Research draft: intelligence without omniscience, epistemic justice, and an intelligence commons

[raelovejoy]

Intelligence Without Omniscience — research, mechanisms, and preservation

Status: Draft / public-safe / AI-assisted / author + domain review pending. No validated general intelligence theory or world/organizational policy follows from the documents.

Substantive research

The linked research/intelligence-without-omniscience/ directory contains an overview, original idea preservation map, and companions for epistemic injustice, knowledge-to-uptake, intelligence commons, epistemic trust, and Mechanisms of Collective Knowledge. The new learning-to-revise-the-rules.md investigates Argyris/Schön double-loop learning, espoused versus enacted theory, Soft Systems Methodology, Critical Systems Heuristics, observer-inclusive cybernetics, the capability approach, and authority over changing goals. It explicitly declines to invent an RS "triple-loop" intelligence hierarchy.

Completed research-only work

• A first analytic fact-coverage model plus a stronger equal-information/five-speaking-slot follow-up. The follow-up averages all 10 role-pair assignments, compares open discussion, equal turns, elicitation, null prompt, and backfire, and includes protected refusal, irrelevant-information, and false-claim-exposure controls.
• Exact mathematical outcomes were calculated and independently cross-checked by a standard-library Python test suite (8/8 checks passed). These verify arithmetic and invariant assumptions not real-world causal effects; response probabilities are invented.
• A fictional source-lineage counterexample and a fictional institutional goal-revision counterfactual distinguish fact volume from independent evidence, and correct knowledge from resources, authorized rule change, consent, and actual practice.

Navigation and existing ownership

Updated research indexes, nearest concept pointers, existing Idea Trails, graph, and generated Idea Trail browser. Root Sequence remains the router/synthesis; cognitive science, organizational learning, governance, and critical systems methods own the claims they can actually test. Private Coherent World probes and Neighborhood v0.2 remain separate and autonomous.

Methodological warnings

• The first targeted-prompt toy had a privileged missing-fact map; later analysis makes that explicit and uses an equal-turn baseline with no location-of-missingness oracle.
• A stipulated p=.70 response probability is not empirical evidence that targeted prompts improve disclosure.
• More information can bring more irrelevant and false claims, review work, and privacy pressures.
• A declared organizational reform does not establish a change of governing variables and actual behavior; neither technical performance nor cognitive expertise grants legitimate decision authority.

No merges, website deployment, Neighborhood mechanics, private data import, policy enactment, or narrative canon changes. Review sources and competing explanations, compare against native expert methods, allow null results and retire redundant terminology.

---

October 8, 2026 — measurement validity and missingness extension

• The main Intelligence Without Omniscience synthesis now makes the generality-as-measurement link explicit (Morris et al. ICML 2024, Chollet 2019, NIST AI 800-3 February 2026, Jacobs & Wallach).
• learning-to-revise-the-rules.md links five identically reported 95% dashboard worlds to pre-existing operational audit, construct validity, sociotechnical abstraction, and critical systems boundary methods. Eight authored consistency checks passed; this is a mathematical insufficiency counterexample, not a trial of actual analysts.
• epistemic-injustice-and-knowledge-recognition.md differentiates logical open/closed-world assumptions, formal missing-data inference, and socially constrained non-reporting, using a 2026 observational study of labor violations and complaints. Neither silence nor non-response is automatically consent, injustice, or evidence of zero incidence.
• concepts/intelligence-ecology.md now connects the environment's selection pressures to which problems become observable and safely reported, with scoped source claims.
• Important null/overlap finding: the existing native methods already account for the five authored dashboard cases. No new RS diagnostic power was demonstrated; its possible value remains accessible cross-domain routing.

Structural link checks passed for the research and simulation navigation (not a usability evaluation). No main-branch merge, new repository, governance decision, public site deployment, empirical human experiment, or story canon change.