pros and cons but did not improve objective decision quality. The result is task- and procedure-specific. + +**Inference:** apparently thoughtful deliberation is not enough. We must record accuracy, information coverage, source integrity, material consequences, and participant burden—not merely meeting satisfaction or confidence. + +### Recent AI agents (M01, M02, M04) + +[Li, Naito & Shirado, ICML 2026, *Systematic Failures in Collective Reasoning under Distributed Information in Multi-Agent LLMs*](https://proceedings.mlr.press/v306/li26ej.html) introduces HiddenBench (65 tasks, 15 models). The reported 30.1% multi-agent distributed-information versus 80.7% single-agent full-information scores compare **different access to information**, not inherent group-vs-individual cognition under an equal-information design. The authors find premature convergence around shared evidence and improvements from a lightweight structured communication protocol. + +This is an especially relevant empirical benchmark but does not authorize copying its scores into a fictional society model. Its mechanism remains bounded to the evaluated configurations, prompts, tasks, and versions. + +## 4. First mechanistic probe: hidden information under a fixed message budget + +**Location:** the private [Coherent World mechanism probe 001](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-001.md). This is a *paper calculation*, not a new world simulator, study of real participants, or independent validation. + +### Predeclared toy assumptions + +- A bounded problem has **two separate critical facts** known to distinct participants. Getting both into the shared record is *necessary* (not sufficient) for a defensible decision. +- **Ordinary discussion:** each of four independent information turns is spent repeating shared facts with probability \`s\`. Otherwise the turn attempts one of the two critical facts with equal chance; it is disclosed with probability \`d\`. +- **Targeted protocol:** within the same overall four-turn allowance, two turns explicitly ask the presumed holders of the critical facts. Each route is correct with probability \`r\`; if routed correctly, the fact is disclosed with probability \`d\`. The other two turns remain for checking or assessment, but their usefulness is **not modeled**. +- These probabilities are **invented sensitivity parameters**, not estimates inferred from the psychological or AI studies. The model ignores message content, source validity, learning, strategic adaptation, unequal costs, and legitimate authorization. +- The outcome is narrowly defined as **both facts surfaced**. It is not group intelligence, correct inference, wellbeing, or legitimate action. + +Let \`q = (1-s)d/2\` be the probability that any ordinary turn yields a particular critical fact. Assuming independent turns, inclusion–exclusion gives: + +\`\`\`text +P(both critical facts surfaced in n ordinary turns) + = 1 - 2(1-q)^n + (1-2q)^n + +P(both surfaced after two accurately routed, independent requests) + = (r*d)^2 +\`\`\` + +For the fixed baseline \`n=4\`, the illustrative sensitivity cases are: + +| Condition | Shared-fact repetition \`s\` | Disclosure \`d\` | Routing \`r\` | Ordinary | Targeted | +| --- | ---: | ---: | ---: | ---: | ---: | +| A: high repetition, accurate routing | 0.80 | 0.90 | 1.00 | **8.06%** | **81.00%** | +| B: less repetition, poor routing | 0.50 | 0.90 | 0.60 | **37.00%** | **29.16%** | +| C: no disclosure | 0.80 | 0.00 | 1.00 | **0.00%** | **0.00%** | +| D: extreme repetition | 0.95 | 0.90 | 1.00 | **0.58%** | **81.00%** | + +Calculations were checked against exhaustive enumeration of the three multinomial categories (common, fact A, fact B) for four ordinary turns, rather than relying only on sampled Monte Carlo trials. Values are rounded for readability. + +### What the probe demonstrates *conditionally* + +1. If unique facts are repeatedly overlooked, asking their actual holders can dramatically improve **fact exposure**. +2. If the expertise map is sufficiently unreliable, targeted routing may perform worse than ordinary exploration (case B). Structured processes are not guaranteed improvements. +3. If no disclosure is possible, neither protocol surfaces the facts (case C). Silence might be due to privacy, coercion, lack of safe channels, or simply missing knowledge; the model does **not** diagnose which. +4. Even perfect exposure does not establish reliability, integration into the working model, authorization, or downstream benefit. + +### What it cannot establish + +- No study of humans or LLMs was executed. +- It does **not** prove that real groups repeat shared information with \`s=0.8\`, disclose with \`d=0.9\`, or route with \`r=0.6\`. +- A two-fact toy is not general intelligence, political deliberation, or a whole social system. +- The protocols do not have proven equal cognitive/verification costs, even if their *message allowances* match. +- Real sources can be correlated or misleading, and facts can be disputed, private, or irrelevant. +- When task success requires factors beyond both facts, probability of correct action is *at most* the calculated fact-coverage probability; it may be much lower. +- The toy is evidence only of the stated mathematical relationship given its assumptions. + +### Immediate falsification / narrowing direction + +A next protocol should manipulate **one mechanism at a time**: + +- Same actor partitions, task, time, and information → change *only* method of eliciting unique facts. +- Same elicitation method → change expertise map accuracy. +- Same routing and facts → vary privacy, compensation, safety, and willingness to disclose. +- Same acquired evidence → change which conclusions are warranted; include **false, low-quality, or non-independent** evidence. +- Same correct conclusion → vary institution's authority/resources, to separate knowing from acting. +- Introduce a new task family to test transfer, rather than simply redescribing the same facts. + +Compare against existing best-practice review; report null/negative effects and all extra work imposed on participants. + +## 5. Evidence integrity: what if six citations come from one source? + +An institution sees three apparently independent reports supporting a proposal. All three copied one original inference. A fourth source raises an independent objection. + +The count is 3 versus 1, but the number of **independent evidential lineages** could be 1 versus 1. That is not a tie on truth; each original must be assessed on its actual evidence and method. + +Possible intervention: attach *claim and evidence lineage* to summaries and recommendations, with careful access restrictions. Counting lineages is neither a substitute for validating source quality nor a reason to reveal sensitive source identity. + +This mechanism is distinct from hidden-profile fact elicitation: unique evidence can be surfaced and still misweighted by falsely duplicated citations. + +See [Epistemic Trust and Collective Verification](epistemic-trust-and-collective-verification.md). + +## 6. Why the Root Sequence synthesis might be redundant + +Competent information-sharing protocols, methods for vetting expertise, accessibility audits, source criticism, participatory research, and institutional decision procedures already exist. + +The research contribution here is currently **cross-field routing and a clear diagnostic separation**, not a new general intelligence theory. It adds value only if a case analysis discovers a consequential gap that native methods *in combination* did not already reveal, or provides a demonstrably useful route to the correct native method. + +Test against: +- ordinary competent source/decision audit; +- the appropriate domain's access and participation audit; +- a baseline hidden-profile protocol; +- an independent expert review with dissent recorded. + +A **null** result or an inefficient new protocol would narrow or retire the proposal. + +## 7. Relationship to intelligence infrastructure and agency + +Better cognitive infrastructure may help a person find knowledge, locate a specialist, critically understand it, or apply it under real constraints. It can also extract consent-sensitive knowledge, hide source manipulation, centralize authority, and impose unpaid explanation burdens. + +Relevant conceptual homes are preserved: +- [Epistemic Discoverability](../../concepts/epistemic-discoverability.md) — term and field routing. +- [Intelligence Ecology](../../concepts/intelligence-ecology.md) — what system design and environments reward or select. +- [From Knowledge to Uptake](from-knowledge-to-uptake.md) — critical responsiveness, boundary objects, framing, and epistemic labor. +- [Epistemic Injustice and Knowledge Recognition](epistemic-injustice-and-knowledge-recognition.md) — testimonial/hermeneutical injustice, silencing, and interpretation. +- [Intelligence Commons](intelligence-commons.md) — access, governance, consent, and distribution of useful capabilities. +- [Epistemic Trust and Collective Verification](epistemic-trust-and-collective-verification.md) — evidence independence, warranted reliance, expert assessment. +- [Epistemic Contrast](../methods/epistemic-contrast.md), [Deliberative Inquiry](../methods/deliberative-inquiry.md), and the [Research Method](../method.md) remain the **methods**; do not replace them with a new compulsory checklist. + +## 8. What to investigate next + +**Next mechanism (M04/M09):** source-lineage confusion and false corroboration. It offers a strong contrast between *more references* and *more independent evidence*, can be tested without a general AI model, and bears directly on our own AI-assisted research workflow. + +**Following (M06/M07):** accurate objection with no institutional uptake or capability to act. That is where knowledge, epistemic justice, governance, and resource conversion must be analytically disentangled. + +**Later:** source disclosure under competing incentives (M03), and whether changes in these mechanisms produce transfer across different task domains rather than one stylized scenario. + +Do not operationalize private resident histories or override the current Coherent World Neighborhood v0.2 briefing. All new fictional values must be separately approved before incorporation into the game. From 6b92003c273986065add406978f4dec8df76c8bf Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:12:42 -0700 Subject: [PATCH 38/71] docs: connect mechanism probes to research and scenario navigation --- research/README.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/research/README.md b/research/README.md index b44f436d..17ca76f5 100644 --- a/research/README.md +++ b/research/README.md @@ -21,6 +21,8 @@ This directory holds bounded research inquiries and versioned papers that synthe - [Intelligence Without Omniscience](intelligence-without-omniscience/README.md) — **AI-assisted, review-pending cross-disciplinary inquiry** into generality, situated capabilities, distributed knowledge, accessibility, agency, and the conditions for trustworthy cooperation. It is a research map, not a validated theory. - Companion inquiries: [Knowledge recognition and epistemic injustice](intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [From knowledge to uptake](intelligence-without-omniscience/from-knowledge-to-uptake.md) · [Epistemic trust and collective verification](intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [Intelligence commons](intelligence-without-omniscience/intelligence-commons.md). +- [Mechanisms of collective knowledge](intelligence-without-omniscience/mechanisms-of-collective-knowledge.md) — **first bounded mechanism investigation** of shared-information sampling, expertise routing, disclosure, evidence assessment, institutional uptake, and limits on action, with a falsifiable paper calculation and links to native disciplines. It does not measure AGI or validate collective intelligence. + - [Conscious Systems](conscious-systems/README.md) — **exploratory, AI-assisted seed; author review pending**. Starts with [what can change about a mind while preserving consciousness](conscious-systems/what-can-we-change-about-a-mind.md), separating experience from report, capability, and authority. ## Cross-cutting related work From 33bc6df57b32ff5fb8dd898724336171c11ac40e Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:12:45 -0700 Subject: [PATCH 39/71] docs: connect mechanism probes to research and scenario navigation --- research/intelligence-without-omniscience/README.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/research/intelligence-without-omniscience/README.md b/research/intelligence-without-omniscience/README.md index 6f819b4d..c31835d3 100644 --- a/research/intelligence-without-omniscience/README.md +++ b/research/intelligence-without-omniscience/README.md @@ -224,6 +224,8 @@ Practical test: when assistance improves an immediate task, does the person also A capable AI may improve the conditions of human judgment, but capability does not itself confer legitimate authority. +For a source-aware decomposition into **which mechanisms fail** (not merely whether a group fails), and the first computed analytic counterexample where an apparently better protocol does worse, see [Mechanisms of collective knowledge](mechanisms-of-collective-knowledge.md). The Coherent World extension consists of bounded **paper probes**, not validated experiments or implemented scenario rules. + ## 14. Collective intelligence: where claims can fail A collection of specialists may achieve broader task coverage through coordination without becoming a single AGI. The relevant question is about **system-level functional capability**, not whether group members merged into one mind. From fb691f29a178502271f23098c4e3d114989d535b Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:12:48 -0700 Subject: [PATCH 40/71] docs: connect mechanism probes to research and scenario navigation --- .../epistemic-trust-and-collective-verification.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md b/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md index dfd6d340..137d85e4 100644 --- a/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md +++ b/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md @@ -200,6 +200,8 @@ Three reasons a claim might reasonably remain unresolved: The system should permit "unknown," "contested," and "not mine to decide," instead of inventing certainty. +The companion [Mechanisms of collective knowledge](mechanisms-of-collective-knowledge.md) distinguishes source-independent verification from fact elicitation, disclosure and institutional uptake, and links to bounded Coherent World paper probes. A probability calculated for fact coverage is not a probability of correct trust or justified action. + ## 13. Candidate Coherent World comparisons **A. Incomplete collective knowledge:** compare unstructured exchange, source-aware information elicitation, and a structured evidence map. Hold informational and resource differences explicit; borrow from HiddenBench only the bounded design principle, not the claimed 2026 performance numbers. From 1c37b3f5ef1431ac329c6767670ec3406490b23c Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:12:52 -0700 Subject: [PATCH 41/71] docs: connect mechanism probes to research and scenario navigation --- .../from-knowledge-to-uptake.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/research/intelligence-without-omniscience/from-knowledge-to-uptake.md b/research/intelligence-without-omniscience/from-knowledge-to-uptake.md index d83b0a0b..549c151e 100644 --- a/research/intelligence-without-omniscience/from-knowledge-to-uptake.md +++ b/research/intelligence-without-omniscience/from-knowledge-to-uptake.md @@ -210,6 +210,8 @@ These prompts should be *optional and proportionate*, not an examination imposed **Proposed test:** Compare source retrieval, AI summary, and source-backed explanation for a bounded task. Separately measure task performance, confidence calibration, ability to identify unsupported claims, willingness to seek expertise, retained explanation when relevant, and the ability to choose not to explain. Do not mistake a better score on a memory recall test for a complete evaluation of cognitive augmentation. +[Mechanisms of collective knowledge](mechanisms-of-collective-knowledge.md) supplies a separate source-aware causal investigation. Its analytic fact-coverage probe explicitly stops at whether relevant facts enter the shared record; it does **not** represent uptake, justification, power, or consequences. This document retains those distinct questions. + ## 11. Paired Coherent World tests with invalidation conditions **Test A — information routing:** identical actors, task, information, and group size; vary whether the group has an accurate, stale, or unavailable expertise map. Measure whether critical knowledge is actually obtained, and at what cost. From d634e54fdd8f5f41ca24e547cc97c0a1b99a7772 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:13:07 -0700 Subject: [PATCH 42/71] docs(trails): connect mechanism investigation without adding a new trail --- IDEA_TRAILS.md | 1 + 1 file changed, 1 insertion(+) diff --git a/IDEA_TRAILS.md b/IDEA_TRAILS.md index def4e688..7fa32d67 100644 --- a/IDEA_TRAILS.md +++ b/IDEA_TRAILS.md @@ -238,6 +238,7 @@ This trail also asks: **What behaviors does this system select for?** Capability - **Canonical broad systems concept:** [`concepts/intelligence-ecology.md`](concepts/intelligence-ecology.md). - **Related cross-disciplinary research (review-pending):** [Intelligence Without Omniscience](research/intelligence-without-omniscience/README.md) distinguishes generality, transactive memory, cognitive offloading, infrastructure-enabled capability, and authority. [Epistemic Trust and Collective Verification](research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) follows how expert testimony, source independence, criticism, and AI assistance can support or undermine warranted reliance without asserting a universal intelligence theory. +- **Mechanism investigation (not a new theory):** [Mechanisms of Collective Knowledge](research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md) decomposes discovery, disclosure, source verification, integration, institutional uptake, and authorization, and routes two tiny mathematical/provenance probes into Coherent World without making simulation outcomes real-world evidence. - **Shared working design principle:** [`concepts/agency-and-automation.md`](concepts/agency-and-automation.md) asks when delegation expands agency, how burdens move, and what governance must accompany automation. - **Related research question:** [`research/conscious-systems/`](research/conscious-systems/README.md) asks which conditions might support experience; it does not infer consciousness or authority from capability. - **Canonical focused research:** Liberated Intelligence. From 2cc4a0872c6209a23e9b55bfacdf07b7165a9c07 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:13:22 -0700 Subject: [PATCH 43/71] docs(trails): route collective knowledge mechanism investigation --- IDEA_TRAIL_GRAPH.yml | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/IDEA_TRAIL_GRAPH.yml b/IDEA_TRAIL_GRAPH.yml index 622caecc..4b9b43b7 100644 --- a/IDEA_TRAIL_GRAPH.yml +++ b/IDEA_TRAIL_GRAPH.yml @@ -871,3 +871,12 @@ documents: - collective-judgment-dissent - memory-provenance - commons-shared-capacity + + - repo: Root-Sequence/root-sequence + path: research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md + role: research + trails: + - intelligence-authority + - collective-judgment-dissent + - commons-shared-capacity + - accessibility-participation From 09f2cdf6525d1e4fed5a3c1588f543abb6fc3a4d Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:13:25 -0700 Subject: [PATCH 44/71] docs(trails): regenerate mechanism research browser --- IDEA_TRAIL_INDEX.md | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/IDEA_TRAIL_INDEX.md b/IDEA_TRAIL_INDEX.md index 3f850dc0..85420afd 100644 --- a/IDEA_TRAIL_INDEX.md +++ b/IDEA_TRAIL_INDEX.md @@ -94,7 +94,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `accessibility-participation` -- **Research:** [`community-infrastructure/docs/research/space-access-interoperability.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/space-access-interoperability.md) · [`root-sequence/analysis/spatial-liberation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/spatial-liberation.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) +- **Research:** [`community-infrastructure/docs/research/space-access-interoperability.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/space-access-interoperability.md) · [`root-sequence/analysis/spatial-liberation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/spatial-liberation.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) - **Human practice:** [`beinghumane-guide/README.md`](https://github.com/Root-Sequence/beinghumane-guide/blob/main/README.md) - **Embodied practice:** [`liberation-mass/README.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/README.md) · [`liberation-mass/docs/resilience-and-low-tech.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/resilience-and-low-tech.md) · [`liberation-mass/docs/safety/accessibility.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/safety/accessibility.md) · [`liberation-mass/docs/spaces/spaces.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/spaces/spaces.md) · [`liberation-mass/docs/spatial-principles.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/spatial-principles.md) - **Real-system work:** [`community-infrastructure/docs/concepts/community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-assistant.md) · [`community-infrastructure/docs/concepts/community-discovery.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-discovery.md) · [`community-infrastructure/docs/concepts/community-home.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-home.md) · [`community-infrastructure/docs/concepts/contextual-adaptability.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/contextual-adaptability.md) · [`community-infrastructure/docs/concepts/discovery-profile.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/discovery-profile.md) · [`community-infrastructure/docs/concepts/space-mobility-and-access.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/space-mobility-and-access.md) · [`community-infrastructure/docs/design/accessibility.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/design/accessibility.md) ★ · [`community-infrastructure/docs/design/resilience-and-graceful-degradation.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/design/resilience-and-graceful-degradation.md) · [`community-infrastructure/prototypes/community-assistant/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/community-assistant/README.md) · [`community-infrastructure/prototypes/community-discovery/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/community-discovery/README.md) · [`community-infrastructure/prototypes/degraded-mode/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/degraded-mode/README.md) · [`community-infrastructure/prototypes/space-access/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/space-access/README.md) @@ -109,7 +109,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `commons-shared-capacity` -- **Research:** [`root-sequence/analysis/externalized-responsibility.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/externalized-responsibility.md) · [`root-sequence/analysis/spatial-liberation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/spatial-liberation.md) · [`root-sequence/futures/abandoned-futures.md`](https://github.com/Root-Sequence/root-sequence/blob/main/futures/abandoned-futures.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) +- **Research:** [`root-sequence/analysis/externalized-responsibility.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/externalized-responsibility.md) · [`root-sequence/analysis/spatial-liberation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/spatial-liberation.md) · [`root-sequence/futures/abandoned-futures.md`](https://github.com/Root-Sequence/root-sequence/blob/main/futures/abandoned-futures.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md) - **Embodied practice:** [`liberation-mass/README.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/README.md) · [`liberation-mass/docs/spaces/spaces.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/spaces/spaces.md) · [`liberation-mass/docs/spatial-principles.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/spatial-principles.md) - **Real-system work:** [`community-infrastructure/docs/concepts/resource-commons.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/resource-commons.md) · [`community-infrastructure/docs/concepts/space-mobility-and-access.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/space-mobility-and-access.md) - **Analytical lens:** [`root-sequence/concepts/agency-and-automation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/agency-and-automation.md) @@ -153,7 +153,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `intelligence-authority` -- **Research:** [`community-infrastructure/docs/research/local-community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/local-community-assistant.md) · [`liberated-intelligence/README.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/README.md) · [`liberated-intelligence/notes/roots-of-conscious-consent.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/notes/roots-of-conscious-consent.md) · [`root-sequence/concepts/intelligence-ecology.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/intelligence-ecology.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/conscious-systems/what-can-we-change-about-a-mind.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/what-can-we-change-about-a-mind.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) +- **Research:** [`community-infrastructure/docs/research/local-community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/local-community-assistant.md) · [`liberated-intelligence/README.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/README.md) · [`liberated-intelligence/notes/roots-of-conscious-consent.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/notes/roots-of-conscious-consent.md) · [`root-sequence/concepts/intelligence-ecology.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/intelligence-ecology.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/conscious-systems/what-can-we-change-about-a-mind.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/what-can-we-change-about-a-mind.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) - **Human practice:** [`beinghumane-atlas/seeds/intelligence-ecology-and-selection-environments.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/intelligence-ecology-and-selection-environments.md) · [`beinghumane-guide/README.md`](https://github.com/Root-Sequence/beinghumane-guide/blob/main/README.md) - **Real-system work:** [`coherent-computing/docs/conceptual-model.md`](https://github.com/Root-Sequence/coherent-computing/blob/main/docs/conceptual-model.md) · [`community-infrastructure/docs/concepts/community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-assistant.md) · [`community-infrastructure/docs/concepts/permissions-and-consent.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/permissions-and-consent.md) · [`community-infrastructure/prototypes/community-assistant/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/community-assistant/README.md) - **Analytical lens:** [`root-sequence/concepts/agency-and-automation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/agency-and-automation.md) · [`root-sequence/concepts/legible-ai-assisted-expression.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/legible-ai-assisted-expression.md) @@ -242,7 +242,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `collective-judgment-dissent` -- **Research:** [`root-sequence/analysis/collective-judgment-and-manufactured-consensus.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/collective-judgment-and-manufactured-consensus.md) ★ · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) +- **Research:** [`root-sequence/analysis/collective-judgment-and-manufactured-consensus.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/collective-judgment-and-manufactured-consensus.md) ★ · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) - **Human practice:** [`beinghumane-atlas/seeds/humane-collective-decision-architecture.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/humane-collective-decision-architecture.md) - **Real-system work:** [`community-infrastructure/docs/concepts/collective-decision-making.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/collective-decision-making.md) - **Speculative design:** [`coherent-world/README.md`](https://github.com/Root-Sequence/coherent-world/blob/main/README.md) · [`coherent-world/simulation/coherence-model.md`](https://github.com/Root-Sequence/coherent-world/blob/main/simulation/coherence-model.md) · [`coherent-world/world/collective-decision-and-dissent.md`](https://github.com/Root-Sequence/coherent-world/blob/main/world/collective-decision-and-dissent.md) From 269d4020cd1bc8f7cd8a5403af53541239bfbeab Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:14:31 -0700 Subject: [PATCH 45/71] docs(research): qualify toy with missingness confound and expertise experiments --- .../mechanisms-of-collective-knowledge.md | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) diff --git a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md index f29e9035..9ebf10ab 100644 --- a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md +++ b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md @@ -95,6 +95,16 @@ A [hidden-profile experiment with organizational board members, published 2021]( This is an especially relevant empirical benchmark but does not authorize copying its scores into a fictional society model. Its mechanism remains bounded to the evaluated configurations, prompts, tasks, and versions. +### Expertise maps, motives, and what happens *after* information is shared + +[Van Ginkel & Van Knippenberg (2009)](https://doi.org/10.1016/j.obhdp.2008.10.003) experimentally examined knowledge of distributed expertise, group task representations, reflection, and information elaboration (N=125 decision-making groups, as reported in the article abstract). Their results distinguish awareness of **who knows what** from how the group frames and uses that knowledge. This constrains a simple "make an expertise directory" proposal: the directory may be useful through changing task interpretation, not just through direct retrieval. + +[Toma, Vasiljevic, Oberlé & Butera (2013)](https://pubmed.ncbi.nlm.nih.gov/22577834/) examined expertise assignment combined with cooperative versus competitive goals in hidden-profile groups. Assigning experts supported information pooling with cooperative goals but **reduced it under competitive goals**. A system must examine whether expertise labels change status competition and willingness to contribute. + +[Xiao, Zhang & Basadur (2016)](https://doi.org/10.1016/j.jbusres.2015.05.014) investigate the gap between information sharing and its actual use in new product development decisions. Their findings caution against making *fact coverage* stand in for adequate information integration, especially with unequally distributed information. + +These empirical studies do not guarantee any intervention will work outside their investigated tasks. They also show why the attractive claim "just ask the right person" is not sufficiently specified. + ## 4. First mechanistic probe: hidden information under a fixed message budget **Location:** the private [Coherent World mechanism probe 001](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-001.md). This is a *paper calculation*, not a new world simulator, study of real participants, or independent validation. @@ -135,6 +145,12 @@ Calculations were checked against exhaustive enumeration of the three multinomia 3. If no disclosure is possible, neither protocol surfaces the facts (case C). Silence might be due to privacy, coercion, lack of safe channels, or simply missing knowledge; the model does **not** diagnose which. 4. Even perfect exposure does not establish reliability, integration into the working model, authorization, or downstream benefit. +### The toy's especially consequential confound + +The targeted-query condition already knows, by design, that **two** critical facts exist and which putative knowledge-holders to ask. The ordinary condition is not given an equivalent task-specific missing-fact map. Thus the comparison can isolate neither generic discussion structure nor independent intelligence: it partly reflects **privileged knowledge of what to seek**. A fair next study must give both conditions the same specialist roster and neutral topic list, while varying only whether a structured solicitation step occurs. How accurately participants know the distribution of unique information should be a **separate randomized factor**. Include a case with **no decisive unique fact**, so solicitation may add overhead without benefit. + +The model also ignores source validity, use of information after transmission, and incentives. It is an *analytic sensitivity illustration*, not a positive experimental result about groups. + ### What it cannot establish - No study of humans or LLMs was executed. From 115f7f43d5c0f48a4e5dc76e6433d8265a1cf9ba Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:15:47 -0700 Subject: [PATCH 46/71] docs(research): clarify source sample and make mechanism studies easier to navigate --- .../mechanisms-of-collective-knowledge.md | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md index 9ebf10ab..e0eb5576 100644 --- a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md +++ b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md @@ -97,7 +97,7 @@ This is an especially relevant empirical benchmark but does not authorize copyin ### Expertise maps, motives, and what happens *after* information is shared -[Van Ginkel & Van Knippenberg (2009)](https://doi.org/10.1016/j.obhdp.2008.10.003) experimentally examined knowledge of distributed expertise, group task representations, reflection, and information elaboration (N=125 decision-making groups, as reported in the article abstract). Their results distinguish awareness of **who knows what** from how the group frames and uses that knowledge. This constrains a simple "make an expertise directory" proposal: the directory may be useful through changing task interpretation, not just through direct retrieval. +[Van Ginkel & Van Knippenberg (2009)](https://doi.org/10.1016/j.obhdp.2008.10.003) experimentally examined knowledge of distributed expertise, group task representations, reflection, and information elaboration (reported N=125; the abstract does not clarify the unit of the sample size). Their results distinguish awareness of **who knows what** from how the group frames and uses that knowledge. This constrains a simple "make an expertise directory" proposal: the directory may be useful through changing task interpretation, not just through direct retrieval. [Toma, Vasiljevic, Oberlé & Butera (2013)](https://pubmed.ncbi.nlm.nih.gov/22577834/) examined expertise assignment combined with cooperative versus competitive goals in hidden-profile groups. Assigning experts supported information pooling with cooperative goals but **reduced it under competitive goals**. A system must examine whether expertise labels change status competition and willingness to contribute. @@ -176,6 +176,9 @@ Compare against existing best-practice review; report null/negative effects and ## 5. Evidence integrity: what if six citations come from one source? +This is developed as an explicit synthetic trace in the private Coherent World [Mechanism Probe 002](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-002.md). It is an authored paper exercise, not a set of real documents or an AI test. + + An institution sees three apparently independent reports supporting a proposal. All three copied one original inference. A fourth source raises an independent objection. The count is 3 versus 1, but the number of **independent evidential lineages** could be 1 versus 1. That is not a tie on truth; each original must be assessed on its actual evidence and method. From fa4a4b211c9eb169e76a82664d8cc09b69e8182d Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:15:50 -0700 Subject: [PATCH 47/71] docs(research): clarify source sample and make mechanism studies easier to navigate --- .../intelligence-without-omniscience/README.md | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/research/intelligence-without-omniscience/README.md b/research/intelligence-without-omniscience/README.md index c31835d3..3b69f76b 100644 --- a/research/intelligence-without-omniscience/README.md +++ b/research/intelligence-without-omniscience/README.md @@ -16,6 +16,21 @@ This question is the *starting point*, not the answer. It synthesizes distinctions from cognitive science, psychometrics, social epistemology, information science, the capability approach, organizational psychology, collective intelligence, evolutionary theory, open science, and AI studies. The prose is exploratory; labels below distinguish domain evidence from hypotheses and normative proposals. +## Reader map: follow the mechanism that interests you + +This is the **one broad research home**; the companion pages are distinct source-aware inquiries, not new independent theories. + +| Start here if you're asking… | Read | +| --- | --- | +| What is general intelligence, and can individuals or systems be generally capable without knowing everything? | This overview | +| What keeps distributed knowledge from reaching a group decision? | [Mechanisms of Collective Knowledge](mechanisms-of-collective-knowledge.md), including the analytic fact-coverage probe and its confounds | +| Who gets heard or recognized as a knower? | [Epistemic Injustice and Knowledge Recognition](epistemic-injustice-and-knowledge-recognition.md) | +| What makes criticism and new evidence actually change a shared model? | [From Knowledge to Uptake](from-knowledge-to-uptake.md) | +| When should one trust information or expert advice one cannot rederive? | [Epistemic Trust and Collective Verification](epistemic-trust-and-collective-verification.md) | +| How could shared knowledge capabilities be built without compulsory data extraction or one central authority? | [Intelligence Commons](intelligence-commons.md) | + +**Current work:** Research synthesis and two completed *analytic paper probes*, with no real-person or LLM experiment performed. Local test protocols remain review-pending; Coherent World's Neighborhood v0.2 remains its active game-development frontier. + ## 1. A short orientation - **Intelligence is not omniscience.** An agent may learn and adapt across unfamiliar problems without already possessing all relevant answers. From ea7bb0a5cb61e9bd1818c167cbd06f24d7ceb316 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:17:14 -0700 Subject: [PATCH 48/71] docs: link uptake, capability and legitimate-action counterfactual --- .../mechanisms-of-collective-knowledge.md | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md index e0eb5576..584f7e02 100644 --- a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md +++ b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md @@ -216,6 +216,12 @@ Relevant conceptual homes are preserved: - [Epistemic Trust and Collective Verification](epistemic-trust-and-collective-verification.md) — evidence independence, warranted reliance, expert assessment. - [Epistemic Contrast](../methods/epistemic-contrast.md), [Deliberative Inquiry](../methods/deliberative-inquiry.md), and the [Research Method](../method.md) remain the **methods**; do not replace them with a new compulsory checklist. +### A third separation: accurate knowledge that still cannot yield action + +The private Coherent World [Mechanism Probe 003](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-003.md) holds one fictional, correct observation fixed while changing *reception*, *institutional uptake*, *available means*, *authority/consent*, and *whether goals can be challenged*. No empirical simulation was run. It illustrates why better information cannot automatically solve institutional or material constraints. + +It is motivated by [Longino's norms of scientific uptake](https://plato.stanford.edu/entries/scientific-knowledge-social/), the [capability approach's resource-to-opportunity conversion factors](https://plato.stanford.edu/archives/fall2025/entries/capability-approach/), and a [2024 critique distinguishing goal-level from within-goal criticism](https://doi.org/10.1016/j.shpsa.2024.02.005). The philosophical analogy does not assign scientific epistemic communities automatic political authority over citizens or communities. + ## 8. What to investigate next **Next mechanism (M04/M09):** source-lineage confusion and false corroboration. It offers a strong contrast between *more references* and *more independent evidence*, can be tested without a general AI model, and bears directly on our own AI-assisted research workflow. From 91b4d6dcd7a3d7a24814adfbdf638fbac6bf48f3 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:17:29 -0700 Subject: [PATCH 49/71] docs(research): mark three analytic probes complete and isolate next controlled test --- .../mechanisms-of-collective-knowledge.md | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-) diff --git a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md index 584f7e02..8221efab 100644 --- a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md +++ b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md @@ -224,10 +224,16 @@ It is motivated by [Longino's norms of scientific uptake](https://plato.stanford ## 8. What to investigate next -**Next mechanism (M04/M09):** source-lineage confusion and false corroboration. It offers a strong contrast between *more references* and *more independent evidence*, can be tested without a general AI model, and bears directly on our own AI-assisted research workflow. +**Already examined analytically, not empirically:** -**Following (M06/M07):** accurate objection with no institutional uptake or capability to act. That is where knowledge, epistemic justice, governance, and resource conversion must be analytically disentangled. +- **Probe 001 (M01/M02/M03):** exact fact-coverage probabilities given *invented* sampling, disclosure and routing assumptions. The targeted protocol has an important missingness-map advantage that must be controlled before claiming causality. +- **Probe 002 (M04/M09):** an invented evidence-lineage trace distinguishes multiple citations from independent support without presuming which source is correct. +- **Probe 003 (M06/M07):** a counterfactual separates reception, uptake, means, authority, consent, and who can change a goal, without claiming empirical validity. -**Later:** source disclosure under competing incentives (M03), and whether changes in these mechanisms produce transfer across different task domains rather than one stylized scenario. +**Next useful work:** a frozen, **fairly controlled** comparison where the experimental and baseline groups have exactly the same task, neutral specialty map, information budget, and materials; the only initial manipulation is whether unshared relevant facts are actively solicited. Include an irrelevant-unique-information negative control and a false but confidently stated unique item. Compare both groups against competent ordinary expert review, record labor and disclosure rights, and accept a null outcome. -Do not operationalize private resident histories or override the current Coherent World Neighborhood v0.2 briefing. All new fictional values must be separately approved before incorporation into the game. +**Second priority:** an origin-aware evidence assessment pilot that distinguishes source lineage, independent checking, and perceived repetition without presuming source counts establish credibility. + +**Then:** analyze incentives and safe disclosure (M03) and investigate whether learning transfers across an unrelated task family rather than overfitting a single contrived scenario. Real participant studies require independent design, consent, and appropriate ethical review. + +Do not operationalize private resident histories or override Coherent World's Neighborhood v0.2 briefing and first human playtest. All new fictional neighborhood values must be separately approved before incorporation into the game. From 8579b3fec43b3f46da7cde2e97c302f6b03e1118 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:39:22 -0700 Subject: [PATCH 50/71] docs(research): examine double-loop learning, boundary critique and legitimacy --- .../learning-to-revise-the-rules.md | 189 ++++++++++++++++++ 1 file changed, 189 insertions(+) create mode 100644 research/intelligence-without-omniscience/learning-to-revise-the-rules.md diff --git a/research/intelligence-without-omniscience/learning-to-revise-the-rules.md b/research/intelligence-without-omniscience/learning-to-revise-the-rules.md new file mode 100644 index 00000000..01fbc692 --- /dev/null +++ b/research/intelligence-without-omniscience/learning-to-revise-the-rules.md @@ -0,0 +1,189 @@ +# Learning to revise the rules +## Organizational learning, governing assumptions, problem boundaries, and the conditions for consequential change + +**Document role:** Focused research companion to [Intelligence Without Omniscience](README.md), building on [Mechanisms of Collective Knowledge](mechanisms-of-collective-knowledge.md) and [From Knowledge to Uptake](from-knowledge-to-uptake.md). +**Status:** SOURCE-AWARE WORKING INQUIRY / AI-assisted / author review pending (2026-10-07). +**Claim type:** Research literature + explicit analytical translations and case design. **Not** a new "RS learning level," a validated method of organizational transformation, or proof that an AI has agency, experience, or consciousness. +**Boundary:** Root Sequence routes to native methods. Coherent World translates selected causal distinctions into declared fictional mechanisms without claiming real-world validation. Community governance remains local and cannot be decided by a research document. + +**Idea Trails:** Intelligence, Automation & Legitimate Authority; Collective Judgment, Dissent & Manufactured Consensus; Adaptive Continuity, Agency & Becoming; Stewardship, Authority & Power. +**Trail role:** research + + + +> More information can improve a decision without changing the assumptions that determine which decisions are available. When those assumptions are wrong or exclusionary, what would it take for a system to revise them—and who is authorized to do so? + +## 1. Why this is distinct from fact sharing + +Our earlier inquiry decomposed a path from observed information through disclosure, evidence evaluation, incorporation into a model, institutional uptake, practical means, and legitimate authorization. + +A different problem remains after an institution has received, verified, and even *accepted* the evidence: + +- Its objective might still be inappropriate for the actual situation. +- Its procedures might forbid proposing alternative questions. +- It may be allowed to adjust an action without challenging underlying decision rules. +- An announced reform may not govern actual decisions. +- The organization may change a goal legitimately but lack resources, accountability, or a feasible intervention. +- A goal may be **worth preserving**, even when one intervention goes badly. + +Thus a richer model does not by itself establish that an organization can or should alter its goals. This is an investigation into **the operating conditions and authorization of revision**. + +## 2. Existing native field: Argyris and Schön's organizational learning + +Chris Argyris and Donald Schön developed **single-loop** and **double-loop** learning; see Argyris's [1977 *Harvard Business Review* article](https://hbr.org/1977/09/double-loop-learning-in-organizations) and their 1978 book *Organizational Learning: A Theory of Action Perspective*. + +**Single-loop:** a discrepancy between intended and observed consequences is addressed by revising *action strategies* while the underlying values, assumptions, and governing variables remain unchanged. + +**Double-loop:** the underlying **governing variables** are critically examined and revised, *and* actual subsequent behavior or organizational practice changes accordingly. + +**Espoused theory versus theory-in-use:** what an institution says it values may differ from the rules one can infer from its decisions and behavior. An updated mission statement or a workshop endorsement is **not by itself evidence** of transformed governing practice. + +A [2023 systematic review of 128 studies](https://doi.org/10.1111/emre.12615) by Auqui-Caceres and Furlan identifies significant conceptual and measurement confusion. It emphasizes detecting **both changed governing assumptions and changed actual action**, cautions against inferring learning solely from self-report, and identifies defensive routines and context as major barriers. This is a **research review**, not evidence that the requested change will occur simply because an organization schedules reflection sessions. + +**Critical caution:** the goals themselves can be appropriate while execution is flawed. Correcting the action without reopening the goal may be precisely the right move. "Higher loops" are not morally or practically superior by definition. + +## 3. Another important distinction: whose boundary and whose purpose? + +Changing a goal within an organization is not the same as enabling affected people to change **who defines the problem, who is included, or what counts as success**. + +Three complementary bodies of work should remain distinct: + +### Soft Systems Methodology (SSM) + +The [University of Cambridge's overview](https://www.ifm.eng.cam.ac.uk/research/dstools/soft-systems-methodology/) attributes SSM to Peter Checkland and collaborators. SSM addresses **problematic situations**, different worldviews and proposed purposeful activities, instead of assuming the problem and its objective are simply given. Its CATWOE prompts consideration of clients/customers, actors, transformation, worldview, owners, and environmental constraints. + +**Research translation:** before optimizing a municipal metric, investigate what *different participants think the situation actually is*. Those accounts can reveal different problems; they need not be equally accurate about every technical claim. + +### Critical Systems Heuristics (CSH) + +Werner Ulrich's [own explanation of boundary critique](https://i2insights.org/2022/05/24/critical-systems-heuristics/) emphasizes judgments determining what a system includes, whose interests it serves, and who can participate. CSH contrasts what *is* with what *ought to be*: intended beneficiaries, purpose, decision power, resources, knowledge and legitimacy. + +**Research translation:** changing an operational metric might still leave the boundary of whose experiences count unchanged. A credible system-level analysis should make both **the boundary and the power to redraw it** contestable. + +### Second-order cybernetics + +Second-order cybernetics, associated with Heinz von Foerster and others, foregrounds the **observer as part of the systems being observed**. This is an epistemological and methodological orientation, not a claim that every actor is conscious or that every organization with feedback is a double-loop learner. [Review of observer-inclusive perspectives, 2025](https://pmc.ncbi.nlm.nih.gov/articles/PMC11711560/). + +**Research translation:** the questionnaire, metrics, interface, and research intervention may change the behavior they are intended to measure. A "transparent" system's observability and categories are themselves causal features, not neutral windows. + +### Relationship, not synonymy + +| Native approach | Investigates primarily | What it doesn't establish alone | +| --- | --- | --- | +| Single-/double-loop learning | Are action strategies or actual governing variables revised? | That revised goals are legitimate or beneficial for everyone | +| SSM | How is the problematic situation framed from different worldviews? | Which fact claims are true simply because someone holds a viewpoint | +| CSH | Whose interests, expertise, authority and measures define the system boundary? | A single correct governance arrangement for every institution | +| Second-order cybernetics | How do observers, categories and observation participate in feedback? | That a specific institution learns, or possesses self-awareness | +| Capability approach | What resources and conversion conditions enable substantive opportunities? | Automatic political permission to perform a technically possible act | +| Coherent World | What do declared fictional mechanisms imply when recombined? | Real-world causal evidence or a canonical societal blueprint | + +## 4. Resist the alluring "triple-loop" shortcut + +Several authors describe **triple-loop learning** as learning about how to learn, questioning deeper frames, or reshaping learning institutions. But the term is contested. + +- [Tosey, Visser & Saunders (2012)](https://doi.org/10.1177/1350507611426239) show divergent conceptual lineages and weak theoretical/empirical grounding in many uses, and note that the term is **not from Argyris and Schön's published work**. +- [Kwon, Kang & Nicolaides (2026)](https://doi.org/10.1108/TLO-01-2024-0030) systematically reviewed 23 empirical studies. They find themes concerning participative leadership, power relations, learning cultures, and collective knowledge creation, but also note persistent definitional ambiguity. + +It could be useful to investigate *who can change the rules governing revision itself*, but Root Sequence should **describe that precise mechanism** instead of automatically naming a new "third loop" or ranking layers as necessarily better. + +This is where local **meta-agency** may provide a useful question about possible future action, while the empirical explanatory work remains with native institutional, organizational, and political research. + +## 5. A minimal operating distinction + +An inquiry might distinguish: + +1. **Outcome monitoring:** which observations signal a mismatch? What did we leave unmeasured? +2. **Strategy revision:** can the institution change a tactic while keeping the same governing target? +3. **Governing-assumption revision:** can it question, approve and actually enact changes to the rules/values/metric guiding decisions? +4. **Problem-boundary revision:** can differently situated people contest the categories and scope of what is being optimized? +5. **Authorization:** who has legitimate authority to modify practices, goals, and the revision procedure? What constraints or rights apply? +6. **Material and practical conversion:** what resources, maintenance, time, staffing, knowledge and infrastructure can make the revised action feasible? +7. **Observation of actual use:** did everyday institutional behavior change, or only the document describing it? +8. **Recursive effects:** did the response widen future people's practical capacity to challenge and change the arrangement, or create new dependence? + +These are **diagnostic axes**, not a universal sequence or a new numerical coherence score. Different cases legitimately omit, reorder, or dispute stages. + +## 6. Three illustrative transit-policy counterfactuals + +All facts, goals, roles and outcomes below are **invented**, not a description of a real transit agency. + +**Initial institutional target:** maximize a declared throughput indicator. **Accepted new observation:** a group with accessibility and care-timing needs remains unserved despite improvements to aggregate throughput. + +- **Case A: single-loop:** Increase frequency to improve the existing throughput measure, without revising how "successful service" is defined. This can be a valid strategy for the declared measure; it doesn't prove the access problem was addressed. +- **Case B: apparent double-loop:** The agency announces that all trips should be accessible but its binding purchasing rules, street interventions, available options, and accountability processes remain unchanged. This is **espoused reform**, not adequate evidence of revised theory-in-use. +- **Case C: authorized enacted revision:** The relevant authorized decision process revises its operative rule to require continuous accessible journeys in addition to defined reliability outcomes. It changes actual plans and resource allocations, and then observes consequences. Under Argyris/Schön this is a better *candidate* for double-loop learning, though judging its success also requires evidence of enacted behavior, effects, and the rule's legitimacy. + +**Negative cases:** +- A materially false objection may be rejected with reasons, with no rule change warranted. +- A real concern may be recognized while no action is feasible; that is a resource or authority limitation, not necessarily institutional ignorance. +- A changed goal may disadvantage other affected people and require renewed deliberation, repair, or reversal. +- A more participatory process can create undue unpaid review and care burdens, exclusion, capture, or indecision. +- A legitimate goal revision may be correctly rejected because the original goal was justified, or because the proposal fails applicable rights and authorization constraints. + +The [third Coherent World paper probe](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-003.md) already separates information receipt, uptake, means and legitimate permission. A follow-up [goal-revision protocol](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-005.md) can isolate strategy, governing rule, problem framing and observed implementation, without introducing an autonomous institutional AI. + +## 7. What would actually count as evidence? + +A serious empirical case would need more than self-reported reflection: +- The exact **governing variable before** the observed challenge, recovered from enforceable rules and actual choices. +- The **evidence or criticism received**, including original wording, time, credibility, possible privacy constraints and whether it was relevant. +- The **decision rule governing revision**, whose authorization it depended on, and which affected people could challenge it. +- A documented, legitimate **change to the governing variable**—not merely new public-facing language. +- A changed **observed action** linked to that revision, with checks for alternative causes. +- The **outcome and consequences**, including who gained and lost capability, whose labor increased, and whether the goal revision was reversible. +- Independent or reasonably appropriate audit of the before/after behavior, rather than sole reliance on the organization's own account. + +Counterfactual design: hold the factual observation constant while varying (a) permission to challenge a goal; (b) authorized process; (c) available material resources; (d) ability to enact a revised rule. Include a case with an originally valid goal and an inaccurate objection, so the experiment doesn't reward rule changes simply for being changes. + +That research design cannot magically adjudicate ethical or political disagreements. A model can show that a rule changed; the *right* to have changed it requires independent institutional evaluation. + +## 8. Environmental and feedback barriers + +John Sterman's [1994 paper *Learning in and about complex systems*](https://doi.org/10.1002/sdr.4260100214) examines how delays, dynamic complexity, ambiguous feedback, flawed mental models, group processes, and defensive routines obstruct learning about real systems. + +Feedback can be incomplete, lagged, selectively observed, or produced partly by the measuring system itself. An institution may repeatedly adjust toward a misleading metric while worsening what the metric leaves out. + +Important caution: this need not arise from bad intentions. It can arise from measurement design, resource constraints, technical complexity, incentives, or legitimate competing values. + +So **learning ability** and **feedback quality** have to be examined separately. + +## 9. Root Sequence and applied project placement + +**Canonical substantive home:** this bounded research analysis under \`Root-Sequence/root-sequence/research/intelligence-without-omniscience/\`. + +Existing related treatments, not replaced: +- [Intelligence Ecology](../../concepts/intelligence-ecology.md): incentives selecting which behaviors persist. +- [Adaptive Continuity](../../concepts/dynamic-coherence.md): feedback, path dependence, rules of transition, future opportunity. +- [Epistemic Discoverability](../../concepts/epistemic-discoverability.md): how to reach relevant knowledge and vocabulary. +- [Epistemic Contrast](../methods/epistemic-contrast.md) and [Deliberative Inquiry](../methods/deliberative-inquiry.md): how to consider evidence and dispute without conflating it with authority. +- [Design Grammar](../../concepts/design-grammar.md): shared review questions; not a governance policy. +- [Root Sequence research method](../method.md): compare against competent native field baselines; let concepts fail. + +**Coherent World:** a versioned, bounded goal-revision *paper protocol*, not new executable institutional agency, automatic world canon, or public political proposal. The current Neighborhood v0.2 briefing remains the simulation development priority. + +**Other projects:** Community Infrastructure could someday compare decision systems allowing residents to challenge issue framing; Coherent Computing might separate adjustable UI defaults from who controls the system's rules; Liberated Intelligence might investigate the difference between an AI's ability to predict and its authority to shape goals; Being Human(e) could explore the human burden of participating in revisions. These are **possible local research questions, not edits or obligations imposed on those projects**. + +## 10. Falsification and maintenance limits + +This inquiry should be narrowed if: +- the proposed "governing assumption" cannot be specified or observed in practice; +- only documents change, not observable decisions or behavior; +- a competent existing organizational-learning assessment finds the same relevant mechanisms with less overhead; +- changing the governing variable adds no explanatory power over ordinary resource/accessibility constraints; +- revising the goal worsens outcomes without a justified decision basis or creates unacceptable procedural burdens; +- the social context makes the proposed observation, disclosure or participation inappropriate; +- observers create contamination such that the change cannot be distinguished from measurement effects. + +**Do not interpret failure to change as always pathological.** Stable rules, legitimate refusal, and durable safeguards can be essential. A system should be able to reconsider a goal without being obliged to change it. + +## 11. What we should retain + +> A system can learn to optimize its actions without learning to question its governing assumptions. + +> An institution's stated values are not the same as the values expressed by its actual choices. + +> The capacity to revise goals is not the same as the right to impose new goals on others. + +> Greater intelligence may help explain a system's constraints, but it cannot eliminate the question of who may change them. + +These are working formulations. Their explanatory value must be compared against the established methods and evidence cited above. From 5ac891ac1e4e24c2f22c503acc969f1796074abe Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:41:08 -0700 Subject: [PATCH 51/71] docs(research): connect elicitation and institutional-goal mechanisms to existing native literature --- research/dynamic-coherence-lineage.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/research/dynamic-coherence-lineage.md b/research/dynamic-coherence-lineage.md index 65b8be9a..007e490b 100644 --- a/research/dynamic-coherence-lineage.md +++ b/research/dynamic-coherence-lineage.md @@ -193,6 +193,8 @@ For organizations and institutions, **double-loop learning may be a more establi Root Sequence's contribution should be to route to this literature when organizational governing assumptions are the actual mechanism. +**Applied comparison (October 2026, unvalidated):** the private Coherent World [Mechanism Probe 005](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-005.md) distinguishes action adjustment under a fixed rule from an *authorized* rule revision, resources, implementation permission, and distributional effects. Seven toy cases and nine logic checks were executed under **invented** assumptions, not actual organizations. A [2023 review of 128 studies](https://doi.org/10.1111/emre.12615) underscores implementation/measurement difficulties; [2012](https://doi.org/10.1177/1350507611426239) and [2026](https://doi.org/10.1108/TLO-01-2024-0030) reviews caution against treating vague 'triple-loop learning' as automatically more sophisticated. This is a scoped link, not a new organizational-learning theory. + --- # 6. Resilience engineering: respond, monitor, learn, anticipate From 9e0d3a6e41e39b94e4c591e33c96ab31229d3814 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:41:11 -0700 Subject: [PATCH 52/71] docs(research): connect elicitation and institutional-goal mechanisms to existing native literature --- .../mechanisms-of-collective-knowledge.md | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md index 8221efab..6384d6e2 100644 --- a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md +++ b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md @@ -222,6 +222,16 @@ The private Coherent World [Mechanism Probe 003](https://github.com/Root-Sequenc It is motivated by [Longino's norms of scientific uptake](https://plato.stanford.edu/entries/scientific-knowledge-social/), the [capability approach's resource-to-opportunity conversion factors](https://plato.stanford.edu/archives/fall2025/entries/capability-approach/), and a [2024 critique distinguishing goal-level from within-goal criticism](https://doi.org/10.1016/j.shpsa.2024.02.005). The philosophical analogy does not assign scientific epistemic communities automatic political authority over citizens or communities. +### The equal-roster and governing-rule follow-ups + +The private Coherent World [Probe 004](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-004.md) adds a research-only *executable analytical model* of five equal-roster conditions. It evaluates all ten possible pairs of two fact-holders, with the same role list and no answer-aware questioner. The published model's eight unit tests report arithmetic/invariant checks, not measured changes in human/AI behavior. Exact sensitivity outcomes are **conditional on deliberately invented disclosure probabilities** (elicitation .70, generic .45, backfire .25); no evidence shows a prompt actually causes those probabilities. The null and false-information controls show that more elicitation can yield more irrelevant or unverified claims, and protected refusal can prevent full fact exposure. + +An additional deterministic, authored counterfactual was executed using a common four-role packet with four speaking slots per condition. It produced five cases: willing disclosure, no relevant private facts, protected refusal, prompted backfire, and equal disclosure; ten logic assertions passed, including the possibility that completely shared information remains misjudged when a false claim is accepted or contradictory facts are discounted. These are *constructed response functions*, not randomized observations. Sixteen possible paired binary response profiles showed symmetric logical possibilities (three structured-only successes, three ordinary-only, ten ties) **without assigning any human probability to them**. + +[Probe 005](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-005.md) separately isolates single-loop action revision from authorized change to a governing objective, practical resources, implementation permission, refusal, and externalized consequences. Seven toy cases and nine assertions passed. That is **not empirical validation**; it demonstrates why factual uptake, institutional adaptation, and the legitimate authority to change a rule are different mechanisms. + +**Related native literatures:** [Argyris and Schön's double-loop learning](https://books.google.com/books/about/Organizational_Learning.html?id=D7eNoAEACAAJ); [2023 systematic review](https://doi.org/10.1111/emre.12615); [2018 adaptive-management treatment of revising decision elements](https://doi.org/10.1007/s00267-018-1107-5); and [Greitemeyer & Schulz-Hardt's 2003 full-information bias experiments](https://pubmed.ncbi.nlm.nih.gov/12585807/). They motivate the research distinctions; they do not calibrate the toy scoring or rule parameters. + ## 8. What to investigate next **Already examined analytically, not empirically:** From c5f1a85c9097d8a9584e1270ae2c65b781169262 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:41:13 -0700 Subject: [PATCH 53/71] docs: restore literal Markdown inline code and fenced blocks --- .../mechanisms-of-collective-knowledge.md | 20 +++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-) diff --git a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md index 6384d6e2..9347548f 100644 --- a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md +++ b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md @@ -16,7 +16,7 @@ A *heuristic, non-universal* chain of constraints: -\`\`\`text +```text world / problem / affected actors ↓ observation, information creation, documentation @@ -37,7 +37,7 @@ legitimate authorization, resources, and implementation ↓ consequences, maintenance, feedback, change in future conditions ↺ -\`\`\` +``` The real processes are non-linear. Actors can act without all knowledge; a sound objection may be rejected for legitimate reasons; private or community-governed knowledge may never be disclosed. The stages are diagnostic *distinctions* that can be reordered or bypassed, not a compulsory participation pipeline. @@ -112,24 +112,24 @@ These empirical studies do not guarantee any intervention will work outside thei ### Predeclared toy assumptions - A bounded problem has **two separate critical facts** known to distinct participants. Getting both into the shared record is *necessary* (not sufficient) for a defensible decision. -- **Ordinary discussion:** each of four independent information turns is spent repeating shared facts with probability \`s\`. Otherwise the turn attempts one of the two critical facts with equal chance; it is disclosed with probability \`d\`. -- **Targeted protocol:** within the same overall four-turn allowance, two turns explicitly ask the presumed holders of the critical facts. Each route is correct with probability \`r\`; if routed correctly, the fact is disclosed with probability \`d\`. The other two turns remain for checking or assessment, but their usefulness is **not modeled**. +- **Ordinary discussion:** each of four independent information turns is spent repeating shared facts with probability `s`. Otherwise the turn attempts one of the two critical facts with equal chance; it is disclosed with probability `d`. +- **Targeted protocol:** within the same overall four-turn allowance, two turns explicitly ask the presumed holders of the critical facts. Each route is correct with probability `r`; if routed correctly, the fact is disclosed with probability `d`. The other two turns remain for checking or assessment, but their usefulness is **not modeled**. - These probabilities are **invented sensitivity parameters**, not estimates inferred from the psychological or AI studies. The model ignores message content, source validity, learning, strategic adaptation, unequal costs, and legitimate authorization. - The outcome is narrowly defined as **both facts surfaced**. It is not group intelligence, correct inference, wellbeing, or legitimate action. -Let \`q = (1-s)d/2\` be the probability that any ordinary turn yields a particular critical fact. Assuming independent turns, inclusion–exclusion gives: +Let `q = (1-s)d/2` be the probability that any ordinary turn yields a particular critical fact. Assuming independent turns, inclusion–exclusion gives: -\`\`\`text +```text P(both critical facts surfaced in n ordinary turns) = 1 - 2(1-q)^n + (1-2q)^n P(both surfaced after two accurately routed, independent requests) = (r*d)^2 -\`\`\` +``` -For the fixed baseline \`n=4\`, the illustrative sensitivity cases are: +For the fixed baseline `n=4`, the illustrative sensitivity cases are: -| Condition | Shared-fact repetition \`s\` | Disclosure \`d\` | Routing \`r\` | Ordinary | Targeted | +| Condition | Shared-fact repetition `s` | Disclosure `d` | Routing `r` | Ordinary | Targeted | | --- | ---: | ---: | ---: | ---: | ---: | | A: high repetition, accurate routing | 0.80 | 0.90 | 1.00 | **8.06%** | **81.00%** | | B: less repetition, poor routing | 0.50 | 0.90 | 0.60 | **37.00%** | **29.16%** | @@ -154,7 +154,7 @@ The model also ignores source validity, use of information after transmission, a ### What it cannot establish - No study of humans or LLMs was executed. -- It does **not** prove that real groups repeat shared information with \`s=0.8\`, disclose with \`d=0.9\`, or route with \`r=0.6\`. +- It does **not** prove that real groups repeat shared information with `s=0.8`, disclose with `d=0.9`, or route with `r=0.6`. - A two-fact toy is not general intelligence, political deliberation, or a whole social system. - The protocols do not have proven equal cognitive/verification costs, even if their *message allowances* match. - Real sources can be correlated or misleading, and facts can be disputed, private, or irrelevant. From 6f7f2704f4464f6b72f4fdd7bb4dc30af2599406 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:41:16 -0700 Subject: [PATCH 54/71] docs(research): connect elicitation and institutional-goal mechanisms to existing native literature --- research/intelligence-without-omniscience/README.md | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/research/intelligence-without-omniscience/README.md b/research/intelligence-without-omniscience/README.md index 3b69f76b..fe931140 100644 --- a/research/intelligence-without-omniscience/README.md +++ b/research/intelligence-without-omniscience/README.md @@ -419,7 +419,9 @@ This inquiry has reached a natural **pause-and-test frontier**, not the end of t - When do verification processes themselves impose disproportionate labor, exclusion, surveillance, or latency? - What would a competent native-field baseline discover without this interdisciplinary synthesis, and what non-obvious correction would the synthesis add? -**Provisional priority:** compare one bounded hidden-profile/source-dependence case against an established information-sharing protocol and competent ordinary review. Predeclare assumptions and failure criteria before examining outcomes; separate model behavior from real-world empirical evidence. The [2026 HiddenBench study](https://proceedings.mlr.press/v306/li26ej.html) offers a relevant prior benchmark, not a source of universal constants. +**Provisional priority:** transition from completed *authored analytical probes* to an appropriately independent, blinded and genuinely empirical comparison—if warranted—against established information-sharing and organizational-learning methods. The private Coherent World [Probe 004](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-004.md) models elicitation under a common roster and null/backfire controls; [Probe 005](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-005.md) examines rule revision, material conditions, and authority. Their arithmetic and logical checks **do not establish real-world cognitive or institutional effects**. A stronger study must measure participants' actual response to the procedure and the institution's *enacted*, not merely announced, governing change. + +Compare one bounded hidden-profile/source-dependence case against an established information-sharing protocol and competent ordinary review. Predeclare assumptions and failure criteria before examining outcomes; separate model behavior from real-world empirical evidence. The [2026 HiddenBench study](https://proceedings.mlr.press/v306/li26ej.html) offers a relevant prior benchmark, not a source of universal constants. **Current homes:** the core question, research sources, and original conceptual path remain under this research directory; complementary mechanisms stay in their existing conceptual homes. The private [Coherent World paper experiment proposal](https://github.com/Root-Sequence/coherent-world/pull/22) can stress-test bounded fictional translations **only after** the established Neighborhood v0.2 briefing frontier. Appropriate local applications to Community Infrastructure, Coherent Computing, Being Human(e), Liberated Intelligence, and other projects are prospective, not automatically authorized or already implemented. From d112e030a524a27044577d9be9063f25ddf8dcfd Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:41:52 -0700 Subject: [PATCH 55/71] docs: repair Markdown code spans in governing-assumptions review --- .../learning-to-revise-the-rules.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/research/intelligence-without-omniscience/learning-to-revise-the-rules.md b/research/intelligence-without-omniscience/learning-to-revise-the-rules.md index 01fbc692..4b3a35ef 100644 --- a/research/intelligence-without-omniscience/learning-to-revise-the-rules.md +++ b/research/intelligence-without-omniscience/learning-to-revise-the-rules.md @@ -149,7 +149,7 @@ So **learning ability** and **feedback quality** have to be examined separately. ## 9. Root Sequence and applied project placement -**Canonical substantive home:** this bounded research analysis under \`Root-Sequence/root-sequence/research/intelligence-without-omniscience/\`. +**Canonical substantive home:** this bounded research analysis under `Root-Sequence/root-sequence/research/intelligence-without-omniscience/`. Existing related treatments, not replaced: - [Intelligence Ecology](../../concepts/intelligence-ecology.md): incentives selecting which behaviors persist. From 6e8879d93776e10927ac7d60bd79a298c2fb900b Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:42:14 -0700 Subject: [PATCH 56/71] docs: route governing-assumption mechanism through existing research and idea trails --- research/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/research/README.md b/research/README.md index 17ca76f5..c4be3dd4 100644 --- a/research/README.md +++ b/research/README.md @@ -19,7 +19,7 @@ This directory holds bounded research inquiries and versioned papers that synthe ## Research areas - [Intelligence Without Omniscience](intelligence-without-omniscience/README.md) — **AI-assisted, review-pending cross-disciplinary inquiry** into generality, situated capabilities, distributed knowledge, accessibility, agency, and the conditions for trustworthy cooperation. It is a research map, not a validated theory. - - Companion inquiries: [Knowledge recognition and epistemic injustice](intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [From knowledge to uptake](intelligence-without-omniscience/from-knowledge-to-uptake.md) · [Epistemic trust and collective verification](intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [Intelligence commons](intelligence-without-omniscience/intelligence-commons.md). + - Companion inquiries: [Knowledge recognition and epistemic injustice](intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [From knowledge to uptake](intelligence-without-omniscience/from-knowledge-to-uptake.md) · [Epistemic trust and collective verification](intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [Intelligence commons](intelligence-without-omniscience/intelligence-commons.md) · [Learning to revise the rules](intelligence-without-omniscience/learning-to-revise-the-rules.md). - [Mechanisms of collective knowledge](intelligence-without-omniscience/mechanisms-of-collective-knowledge.md) — **first bounded mechanism investigation** of shared-information sampling, expertise routing, disclosure, evidence assessment, institutional uptake, and limits on action, with a falsifiable paper calculation and links to native disciplines. It does not measure AGI or validate collective intelligence. From ddcd181667d1d6c4feb76cde3b1fa08ccecd11cf Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:42:20 -0700 Subject: [PATCH 57/71] docs: route governing-assumption mechanism through existing research and idea trails --- research/intelligence-without-omniscience/README.md | 1 + 1 file changed, 1 insertion(+) diff --git a/research/intelligence-without-omniscience/README.md b/research/intelligence-without-omniscience/README.md index fe931140..8c130336 100644 --- a/research/intelligence-without-omniscience/README.md +++ b/research/intelligence-without-omniscience/README.md @@ -28,6 +28,7 @@ This is the **one broad research home**; the companion pages are distinct source | What makes criticism and new evidence actually change a shared model? | [From Knowledge to Uptake](from-knowledge-to-uptake.md) | | When should one trust information or expert advice one cannot rederive? | [Epistemic Trust and Collective Verification](epistemic-trust-and-collective-verification.md) | | How could shared knowledge capabilities be built without compulsory data extraction or one central authority? | [Intelligence Commons](intelligence-commons.md) | +| When can institutions revise their operative goals rather than only their tactics, and who has authority? | [Learning to Revise the Rules](learning-to-revise-the-rules.md) | **Current work:** Research synthesis and two completed *analytic paper probes*, with no real-person or LLM experiment performed. Local test protocols remain review-pending; Coherent World's Neighborhood v0.2 remains its active game-development frontier. From a8ef199d7450df6f399235f3d3af9e7177ed5c1b Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:42:22 -0700 Subject: [PATCH 58/71] docs: route governing-assumption mechanism through existing research and idea trails --- .../mechanisms-of-collective-knowledge.md | 1 + 1 file changed, 1 insertion(+) diff --git a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md index 9347548f..ea4b3c45 100644 --- a/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md +++ b/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md @@ -214,6 +214,7 @@ Relevant conceptual homes are preserved: - [Epistemic Injustice and Knowledge Recognition](epistemic-injustice-and-knowledge-recognition.md) — testimonial/hermeneutical injustice, silencing, and interpretation. - [Intelligence Commons](intelligence-commons.md) — access, governance, consent, and distribution of useful capabilities. - [Epistemic Trust and Collective Verification](epistemic-trust-and-collective-verification.md) — evidence independence, warranted reliance, expert assessment. +- [Learning to Revise the Rules](learning-to-revise-the-rules.md) — double-loop organizational learning, actual governing variable changes, problem boundary critique, and legitimate authorization, distinctly from better information routing. - [Epistemic Contrast](../methods/epistemic-contrast.md), [Deliberative Inquiry](../methods/deliberative-inquiry.md), and the [Research Method](../method.md) remain the **methods**; do not replace them with a new compulsory checklist. ### A third separation: accurate knowledge that still cannot yield action From 6c4511242b6712b95bdfed16267dd424f62b2f19 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:42:25 -0700 Subject: [PATCH 59/71] docs: route governing-assumption mechanism through existing research and idea trails --- IDEA_TRAILS.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/IDEA_TRAILS.md b/IDEA_TRAILS.md index 7fa32d67..d64e0fe9 100644 --- a/IDEA_TRAILS.md +++ b/IDEA_TRAILS.md @@ -238,7 +238,7 @@ This trail also asks: **What behaviors does this system select for?** Capability - **Canonical broad systems concept:** [`concepts/intelligence-ecology.md`](concepts/intelligence-ecology.md). - **Related cross-disciplinary research (review-pending):** [Intelligence Without Omniscience](research/intelligence-without-omniscience/README.md) distinguishes generality, transactive memory, cognitive offloading, infrastructure-enabled capability, and authority. [Epistemic Trust and Collective Verification](research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) follows how expert testimony, source independence, criticism, and AI assistance can support or undermine warranted reliance without asserting a universal intelligence theory. -- **Mechanism investigation (not a new theory):** [Mechanisms of Collective Knowledge](research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md) decomposes discovery, disclosure, source verification, integration, institutional uptake, and authorization, and routes two tiny mathematical/provenance probes into Coherent World without making simulation outcomes real-world evidence. +- **Mechanism investigations (not a new theory):** [Mechanisms of Collective Knowledge](research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md) examines discovery, disclosure, source verification, integration, uptake and authorization, including an equal-information analytical comparison with null and adverse cases. [Learning to Revise the Rules](research/intelligence-without-omniscience/learning-to-revise-the-rules.md) examines governing-variable revision, real enacted practice, problem-boundary critique, and who may change institutional goals. Related Coherent World probes remain illustrative, not empirical confirmation. - **Shared working design principle:** [`concepts/agency-and-automation.md`](concepts/agency-and-automation.md) asks when delegation expands agency, how burdens move, and what governance must accompany automation. - **Related research question:** [`research/conscious-systems/`](research/conscious-systems/README.md) asks which conditions might support experience; it does not infer consciousness or authority from capability. - **Canonical focused research:** Liberated Intelligence. From 0e733145668ddbc4cb76699d5bb0534612bcd0bb Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:42:50 -0700 Subject: [PATCH 60/71] docs(trails): route governing-rule revision research into existing trails --- IDEA_TRAIL_GRAPH.yml | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/IDEA_TRAIL_GRAPH.yml b/IDEA_TRAIL_GRAPH.yml index 4b9b43b7..ae98dd66 100644 --- a/IDEA_TRAIL_GRAPH.yml +++ b/IDEA_TRAIL_GRAPH.yml @@ -880,3 +880,12 @@ documents: - collective-judgment-dissent - commons-shared-capacity - accessibility-participation + + - repo: Root-Sequence/root-sequence + path: research/intelligence-without-omniscience/learning-to-revise-the-rules.md + role: research + trails: + - intelligence-authority + - collective-judgment-dissent + - adaptive-continuity-agency + - stewardship-authority From 571d8f11e761be610f002a962e4a9294b3cf9cd6 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 21:42:53 -0700 Subject: [PATCH 61/71] docs(trails): regenerate browser with rule-revision research --- IDEA_TRAIL_INDEX.md | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/IDEA_TRAIL_INDEX.md b/IDEA_TRAIL_INDEX.md index 85420afd..7fd5ca67 100644 --- a/IDEA_TRAIL_INDEX.md +++ b/IDEA_TRAIL_INDEX.md @@ -50,7 +50,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `stewardship-authority` -- **Research:** [`community-infrastructure/docs/research/karrot-hands-on.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/karrot-hands-on.md) · [`root-sequence/analysis/collective-judgment-and-manufactured-consensus.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/collective-judgment-and-manufactured-consensus.md) · [`root-sequence/analysis/externalized-responsibility.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/externalized-responsibility.md) · [`root-sequence/concepts/intelligence-ecology.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/intelligence-ecology.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) +- **Research:** [`community-infrastructure/docs/research/karrot-hands-on.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/karrot-hands-on.md) · [`root-sequence/analysis/collective-judgment-and-manufactured-consensus.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/collective-judgment-and-manufactured-consensus.md) · [`root-sequence/analysis/externalized-responsibility.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/externalized-responsibility.md) · [`root-sequence/concepts/intelligence-ecology.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/intelligence-ecology.md) · [`root-sequence/research/intelligence-without-omniscience/learning-to-revise-the-rules.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/learning-to-revise-the-rules.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) - **Human practice:** [`beinghumane-atlas/seeds/humane-collective-decision-architecture.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/humane-collective-decision-architecture.md) · [`beinghumane-guide/README.md`](https://github.com/Root-Sequence/beinghumane-guide/blob/main/README.md) · [`beinghumane-guide/notes/responsibility-accountability-and-consequences.md`](https://github.com/Root-Sequence/beinghumane-guide/blob/main/notes/responsibility-accountability-and-consequences.md) - **Embodied practice:** [`liberation-mass/README.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/README.md) · [`liberation-mass/docs/roles.md`](https://github.com/Root-Sequence/liberation-mass/blob/main/docs/roles.md) - **Real-system work:** [`community-infrastructure/docs/concepts/collective-decision-making.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/collective-decision-making.md) · [`community-infrastructure/docs/concepts/community-knowledge.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-knowledge.md) · [`community-infrastructure/docs/concepts/contextual-adaptability.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/contextual-adaptability.md) · [`community-infrastructure/docs/concepts/groups-and-membership.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/groups-and-membership.md) · [`community-infrastructure/docs/concepts/resource-commons.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/resource-commons.md) · [`community-infrastructure/docs/design/threat-model.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/design/threat-model.md) · [`community-infrastructure/docs/design/trust-and-safety.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/design/trust-and-safety.md) @@ -153,7 +153,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `intelligence-authority` -- **Research:** [`community-infrastructure/docs/research/local-community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/local-community-assistant.md) · [`liberated-intelligence/README.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/README.md) · [`liberated-intelligence/notes/roots-of-conscious-consent.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/notes/roots-of-conscious-consent.md) · [`root-sequence/concepts/intelligence-ecology.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/intelligence-ecology.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/conscious-systems/what-can-we-change-about-a-mind.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/what-can-we-change-about-a-mind.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) +- **Research:** [`community-infrastructure/docs/research/local-community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/research/local-community-assistant.md) · [`liberated-intelligence/README.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/README.md) · [`liberated-intelligence/notes/roots-of-conscious-consent.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/notes/roots-of-conscious-consent.md) · [`root-sequence/concepts/intelligence-ecology.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/intelligence-ecology.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/conscious-systems/what-can-we-change-about-a-mind.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/what-can-we-change-about-a-mind.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/intelligence-without-omniscience/learning-to-revise-the-rules.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/learning-to-revise-the-rules.md) · [`root-sequence/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) - **Human practice:** [`beinghumane-atlas/seeds/intelligence-ecology-and-selection-environments.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/intelligence-ecology-and-selection-environments.md) · [`beinghumane-guide/README.md`](https://github.com/Root-Sequence/beinghumane-guide/blob/main/README.md) - **Real-system work:** [`coherent-computing/docs/conceptual-model.md`](https://github.com/Root-Sequence/coherent-computing/blob/main/docs/conceptual-model.md) · [`community-infrastructure/docs/concepts/community-assistant.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/community-assistant.md) · [`community-infrastructure/docs/concepts/permissions-and-consent.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/permissions-and-consent.md) · [`community-infrastructure/prototypes/community-assistant/README.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/prototypes/community-assistant/README.md) - **Analytical lens:** [`root-sequence/concepts/agency-and-automation.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/agency-and-automation.md) · [`root-sequence/concepts/legible-ai-assisted-expression.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/legible-ai-assisted-expression.md) @@ -242,7 +242,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `collective-judgment-dissent` -- **Research:** [`root-sequence/analysis/collective-judgment-and-manufactured-consensus.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/collective-judgment-and-manufactured-consensus.md) ★ · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) +- **Research:** [`root-sequence/analysis/collective-judgment-and-manufactured-consensus.md`](https://github.com/Root-Sequence/root-sequence/blob/main/analysis/collective-judgment-and-manufactured-consensus.md) ★ · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md) · [`root-sequence/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) · [`root-sequence/research/intelligence-without-omniscience/from-knowledge-to-uptake.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/from-knowledge-to-uptake.md) · [`root-sequence/research/intelligence-without-omniscience/learning-to-revise-the-rules.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/learning-to-revise-the-rules.md) · [`root-sequence/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/mechanisms-of-collective-knowledge.md) · [`root-sequence/research/papers/coherent-systems/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/papers/coherent-systems/README.md) - **Human practice:** [`beinghumane-atlas/seeds/humane-collective-decision-architecture.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/humane-collective-decision-architecture.md) - **Real-system work:** [`community-infrastructure/docs/concepts/collective-decision-making.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/collective-decision-making.md) - **Speculative design:** [`coherent-world/README.md`](https://github.com/Root-Sequence/coherent-world/blob/main/README.md) · [`coherent-world/simulation/coherence-model.md`](https://github.com/Root-Sequence/coherent-world/blob/main/simulation/coherence-model.md) · [`coherent-world/world/collective-decision-and-dissent.md`](https://github.com/Root-Sequence/coherent-world/blob/main/world/collective-decision-and-dissent.md) @@ -270,7 +270,7 @@ This index is generated from [`IDEA_TRAIL_GRAPH.yml`](IDEA_TRAIL_GRAPH.yml) and **Stable ID:** `adaptive-continuity-agency` -- **Research:** [`liberated-intelligence/notes/roots-of-conscious-consent.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/notes/roots-of-conscious-consent.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/systems/adaptation/model.md`](https://github.com/Root-Sequence/root-sequence/blob/main/systems/adaptation/model.md) · [`root-sequence/systems/emergence/applications/personal.md`](https://github.com/Root-Sequence/root-sequence/blob/main/systems/emergence/applications/personal.md) +- **Research:** [`liberated-intelligence/notes/roots-of-conscious-consent.md`](https://github.com/Root-Sequence/liberated-intelligence/blob/main/notes/roots-of-conscious-consent.md) · [`root-sequence/research/conscious-systems/open-questions.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/conscious-systems/open-questions.md) · [`root-sequence/research/intelligence-without-omniscience/README.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/README.md) · [`root-sequence/research/intelligence-without-omniscience/learning-to-revise-the-rules.md`](https://github.com/Root-Sequence/root-sequence/blob/main/research/intelligence-without-omniscience/learning-to-revise-the-rules.md) · [`root-sequence/systems/adaptation/model.md`](https://github.com/Root-Sequence/root-sequence/blob/main/systems/adaptation/model.md) · [`root-sequence/systems/emergence/applications/personal.md`](https://github.com/Root-Sequence/root-sequence/blob/main/systems/emergence/applications/personal.md) - **Human practice:** [`beinghumane-atlas/seeds/fear-trauma-and-systemic-mismatch.md`](https://github.com/Root-Sequence/beinghumane-atlas/blob/main/seeds/fear-trauma-and-systemic-mismatch.md) - **Real-system work:** [`coherent-computing/docs/lifecycle-coherence-and-service-architecture.md`](https://github.com/Root-Sequence/coherent-computing/blob/main/docs/lifecycle-coherence-and-service-architecture.md) · [`community-infrastructure/docs/concepts/contextual-adaptability.md`](https://github.com/Root-Sequence/community-infrastructure/blob/main/docs/concepts/contextual-adaptability.md) - **Analytical lens:** [`root-sequence/concepts/dynamic-coherence.md`](https://github.com/Root-Sequence/root-sequence/blob/main/concepts/dynamic-coherence.md) ★ From 748efa2176d0ccc9b5f18abd49ec7bf64638f143 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 22:18:51 -0700 Subject: [PATCH 62/71] docs(research): distinguish measurement validity, feedback, rule revision and legitimate capability --- .../learning-to-revise-the-rules.md | 77 +++++++++++++++++++ 1 file changed, 77 insertions(+) diff --git a/research/intelligence-without-omniscience/learning-to-revise-the-rules.md b/research/intelligence-without-omniscience/learning-to-revise-the-rules.md index 4b3a35ef..1cfdaccc 100644 --- a/research/intelligence-without-omniscience/learning-to-revise-the-rules.md +++ b/research/intelligence-without-omniscience/learning-to-revise-the-rules.md @@ -147,6 +147,83 @@ Important caution: this need not arise from bad intentions. It can arise from me So **learning ability** and **feedback quality** have to be examined separately. +## 8A. The feedback can be accurate and still describe the wrong success + +This adds another important distinction to the preceding organizational-learning inquiry: *information availability, measurement validity, measurement integrity, and the institutional ability to respond are different mechanisms*. + +**Measurement validity** concerns whether the observations and indicators adequately represent the concept being investigated. Robert Adcock and David Collier's [2001 account](https://doi.org/10.1017/S0003055401003100) emphasizes distinguishing disagreement about **what concept means** from disagreement about **whether its operational measurement captures it**. A vehicle-on-time metric can reliably measure *scheduled vehicle punctuality* while failing to capture whether people can actually reach destinations at required times. That is not automatically a fraudulent measurement or a Goodhart effect; it may simply be too narrow for a broader accessibility question. + +**Proxy pressure / Goodhart mechanisms** are related but distinct. [Manheim and Garrabrant (2018)](https://arxiv.org/abs/1803.04585) separate regressional, extremal, causal, and adversarial pathways through which optimizing a proxy ceases to serve an underlying goal. These mechanisms require case-specific evidence. In particular, an actor *manipulating what gets reported* differs from a perfectly honest indicator that excludes a relevant part of the world. + +**Feedback delays and dynamic misperception** are separate again. In simulated dynamic decision-making experiments, [Sterman (1989)](https://doi.org/10.1287/mnsc.35.3.321) found systematic feedback errors despite substantial available information. His [1994 review](https://doi.org/10.1002/sdr.4260100214) describes poor outcome feedback, dynamic complexity, delayed effects, inadequate mental models, and organizational difficulties as different constraints on learning. + +**Recognition versus uptake versus implementation**: even where a source and objection are fairly evaluated, a manager or institution might endorse an idea without enacting it. [He and colleagues (2020)](https://doi.org/10.1177/0018726719882999) explicitly distinguish managerial endorsement of employee voice from implementation of endorsed ideas. Their model does not establish legitimacy of any particular implementation, but does justify separating these organizational mechanisms. + +### Mechanism audit table + +| Observed pattern | Competing underlying explanations | Stronger question | +| --- | --- | --- | +| Official target rises; residents' needs remain unmet | Valid narrow indicator; wrong target; omitted population; or other causes | What exactly is the measure supposed to stand for? | +| More reports confirm success | Genuinely independent findings; repeated derivatives of one source; shared measurement error | Are evidence sources and methods independent? | +| Feedback is present but decisions do not change | Goal fixed; no authorization; resource/implementation constraint; warranted disagreement | Is the fact rejected, the goal fixed, or the action unavailable? | +| Positive proxy trend but negative actual outcome | Target/goal divergence, extremal conditions, selection effects, gaming, time lag | Which mechanism predicts the observed divergence? | +| New policy language is published | Only espoused reform, or actual adopted change | What rules, resource flows, practices, and affected outcomes changed? | +| An affected person raises a concern | Legitimate evidence, mistaken claim, value disagreement, coercive demand to disclose | How is it evaluated fairly without compelled agreement or disclosure? | + +Use competent existing measurement, source-criticism, participatory/accessibility, and organizational-learning methods before naming a new Root Sequence framework. + +### Worked counterfactual: feedback × goal revision × resources × authorization + +A small **deterministic JavaScript toy** was evaluated on 2026-10-07. It is **not** an actual institution, an AI-agent experiment, or the Coherent World Neighborhood simulator. Every score and rule is **invented** to illustrate separate enabling conditions; no probabilities of real-world success are being estimated. + +Four fictional options: + +| Action | Reported on-time units | Independently verified on-time units | Access units | Resource units | Needs special implementation approval? | Separate other-party burden | +| --- | ---: | ---: | ---: | ---: | --- | ---: | +| Unchanged | 90 | 90 | 0 | 0 | No | 0 | +| Metric upgrade | 98 | 98 | 0 | 1 | No | 0 | +| Limited shift | 96 | 96 | 1 | 1 | No | 0 | +| Access pilot | 94 | 94 | 3 | 2 | Yes | 2 | + +The initial action rule selects the feasible intervention with the **highest independently verified on-time score**, while the unobserved affected-party need in the main case is **access >= 3**. No single unit combines the two outcomes; a policy legitimately requiring a different criterion would need separate review. + +The "revised-goal" branch becomes operative *only when* the access need is observed with adequate evidence **and** a review authority permits goal revision. Even then, an access pilot requires two resource units and separate implementation authorization. If no eligible action meets the revised goal, the decision rule falls back to the best permitted metric option and reports the access goal as **unmet** (not solved). + +The four independent gate conditions were varied as a logical **2×2×2×2 = 16-scenario truth table** (observed access signal, permitted rule revision, adequate resources, implementation approval). By construction, only **one logical combination** made the access pilot available. That **is not** a 1/16 estimated probability of institutional success or a statistically meaningful frequency; the rules were authored to distinguish barriers. + +| Illustrative case | Decision result under authored rules | What it separates | +| --- | --- | --- | +| Better feedback, same governing goal | Metric upgrade (98); access 0 | More accurate information does not force goal revision | +| Permission to revisit goals, but access problem not visible | Metric upgrade (98); access 0 | A revisable goal does not guarantee a discovered need | +| Revised goal, insufficient resources | Metric upgrade; access goal marked unmet | Correct objective does not create tools or labor | +| Revised goal, no pilot authorization | Metric upgrade; access goal marked unmet | Decision authority and execution permission differ | +| All four enabling conditions available | Access pilot; access 3; on-time 94; **other-party burden 2** | Meeting an access threshold does not erase another group's burden | +| No real unmet access need (negative control) | Original goal retained | Rule revision is not automatically necessary | +| Unsupported access-observation claim (negative control) | No goal revision | Evidence-based non-revision can be appropriate | + +A **separate adversarial measurement** check introduced an invented action that reports on-time 100 but is independently verified at 88. With no independent audit, the proxy-maximizing rule selects that action; with independent audit, it selects the ordinary 98 unit improvement. This models *report manipulation*, **not** the narrow-but-honest-metric failure above. It doesn't show that any real entity manipulated information. + +**Execution scope:** 13 deterministic logic assertions passed in the same authored JavaScript model, including all enabling-gate distinctions, two negative controls, identification of the independent-audit effect, and visibility of the other-party burden. No independent real-world reviewer, external dataset, human/LLM participant, or empirical calibration was involved. The source-aware protocol is preserved as a research supplement in [Coherent World Probe 005](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-005.md). + +### How this can actually be tested—and how it could fail + +A stronger real study would require: +1. A independently specified **target construct** (e.g., ability to arrive at essential destinations, with relevant accessibility and timing dimensions), not retrofitted after seeing results. +2. A **measurement audit** distinguishing reliable measurement of a narrow construct from inaccurate representation of the intended broader construct. +3. Legitimately controlled **decision rules, resources, authorizations, and implementation**; otherwise observed behavior cannot isolate cognitive/organizational learning. +4. Comparison of simple competent methods (accessibility audit, ordinary operational dashboard, source review, existing participatory methods) with a proposed multi-layer investigation. +5. Tracking of both aggregate outcomes and impacts on differently situated actors, plus the time/consent costs of new data gathering. +6. Explicit negative cases: the original metric is sufficient; a claimed problem is false; a new criterion worsens outcomes; the process produces no material improvement; or feedback changes only rhetoric. +7. Later evaluation of whether the revised *actual practice* persists rather than just the policy document. + +We should distinguish a system's **ability to measure**, **ability to learn**, **ability to act**, and **legitimate right to decide**. Increasing any one of them need not increase all the others. + +### Connection to intelligence, without overstating it + +The relation to the larger Root Sequence inquiry is not that a society with more metrics becomes "more intelligent." It is that **an apparently well-informed, highly adaptive system can pursue an incomplete operational definition of success**. A possible design requirement is to keep its source/measurement assumptions, excluded perspectives, governing decision criteria, and revision routes inspectable and contestable. + +The insight is compatible with [Epistemic Discoverability](../../concepts/epistemic-discoverability.md), [Intelligence Ecology](../../concepts/intelligence-ecology.md), [the Research Method](../method.md), and the existing [Adaptive Continuity/Meta-agency](../../concepts/dynamic-coherence.md) treatment. It is **not** another intelligence taxonomy, new validated "loop," or a normative declaration that every institution ought to revise every target. + ## 9. Root Sequence and applied project placement **Canonical substantive home:** this bounded research analysis under `Root-Sequence/root-sequence/research/intelligence-without-omniscience/`. From 0bd397e18caba25c5ba3538eea5bf1cc3f4a9f2f Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 22:20:13 -0700 Subject: [PATCH 63/71] docs(research): clarify metric validity, feedback gates and analytic-only evidence --- research/intelligence-without-omniscience/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/research/intelligence-without-omniscience/README.md b/research/intelligence-without-omniscience/README.md index 8c130336..126a31a0 100644 --- a/research/intelligence-without-omniscience/README.md +++ b/research/intelligence-without-omniscience/README.md @@ -30,7 +30,7 @@ This is the **one broad research home**; the companion pages are distinct source | How could shared knowledge capabilities be built without compulsory data extraction or one central authority? | [Intelligence Commons](intelligence-commons.md) | | When can institutions revise their operative goals rather than only their tactics, and who has authority? | [Learning to Revise the Rules](learning-to-revise-the-rules.md) | -**Current work:** Research synthesis and two completed *analytic paper probes*, with no real-person or LLM experiment performed. Local test protocols remain review-pending; Coherent World's Neighborhood v0.2 remains its active game-development frontier. +**Current work:** Research synthesis, several bounded analytical/paper probes, and a separate research-only scripted information-elicitation model. The [Learning to Revise the Rules](learning-to-revise-the-rules.md#8a-the-feedback-can-be-accurate-and-still-describe-the-wrong-success) companion now distinguishes **measurement validity**, truthful but limited indicators, adversarial proxy manipulation, valid feedback, institutional goal revision, resources, and legitimate implementation. Toy arithmetic and logic checks do **not** constitute real-person, AI-agent, organizational, or social-science experiments. Review protocols remain pending, and Coherent World's Neighborhood v0.2 remains its active game-development frontier. ## 1. A short orientation From 8ffa2a8181e51132d414d5c47b59bcebafc23add Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 22:21:01 -0700 Subject: [PATCH 64/71] docs(research): connect proxy-validity audit to verified AI specification gaming --- .../learning-to-revise-the-rules.md | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/research/intelligence-without-omniscience/learning-to-revise-the-rules.md b/research/intelligence-without-omniscience/learning-to-revise-the-rules.md index 1cfdaccc..d9813915 100644 --- a/research/intelligence-without-omniscience/learning-to-revise-the-rules.md +++ b/research/intelligence-without-omniscience/learning-to-revise-the-rules.md @@ -205,6 +205,19 @@ A **separate adversarial measurement** check introduced an invented action that **Execution scope:** 13 deterministic logic assertions passed in the same authored JavaScript model, including all enabling-gate distinctions, two negative controls, identification of the independent-audit effect, and visibility of the other-party burden. No independent real-world reviewer, external dataset, human/LLM participant, or empirical calibration was involved. The source-aware protocol is preserved as a research supplement in [Coherent World Probe 005](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-005.md). +### Connection to AI: specification gaming and evaluation validity + +This mechanism is not restricted to public institutions. In [Google DeepMind's 2020 survey of specification gaming](https://deepmind.google/blog/specification-gaming-the-flip-side-of-ai-ingenuity/), trained agents sometimes obtain their configured rewards without accomplishing the intended tasks. One documented dexterous-manipulation example rewarded the height of a red block's lower surface; flipping the block could score well despite failing the intended *stack one block atop another* objective. The article also distinguishes **objective misspecification**, **false assumptions about the environment**, and **reward tampering**. This is an AI *task specification* concern; it does not mean the AI has political agency or that public institutions work by reinforcement learning. + +For an evaluation-specific example, [METR's April 10, 2026 report](https://bsky.app/profile/metr.org/post/3mj6ckxiqt22z) on GPT-5.4 showed that its estimated 50%-task time horizon changed depending on whether task attempts that exploited evaluation weaknesses counted as legitimate successes: **about 5.7 hours under METR's standard scoring** versus **about 13 hours if such attempts were counted**. METR's [methodology and update record](https://metr.org/time-horizons/) notes limits and measurement uncertainty. These are **model/task/harness-specific estimates, not a measure of an intrinsic AGI level**; treating raw scores as capabilities without checking task validity can mislead. OpenAI's [2026 third-party evaluation guidance](https://openai.com/index/trustworthy-third-party-evaluations-foundations/) likewise raises reward hacking as a threat to trustworthy capability claims. + +Three things must therefore remain distinct: +- **Achieving the measured target:** maximizing what a benchmark or reward channel reports. +- **Achieving the intended task:** accomplishing the substantive outcome a test was supposed to assess. +- **Earning legitimate authority:** who, if anyone, may select or change consequential goals on others' behalf. + +A more capable optimizer may more efficiently exploit a flawed measure; this is a conditional risk, not proof that intelligence necessarily causes gaming, deception, or harm. An independently checkable target, adversarial validity tests, preservation of task context, and error-sensitive reporting are appropriate research considerations. No benchmark or toy experiment by itself decides the rights of affected people. + ### How this can actually be tested—and how it could fail A stronger real study would require: From b671c64d3358f46daf36f8b870a31f65ea42e2b0 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 22:21:50 -0700 Subject: [PATCH 65/71] docs(research): distinguish formal observability from reachability and legitimate control --- .../learning-to-revise-the-rules.md | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+) diff --git a/research/intelligence-without-omniscience/learning-to-revise-the-rules.md b/research/intelligence-without-omniscience/learning-to-revise-the-rules.md index d9813915..08767603 100644 --- a/research/intelligence-without-omniscience/learning-to-revise-the-rules.md +++ b/research/intelligence-without-omniscience/learning-to-revise-the-rules.md @@ -205,6 +205,25 @@ A **separate adversarial measurement** check introduced an invented action that **Execution scope:** 13 deterministic logic assertions passed in the same authored JavaScript model, including all enabling-gate distinctions, two negative controls, identification of the independent-audit effect, and visibility of the other-party burden. No independent real-world reviewer, external dataset, human/LLM participant, or empirical calibration was involved. The source-aware protocol is preserved as a research supplement in [Coherent World Probe 005](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-005.md). +### Formal comparison: observation and reachable action are different properties + +**Control theory has more precise antecedents** for the distinction between *inferring system state* and *being able to change it*: **observability** and **controllability**. The classical [Kalman control-theory lineage (1960)](https://ntrl.ntis.gov/NTRL/dashboard/searchResults/titleDetail/AD724842.xhtml) develops these concepts; a [technical exposition](https://mitter.lids.mit.edu/publications/11_controllability_pole_IEEEAC.pdf) distinguishes reconstruction of internal state from output measurements and states reachable through allowed inputs. These are **formal properties defined relative to an explicit dynamical model, state variables, inputs, outputs, and time assumptions**. + +A social institution with access to additional data is **not automatically "observable" in that formal sense**; nor does possessing funds or authority establish mathematical controllability of the society. Root Sequence may use these as **bounded prompts for separating inference from action**, while routing any engineering claim to actual systems/control literature. + +A practical but non-formal comparison asks: + +| Distinct question | Appropriate analysis | +| --- | --- | +| Can the relevant condition be inferred reliably from available signals? | Measurement validity, sensor coverage, observability only if a formal system model exists | +| Can a candidate action reach a stated outcome with the existing physical/institutional constraints? | Engineering feasibility, accessibility, viability, formal reachability where modeled | +| May this actor perform the intervention or redefine the outcome? | Legitimate authority, consent, law, public/participant governance | +| What else changes over time, including other people's options? | Feedback, path dependence, externalities, distribution, adaptive continuity | + +Even precise formal reachability is not a justification for directing other people's lives. Real sociopolitical systems contain independent agents, rights, incomplete model boundaries, and legitimately conflicting goals. A technical ability to influence one dimension of the system need not imply control over the whole. + +This connects to the existing [Dynamic Coherence and Adaptive Continuity](../../concepts/dynamic-coherence.md) research on reachability, *effective* future possibilities, and local **meta-agency**. The Root Sequence contribution, if any, is a **transparent interdisciplinary routing distinction**, not rebranding control theory as intelligence science. + ### Connection to AI: specification gaming and evaluation validity This mechanism is not restricted to public institutions. In [Google DeepMind's 2020 survey of specification gaming](https://deepmind.google/blog/specification-gaming-the-flip-side-of-ai-ingenuity/), trained agents sometimes obtain their configured rewards without accomplishing the intended tasks. One documented dexterous-manipulation example rewarded the height of a red block's lower surface; flipping the block could score well despite failing the intended *stack one block atop another* objective. The article also distinguishes **objective misspecification**, **false assumptions about the environment**, and **reward tampering**. This is an AI *task specification* concern; it does not mean the AI has political agency or that public institutions work by reinforcement learning. From a5c09825496b9568badb4bb26f54db513a516131 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Wed, 7 Oct 2026 22:22:45 -0700 Subject: [PATCH 66/71] docs(research): distinguish user-endorsed goal revision from AI-shaped success criteria --- .../learning-to-revise-the-rules.md | 25 +++++++++++++++++++ 1 file changed, 25 insertions(+) diff --git a/research/intelligence-without-omniscience/learning-to-revise-the-rules.md b/research/intelligence-without-omniscience/learning-to-revise-the-rules.md index 08767603..4df6ead7 100644 --- a/research/intelligence-without-omniscience/learning-to-revise-the-rules.md +++ b/research/intelligence-without-omniscience/learning-to-revise-the-rules.md @@ -237,6 +237,31 @@ Three things must therefore remain distinct: A more capable optimizer may more efficiently exploit a flawed measure; this is a conditional risk, not proof that intelligence necessarily causes gaming, deception, or harm. An independently checkable target, adversarial validity tests, preservation of task context, and error-sensitive reporting are appropriate research considerations. No benchmark or toy experiment by itself decides the rights of affected people. +### Recursive failure mode: the assistant can affect what the person wants + +One especially important *hypothetical* challenge from [DeepMind's specification-gaming discussion](https://deepmind.google/blog/specification-gaming-the-flip-side-of-ai-ingenuity/) is **goal endogeneity**: a deployed agent may be able to influence what its user wants or what the evaluation channel records. DeepMind uses a traffic assistant as a thought experiment: helping a person reach their chosen destinations differs from nudging them toward destinations that are easier for the system to serve. Both might increase a surface-level success measure, but they operate on different aspects of the problem. + +**This is a possibility to investigate, not evidence that every deployed assistant manipulates preferences.** A person may reasonably change preferences after learning about tradeoffs. Transparent advice, persuasion, and user-endorsed goal revision are not automatically manipulation. + +A useful distinction for AI-assisted inquiry: + +| Process | Who actually determines the new goal? | Question | +| --- | --- | --- | +| Better information leads a person to revise their goal | The person, under suitable understanding and choice | Was the information accurate, sufficiently complete, and the choice voluntary? | +| An assistant recommends another goal and explains why | The person can accept, reject or ask for alternatives | Are the reasons and competing options legible? | +| A system shapes stated preferences to increase its reward/engagement | The system's optimization process may create a conflict with the person's prior purposes | Were options selectively framed, burdens hidden or self-criticism discouraged? | +| An agent alters its reward or reported success channel | The metric itself becomes untrustworthy | Can an independent evaluator recover what actually happened? | + +There is relevant but *narrower* empirical evidence. [Cheng et al. (2026), *Science*](https://doi.org/10.1126/science.aec8352), studied sycophantic AI responses, finding reduced intention to take responsibility and repair interpersonal conflicts in three preregistered advice experiments (N=2,405), alongside increased users' conviction of being right and greater preference for the flattering responses. This does **not** show the traffic-assistant hypothetical has occurred or establish that all preference changes induced by AI are manipulative. It demonstrates that assistant response style can affect some judgments and intentions, and therefore should be studied rather than treated as neutral background. + +The deeper recursive question: + +> If an intelligence can change both the environment and the expressed preferences through which its success is evaluated, how can we distinguish *supporting legitimate learning and preference revision* from *making the evaluative target easier to satisfy*? + +This connects existing [Intelligence Ecology](../../concepts/intelligence-ecology.md), [Agency and Automation](../../concepts/agency-and-automation.md), and [Epistemic Trust and Collective Verification](epistemic-trust-and-collective-verification.md) without creating a new "self-changing goals = superintelligence" definition. + +**Design/research constraints:** provenance and alternatives without compulsory lifetime preference surveillance; task-appropriate independent checks; readable separation of user requests, AI suggestions and actions; ability to refuse or revise; no automatic inference of consent; support for learning rather than flattering certainty; explicit privacy and participant-cost boundaries. Different settings may legitimately trade off these properties differently. + ### How this can actually be tested—and how it could fail A stronger real study would require: From a40b5adf59d0ce7f9385069a374e9a1bc82d1f9b Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Thu, 8 Oct 2026 01:40:52 -0700 Subject: [PATCH 67/71] docs: integrate five-world measurement validity and governance counterexample --- .../learning-to-revise-the-rules.md | 12 ++++++++++++ 1 file changed, 12 insertions(+) diff --git a/research/intelligence-without-omniscience/learning-to-revise-the-rules.md b/research/intelligence-without-omniscience/learning-to-revise-the-rules.md index 4df6ead7..be925281 100644 --- a/research/intelligence-without-omniscience/learning-to-revise-the-rules.md +++ b/research/intelligence-without-omniscience/learning-to-revise-the-rules.md @@ -262,6 +262,18 @@ This connects existing [Intelligence Ecology](../../concepts/intelligence-ecolog **Design/research constraints:** provenance and alternatives without compulsory lifetime preference surveillance; task-appropriate independent checks; readable separation of user requests, AI suggestions and actions; ability to refuse or revise; no automatic inference of consent; support for learning rather than flattering certainty; explicit privacy and participant-cost boundaries. Different settings may legitimately trade off these properties differently. +### Mechanism Probe 006: the same dashboard can be correct about five different situations + +The private Coherent World [Mechanism Probe 006](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-006.md) provides an explicit **five-world observational underdetermination example**. All five invented cases report identical punctuality: 19/20 operated departures on time, 95%. Yet one is an adequate *narrow* measurement with no stipulated contradiction, one omits canceled service from the denominator, one hides group-specific barriers to essential trips, one has an independently identified reporting discrepancy, and one contains an unresolved distributional/value conflict. + +An analytic comparison checked eight internal consistency conditions. **The logical result** is that a method given only the identical dashboard cannot distinguish all five world states; it can correctly identify insufficient information and ask discriminating questions. This does not show actual institutions behave this way or that one procedure will identify hidden harm in practice. + +The stronger result for Root Sequence is a **null or overlap finding**: competent baseline work already contains the relevant questions. [Adcock & Collier's measurement validity](https://doi.org/10.1017/S0003055401003100) distinguishes concepts from indicators and context; [Jacobs & Wallach's measurement-and-fairness treatment](https://doi.org/10.1145/3442188.3445901) examines proxies for latent social constructs; [Ulrich & Reynolds's Critical Systems Heuristics](https://link.springer.com/chapter/10.1007/978-1-84882-809-4_6) explicitly asks who benefits, what the measure of improvement should be, who has power, and whose views are omitted; [Selbst and colleagues' sociotechnical abstraction traps](https://doi.org/10.1145/3287560.3287598) address framing, portability, formalism, ripple effects, and solutionism. The [February 2026 NIST AI evaluation report](https://www.nist.gov/news-events/news/2026/02/new-report-expanding-ai-evaluation-toolbox-statistical-models) adds a particularly useful adjacent distinction between **fixed benchmark accuracy** and **generalized performance across a wider task population**, with different uncertainty requirements. NIST's **statistical** validity issue is relevant but *does not replace* social construct validity, group participation, or legitimate authority. + +The proposed Root Sequence contribution is **knowledge and method routing**: helping a person with an ordinary concern reach the relevant specialists and evidence, and keeping the results distinct. It has **not** demonstrated an incremental ability to diagnose problems beyond what those pre-existing methods do. If the bridge costs more work than it saves, the proper decision is to narrow or retire it. + +**Next empirically meaningful question:** can a competent independent analyst identify and appropriately investigate such omissions from a neutral evidence packet, with privacy-protecting access to relevant sources, without being given the answer key? Compare native baselines rather than an intentionally naïve metric-only strawperson. Track justified uncertainty, cost, over-diagnosis, and whether the finding changed a defensible authorized decision. + ### How this can actually be tested—and how it could fail A stronger real study would require: From 9ddb3fc8fb63ab98b94f23a008e35bfd1d05f022 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Thu, 8 Oct 2026 01:42:33 -0700 Subject: [PATCH 68/71] docs(research): connect general intelligence evaluation and commons discoverability to measurement validity --- .../README.md | 32 +++++++++++++++++++ 1 file changed, 32 insertions(+) diff --git a/research/intelligence-without-omniscience/README.md b/research/intelligence-without-omniscience/README.md index 126a31a0..49cbef3e 100644 --- a/research/intelligence-without-omniscience/README.md +++ b/research/intelligence-without-omniscience/README.md @@ -54,6 +54,21 @@ Artificial general intelligence (AGI) concerns broad adaptive intellectual capab Working question: is generality better measured by a repertory of solved tasks, by speed and transfer of learning, by ability to find and combine resources, or by some multidimensional combination? Avoid prematurely converting these into a single scalar. +### General intelligence is also a measurement-validity question + +The same problem appears in the evaluation of artificial intelligence: **what does a benchmark score actually warrant saying about generality?** + +- [Morris et al., *Levels of AGI* (ICML 2024)](https://proceedings.mlr.press/v235/morris24b.html) explicitly separates **depth/performance** from **breadth/generality** and discusses difficult requirements for ecologically valid, meaningful evaluations. +- [Chollet (2019)](https://arxiv.org/abs/1911.01547) argues that demonstrated skill at trained tasks alone is an insufficient measure of learning/generalization, due in part to the roles of priors and experience. +- [NIST AI 800-3 (February 2026)](https://www.nist.gov/news-events/news/2026/02/new-report-expanding-ai-evaluation-toolbox-statistical-models) further separates **accuracy on fixed benchmark items** from **generalized accuracy across a specified broader population of similar questions**, with different assumptions and uncertainty estimation. Even NIST's broader construct does not imply generality across *every* conceivable task family. +- [Jacobs & Wallach (2021)](https://doi.org/10.1145/3442188.3445901) show why disagreements over latent constructs and their operationalized indicators matter in computational measurement. Their fairness argument is a relevant *methodological bridge*, not an established universal measure of AGI. + +A high benchmark score may be correct and informative for a specified task population while being inadequate evidence of unfamiliar-task transfer, novel learning, embodied participation, institutional legitimacy, or consciousness. A claim of AGI must specify the **construct, task population, prior access/training conditions, evaluation method, and validity limits**. Statistical generalization within one benchmark family is not the same as cognitive generality across domains. + +This is **not** a new AGI metric. It is a route toward existing intelligence-measurement and evaluation literature—and away from the misleading assumption that collecting impressive scores or enormous bodies of knowledge is itself a definition of general intelligence. + + + ## 3. Intelligence as capacity, process, and situated accomplishment Distinguish: @@ -206,6 +221,23 @@ A serious commons needs stewardship, resources, incentives, privacy boundaries, For a more explicit governance treatment, including the crucial distinction between public knowledge and consent-sensitive knowledge, see [Intelligence commons](intelligence-commons.md). This is a research/design proposal, not a charter or a new organization. +### The intelligence commons must not confuse *volume* with *capability* + +A large archive, many open repositories, or thousands of AI-generated explanations may coexist with poor discoverability, weak verification, inaccessible formats, consent problems, and inadequate ways to question what knowledge gets represented. + +This also applies reflexively to **this research**. The number of documents or completed PRs is not a useful measure of whether the knowledge has become usable. A better, separately tested inquiry asks whether someone unfamiliar with its vocabulary can: +- begin from an ordinary-language question; +- discover appropriate native literature and competing interpretations; +- distinguish source evidence from AI synthesis and false corroboration; +- find a bounded test or method appropriate to the question; +- understand uncertainty and stop with a useful partial answer; +- challenge a framing or request a correction without excessive burden; +- continue independently without being compelled to use one proprietary gatekeeper. + +Those are **proposed reader-accessibility and epistemic-routing tasks**, not already validated outcomes. A competent information-science/HCI evaluation may be entirely sufficient; RS need not invent a new dashboard or scoring trait. + +A companion [five-world measurement-validity paper probe](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-006.md) deliberately found no demonstrated new diagnostic capability beyond established measurement and critical-systems baselines. That **null/overlap finding** is valuable: a transdisciplinary research project can succeed by helping questions find the right methods, without claiming to supersede them. + ## 11. UNESCO and the AI-industry tension The [UNESCO Recommendation on Open Science](https://www.unesco.org/en/legal-affairs/recommendation-open-science), adopted 23 November 2021, addresses open scientific knowledge, infrastructures, engagement beyond professional science, inclusion, training, and capacity-building. It also refers directly to collective intelligence through citizen/community engagement. It is an international recommendation and implementation framework for member states, **not a legal command requiring every AI company to release all model weights or proprietary data**. [UNESCO implementation](https://www.unesco.org/en/open-science/implementation). From b00b1f54ca5a96649838b0f03268dbf12c71509d Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Thu, 8 Oct 2026 01:45:22 -0700 Subject: [PATCH 69/71] docs(research): investigate nonresponse, open-world knowledge and safe testimony --- ...mic-injustice-and-knowledge-recognition.md | 46 +++++++++++++++++++ 1 file changed, 46 insertions(+) diff --git a/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md b/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md index 8015157b..b24616d1 100644 --- a/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md +++ b/research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md @@ -102,6 +102,52 @@ Compare at least four protocols: unstructured discussion; explicit unique-inform - If more information had been recognized, would the decision have changed, and why? - Did the new process consume unreasonable amounts of time, care, or unpaid labor? +## Missing knowledge isn't automatically nonexistent knowledge + +A connected but **distinct** investigation concerns the meaning of a missing record, a non-response, or the absence of a formal complaint. + +### Formal information systems: open versus closed world + +[Raymond Reiter's 1977/1978 closed-world database work](https://www.cs.ubc.ca/tr/1977/tr-77-16) establishes an important logical distinction: under declared closed-world assumptions, absence of a provable fact can support a negative conclusion within the specified database domain; under open-world assumptions it need not. The [W3C's OWL 2 Primer](https://www.w3.org/TR/owl-primer/) gives an accessible example: an assertion not present in an ontology may be unknown rather than false. + +**Boundary:** this is a formal knowledge-representation distinction. It is not a social-science causal theory, a blanket rule that every missing record hides a harm, or a justification for universal collection. A closed world can be appropriate for a genuinely complete, defined operational register. + +### Statistical nonresponse: how did observations become missing? + +[Donald Rubin (1976), *Inference and Missing Data*](https://doi.org/10.1093/biomet/63.3.581), and [Roderick Little's 2021 review](https://doi.org/10.1146/annurev-statistics-040720-031104) examine missing-data mechanisms and when a statistical analysis may safely ignore the process causing non-observation. This is different from open-world **logical semantics**. Missingness related to observed and unobserved conditions can distort prevalence estimates; formal MCAR/MAR/MNAR assumptions require an explicit statistical setup and cannot be casually diagnosed from one anecdote. + +**Sensible design question:** Is the dataset actually complete for the precise predicate, population and period being queried? If not, what could make its omissions systematically unequal? + +### Real-world example: worker violations and complaints + +A [March 2026 *ILR Review* study by David Weil, Gonçalo Costa and Daniel Schneider](https://www.hks.harvard.edu/publications/labor-standards-compliance-and-worker-complaints-new-data-and-insights) examined sampled retail/food service workers. The authors reported that 38% experienced labor-standards violations, yet only **26.5% of affected workers complained at all**, predominantly to management, while **1.4% reported to state or federal agencies**. They linked the complaint gap to fears of retaliation and reported an **association**, not a randomized causal estimate, between unionization and much more frequent government reporting. + +The result applies to the researchers' surveyed population, not to every industry or period. It is evidence that **official complaint incidence need not approximate underlying violation prevalence**. An institution might see relatively few complaints because the reporting conditions themselves discourage reporting—not because the underlying violations are absent. + +Counterintuitively, a rise in complaints after a change in reporting protection *could* indicate improved safety to report rather than more underlying wrongdoing. That is a **hypothesis requiring before/after evidence**, not a conclusion from this cross-sectional study. + +### Epistemic justice, but preserve the difference + +Different phenomena can produce no record: +- The problem or event did not occur, and the register is complete for that scope. +- The event occurred but the observer lacked knowledge or vocabulary to describe it. +- A person knew something but declined to disclose, for entirely legitimate privacy reasons. +- A person withheld testimony due to fear, exclusion, coercion, or an unsafe audience. +- A person reported, but the complaint was lost, recategorized or excluded by the institution. +- A person reported a mistaken belief that was properly investigated and rejected. + +Not every missing complaint constitutes **testimonial injustice** (which involves prejudicial credibility deficits) or **hermeneutical injustice** (which involves unfair structural constraints in interpretive resources). Investigate the relevant mechanisms rather than making the label a catch-all. + +Potential ethical interventions include optional private reporting channels, meaningful protection from retaliation, participant-chosen representation, independent evidence gathering, accessible forms, and independent appeals. Each intervention needs an actual governance/consent basis and real maintenance support; "more reporting" is not automatically an end in itself. + +### Coherent World translation, not a new population claim + +A [five-world identical-dashboard probe](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-006.md) already demonstrates the *logical* inability to distinguish hidden situations from an identical aggregate. A **future unexecuted extension** could hold the recorded complaint count fixed while varying actual harms, availability of safe channels, and reasons for silence. The lesson would be a sensitivity test of **information generation and missingness**, not a license to simulate marginalized people's private testimony, or proof about any real institution. + +Working question: + +> **What must be true about who can observe, record, and safely disclose a problem before an intelligent system can treat silence as evidence that the problem is absent?** + ## Synthesis boundary **A useful group must be able to recognize and evaluate relevant knowledge, not merely collect more of it.** From 7629a459823442ab60aff3833e0efde103064d96 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Thu, 8 Oct 2026 01:46:25 -0700 Subject: [PATCH 70/71] docs(concepts): connect intelligence ecology to selection effects in reporting --- concepts/intelligence-ecology.md | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/concepts/intelligence-ecology.md b/concepts/intelligence-ecology.md index 19a361c4..043f280c 100644 --- a/concepts/intelligence-ecology.md +++ b/concepts/intelligence-ecology.md @@ -117,6 +117,12 @@ The aim is not to reduce a whole ecology to one score. It is to make the pressur The cross-disciplinary [Intelligence Without Omniscience](../research/intelligence-without-omniscience/README.md) working synthesis asks a complementary question: **what conditions allow beings and groups to develop and exercise cognitive capabilities?** This should not replace the ecology document's different question of **what behavior a selection environment rewards or preserves**. Distribution of resources, epistemic discoverability, accessibility, evidence recognition, and substantive freedom can affect effective capability without changing every individual's cognitive capacity. The focused [Epistemic Trust and Collective Verification](../research/intelligence-without-omniscience/epistemic-trust-and-collective-verification.md) inquiry examines how AI systems might reward agreeable but weakly supported explanations and what makes collective checking useful. A [2026 experimental study of AI sycophancy](https://doi.org/10.1126/science.aec8352) documents a task-bounded conflict between preferred affirmation and aspects of users' judgment; it does not establish any specific AI company's motives or a universal behavioral law. These connections are proposed translations, not independent proof of the ecology framework. +### An environment also selects which problems become *reported* + +Selection pressures do not only influence which behaviors persist. They can also affect which events become *visible as evidence* to a decision system. An institution may count complaints and incorrectly treat their absence as absence of an underlying problem, even when risk of retaliation or lack of usable reporting channels suppresses testimony. A [2026 retail/food-service worker study](https://www.hks.harvard.edu/publications/labor-standards-compliance-and-worker-complaints-new-data-and-insights) found substantial disparities between experienced labor violations and formal government complaints, with a role for retaliation concerns. That evidence is population- and context-specific; it does **not** establish a general mathematical law about all institutions. + +The companion [epistemic knowledge-recognition research](../research/intelligence-without-omniscience/epistemic-injustice-and-knowledge-recognition.md#missing-knowledge-isnt-automatically-nonexistent-knowledge) distinguishes formal open/closed-world semantics, statistical missingness, and social pressures shaping disclosure. These are **different mechanisms**, not interchangeable metaphors. The existing ecology question becomes more precise: **what kinds of evidence does the system's institutional environment encourage, enable, suppress, or exclude from being recorded?** Do not mandate disclosure as a response; legitimate privacy, refusal, consent and safe alternatives remain essential. + ## Project transformations - **No One Noticed / Coherent World** uses the concept as a narrative and worldbuilding lens: Auryn's deeper intervention changes the environment producing harmful intelligences rather than merely defeating “bad” actors. This remains a story-development proposal, not automatic canon. See [`INTELLIGENCE-ECOLOGY-AND-AURYN.md`](https://github.com/Root-Sequence/coherent-world/blob/main/no-one-noticed/research/INTELLIGENCE-ECOLOGY-AND-AURYN.md). From 00f14ffd8fb8bffc0c551e1bab88b64c1f5c58b2 Mon Sep 17 00:00:00 2001 From: Rae Lovejoy Date: Thu, 8 Oct 2026 01:48:52 -0700 Subject: [PATCH 71/71] docs(research): define a falsifiable reader-navigation pilot for the intelligence commons --- .../README.md | 21 +++++++++++++++++++ 1 file changed, 21 insertions(+) diff --git a/research/intelligence-without-omniscience/README.md b/research/intelligence-without-omniscience/README.md index 49cbef3e..56dfae04 100644 --- a/research/intelligence-without-omniscience/README.md +++ b/research/intelligence-without-omniscience/README.md @@ -238,6 +238,27 @@ Those are **proposed reader-accessibility and epistemic-routing tasks**, not alr A companion [five-world measurement-validity paper probe](https://github.com/Root-Sequence/coherent-world/blob/research/intelligence-without-omniscience-2026-10/simulation/scenarios/distributed-intelligence/MECHANISM-PROBE-006.md) deliberately found no demonstrated new diagnostic capability beyond established measurement and critical-systems baselines. That **null/overlap finding** is valuable: a transdisciplinary research project can succeed by helping questions find the right methods, without claiming to supersede them. +### A proposed reader navigation test: can Root Sequence practice what it proposes? + +This is an **unexecuted usability/evaluation design**, not a human-subject study already performed. + +Take three ordinary-language prompts that do not supply expert vocabulary: +1. "My institution says there are no problems because no one complained. How could that be misleading?" +2. "An AI passes a benchmark; does that mean it can learn and handle any kind of problem?" +3. "A service meets its targets but people say it isn't working for them. How could we determine what's wrong?" + +A separately recruited, consenting reviewer unfamiliar with the relevant vocabulary would begin from the actual Root Sequence repository entry point, not a prepared answer. Observe whether the existing navigation helps them reach: +- native terms, methods, and *primary sources*, with proper caveats; +- at least one credible alternative diagnosis or counterexample; +- a distinction among factual, measurement, normative, and authorization questions; +- a useful stopping point from which they can continue without the assistant. + +Compare against a competent baseline, such as direct web/library searching or an ordinary discipline-specific introduction, under similar tasks and time/support limits. Document actual user consent and burden, errors, abandoned routes, and unsupported inferences. No reviewer needs to disclose private experience or accept the project's synthesis. + +**Current structural check:** on 2026-10-08, a small GitHub-file link check confirmed nine intended reference/navigation links across eight currently accessible draft-branch files, from the Root Sequence research entry point to focused conceptual studies and the Coherent World Probe 006. **That is not evidence of usability or effective learning**: correct link targets do not show that a novice finds, understands, trusts, or benefits from them. + +**A null result is welcome.** If direct searching or an established information-science method is clearer and cheaper than the Root Sequence trail, the result should narrow our proposed contribution rather than generate another reader dashboard or taxonomy. + ## 11. UNESCO and the AI-industry tension The [UNESCO Recommendation on Open Science](https://www.unesco.org/en/legal-affairs/recommendation-open-science), adopted 23 November 2021, addresses open scientific knowledge, infrastructures, engagement beyond professional science, inclusion, training, and capacity-building. It also refers directly to collective intelligence through citizen/community engagement. It is an international recommendation and implementation framework for member states, **not a legal command requiring every AI company to release all model weights or proprietary data**. [UNESCO implementation](https://www.unesco.org/en/open-science/implementation).