Reconstructible ancestry
Consequential claims retain sources, transformations, assumptions, model/version state, challenges and overrides.
Transductive / constitutional systems research
PolityAI is a research proposal for a constitutional cognitive commons: public infrastructure in which consequential machine-assisted reasoning remains traceable, contestable, forkable and practically replaceable.
Decisional authority is not the same thing as cognitive authority. A human signature does not neutralize upstream control over framing, evidence, options, compression or uncertainty.
The argument
The constitutional object is not a singular AI ruler. It is the coupled process by which questions are framed, evidence selected, models invoked, options generated, disagreement compressed, decisions made and outcomes fed back into later cognition.
Consequential claims retain sources, transformations, assumptions, model/version state, challenges and overrides.
A successful challenge propagates through dependencies and triggers bounded recomputation rather than becoming a dead annotation.
Material dissent survives as an active reasoning object able to accept later evidence and test later claims.
Independence is measured by how systems fail, not inferred from provider names or model counts.
Models, vendors and operators remain replaceable without destroying standing, provenance, pending challenges or institutional memory.
Public challenge must have enough analytic capability to find meaningful defects in official machine-assisted reasoning.
Comparative field
Representative deliberation, civic platforms, computational democracy, AI mediation, public-sector assurance and commons governance already solve important parts of the problem. The research task is to identify exactly which institutional functions each supplies—and which remain outside its scope.
| Initiative | Public input | Open/forkable | Traceability | Machine mediation | Executable challenge | Vendor exit |
|---|
Taiwan / vTaiwan
Pol.is maps clusters of opinion and surfaces statements that bridge groups. It is a strong precedent for treating agenda formation and shared understanding as political infrastructure rather than mere comment collection.
Barcelona / Decidim
Decidim combines open software with explicit commitments to transparency, traceability and integrity across participatory objects and processes.
AI mediation / Habermas Machine
A 2024 Science study tested LLM mediation with 5,734 participants and a virtual citizens’ assembly. PolityAI treats this as evidence about one cognitive function—not as a complete theory of legitimate public authority.
AI listening / Talk to the City
LLMs cluster and summarize large volumes of qualitative input while preserving drill-down into individual statements. This directly addresses the compression bottleneck that ordinary consultation faces.
Technical constitutionalism
Claims, evidence, assumptions, uncertainty, dissent and dependencies live as structured objects. Documents become readable views of that deeper record.
Material state transitions append signed manifests. The current state can change; its historical ancestry cannot silently disappear.
Different models and toolchains can run the same challenge or reconstruction task. Correlated failure is measured rather than assumed away.
A challenge to a source, category, causal link, criterion or uncertainty estimate resolves the affected dependency subgraph and recomputes what can be recomputed.
Identity, standing, privacy and entitlement to challenge are governed separately from epistemic weight. Access can be proven without universal public identification.
Provider exit is tested as a constitutional transition: open matters, dissent, provenance, audit state and challenge rights must survive the move.
public reasoning state
├── claims / evidence / assumptions
├── provenance / model manifests
├── dissent / alternatives / uncertainty
├── standing / challenges / appeals
├── audit results / failure correlations
└── append-only transition history
models + retrieval + simulators + human experts
⇅ replaceable execution plane
constitutional invariants + public challenge rightsPolitical theory
PolityAI treats dependence on an unanswerable cognitive intermediary as a structural power problem even when that intermediary happens to behave benevolently. The question is not only interference, but whether another actor possesses uncontrolled capacity over the conditions of public reasoning.
SourceHabermasian public-sphere theory foregrounds the circulation of reasons between civil society and formal decision institutions. PolityAI extends the problem to the machine systems that increasingly filter, synthesize and generate those reasons before they reach formal authority.
SourceMansbridge and colleagues shift attention from an idealized single forum to the division of deliberative labor across a whole system. PolityAI applies the same systems-level discipline to human-machine cognitive pipelines.
SourceOstrom’s work is not a blueprint for AI governance, but it breaks the false binary between centralized state control and private ownership. PolityAI borrows the institutional question: under what rules can a shared resource remain governable without a single proprietor becoming sovereign over it?
SourceResearch registry
The registry distinguishes peer-reviewed studies, official policy and practice accounts, platform documentation, political theory and engineering precedents. A source can be useful without being treated as dispositive.
Project status
Markdown first.
The Google Docs source of record is deliberately untouched. Research revisions occur in versioned Markdown.
Comparisons are scope maps.
The matrix describes institutional functions; it is not a political ranking or endorsement.
Architecture is falsifiable.
Later releases will publish executable acceptance tests for reconstructibility, contestability, migration, independence and recovery.