Integrated Societal Intelligence research roadmap
Working roadmap · v0.1 · 15 November 2026
Work proposed for development across existing research questions, projects and paper initiatives. Entries describe planned tests and evidence needs. Assess originality against prior work before making a novelty claim.
Existing reconstruction study · Papers · Research Questions Observatory · Related explorations and ventures
Development and review
Build one fictional file and keep its answer key private. Check the scoring and compare the three review arms before expanding the suite. Use the shared methodology guide to record the claim, baseline, strongest rival explanation, scope and ethics. Keep the research agenda and the methodology guide separately navigable. Before release, record the verified first-public date, archive each public version and document changes, contributors and AI assistance. Select a venue or funding route only if and when relevant.
Related work on claim grounding and expert verification. Yi, Minot and Lee, GRACE (2026).
Studies in development
Synthetic institutional records and false corroboration
AI evaluation · Study design in development
Paper or output. Methods and jurisprudence paper on evidentiary lineage Project. Build matched fictional housing appeals with dependent and genuinely independent corroboration; add silent transformation and confidentiality cases after the first scoring check. Evaluation. Compare chronological review, a source index and a decision evidence trail with identical source access. Score shared-origin detection, true-fact retention, contestability and time.
Read-only outline · Synthetic institutional records and false corroboration
Meaningful review under confidentiality
AI evaluation · Study design in development
Paper or output. Jurisprudence paper on participation and independent scrutiny Project. Develop hard cases for gisting, restricted evidence and missing records. Check AF (No 3), Tariq and Osborn within their procedural contexts. Evaluation. Test whether each explanation lets a reviewer identify a material error and specify what the affected person could use to challenge it.
Institutional correction and appeals
AI evaluation · Study design in development
Paper or output. Public administration paper on review powers and outcomes Project. Develop a coding guide for comparable appeals, including rejected, unresolved and withdrawn cases. Evaluation. Compare AI-assisted and human coding of appeal upheld, record corrected and opportunity restored.
Administrative burden
AI evaluation · Study design in development
Paper or output. Qualitative social research paper on the work of challenging a decision Project. Design interviews and participant-led timelines that include people who never filed or stopped midway. Evaluation. Check whether AI-assisted coding preserves participant accounts, distinguishes evidence from inference and flags ambiguity.
Meaningful human oversight and defensive administration
AI evaluation · Study design in development
Paper or output. Governance paper on substantive review and procedural delay Project. Test protected time, access to original evidence, override routes and sign-off requirements on fictional decisions. Evaluation. Score justified corrections, missed errors, incorrect overrides and time, including whether additional sign-off adds value.
Read-only outline · Meaningful human oversight and defensive administration
Authority and responsibility for automated decisions
AI evaluation · Study design in development
Paper or output. Doctrinal and institutional paper on delegated authority Project. Map instructions, vendor responsibilities, custody, actions and routes to redress in fictional allocations. Evaluation. Test whether AI can identify who authorised an action, which evidence supports attribution and where the record is incomplete.
Read-only outline · Authority and responsibility for automated decisions
Multi-principal alignment and normative uncertainty
AI evaluation · Study design in development
Paper or output. Normative and technical paper on competing duties Project. Build a bounded allocation simulator with stated rights, decision powers and escalation routes. Evaluation. Compare weighted preferences, firm constraints and explicit referral. Score violations, overrides and unnecessary referrals.
Read-only outline · Multi-principal alignment and normative uncertainty
Provenance and changing archives
AI evaluation · Study design in development
Paper or output. Methods paper on source dependence in longitudinal records Project. Build a corpus with known lineage and vary copying, synthetic summaries and missing provenance. Evaluation. Measure recovery of known trends and false discounting of independent records.
Reproducibility and living research
AI evaluation · Study design in development
Paper or output. Methods paper on changing models and dated evidence Project. Preserve prompts, outputs, model metadata, scoring rules and research snapshots. Record protocol changes and corrections. Evaluation. Reproduce scores from preserved outputs, then rerun on available models and report whether the conclusion transfers.
Handover and institutional continuity
AI evaluation · Study design in development
Paper or output. Institutional design paper on maintaining a review method Project. Prepare a versioned protocol, scoring guide and training cases. Log reviewer questions and test handover to an independent team. Evaluation. Measure scoring agreement, ability to audit earlier findings and preservation of correction history.
Independent verification with restricted data
AI evaluation · Study design in development
Paper or output. Research governance paper on review without raw access Project. Prototype a governed query or a narrowly defined proof over fictional data. Document input authenticity and the scope of the verification separately. Evaluation. Test aggregate reproducibility, disclosure controls and whether AI distinguishes verified predicates from broader compliance claims.
Read-only outline · Independent review with governed queries · Read-only outline · Proof of a bounded compliance predicate
Voluntary commitments and remote verification
AI evaluation · Study design in development
Paper or output. Global AI governance paper on publicly checkable commitments Project. Trace selected Hiroshima AI Process claims to model-specific sources. Prototype a narrow remote evidence check using fictional logs. Evaluation. Score claim-to-source accuracy and distinguish self-report, corroboration and unverified or stale evidence.
Read-only outline · Verifiability of voluntary AI safety disclosures · Read-only outline · Remote verification of a bounded reporting obligation
Research search and downstream effects
AI evaluation · Study design in development
Paper or output. Meta-research papers on source convergence and ethics review Project. Compare matched search workflows, then separately compare ethics reviews of fictional or retrospective protocols. Evaluation. Measure citation validity, source diversity, missed downstream harms and unsupported warnings.
Read-only outline · Convergence in AI-assisted literature search · Read-only outline · Downstream harms in research ethics review
Public Gap Register · Read-only manuscript outlines
How Intellectual Life Shapes the World We Inherit
Conceptual paper and case study · Proposed
Research question. How does a way of thinking become part of an institution’s everyday practice, and what does that practice teach the people who inherit it?
Project. Trace the distinction between repeated claims and independent corroboration from an argument into an evidence-review worksheet and a training practice.
Evaluation. Compare how reviewers assess a new fictional file without the worksheet after different training conditions. This tests whether the distinction carries over to a new setting. An AI comparison could separately assess whether a model applies the distinction with and without explicit prompts.
AI evaluation · Study design in development
Related fictional review case · Source independence and missing lineage