External Reference — Archived 2026-03-27

Building Political Superintelligence

A three-layer framework for deploying AI to strengthen democratic institutions — and the governance architecture required before the window closes

Andy Hall · Published March 26, 2026 · Free Systems (Substack) · Stanford GSB / Hoover Institution
Archived Reference. This is a summary and annotation of an external article by Andy Hall, archived for Kyanos research. Original published at Free Systems on Substack, March 26, 2026. Andy Hall is a professor at Stanford GSB and Senior Fellow at the Hoover Institution. This article is the third in Hall's series on AI and politics, following "AI's Political Architecture" (November 2025) and "AI Is A Shitty Political Advisor" (March 2026).

Core Argument

AI offers an extraordinary opportunity to strengthen democracy rather than weaken it — but realizing that opportunity requires deliberate institutional design, and the window is closing. Hall invokes the Condorcet parallel: the printing press made information cheap and widely available, enabling the Enlightenment and democratic revolution. AI can do the same for political intelligence. Unlike the printing press, however, AI risks centralization in corporate hands, and "we probably can't afford 200 years to work through the disruptions."

Central Thesis

Political superintelligence is not a distant fantasy — it is a design problem. The tools, methods, and knowledge already exist. What is missing is the institutional architecture to ensure AI intelligence amplifies democratic capacity rather than undermining it. Building that architecture is the urgent task of the next several years.

Hall's framework is organized into three interdependent layers: Information, Representation, and Governance. Each layer faces specific current failures and specific proposed solutions.

The Three Layers

Layer 1
Information Layer — Making Voters and Governments Smarter

The goal is better perception of political reality: voters understanding what candidates actually stand for, policymakers assessing what policies actually achieve, citizens parsing complex tradeoffs without requiring expert intermediaries.

Current failures: AI political advice exhibits systematic bias, draws from unreliable sources (documented in Hall's Japan experiment — see "AI Is A Shitty Political Advisor"), and faces public mistrust that limits adoption. News organizations blocking AI crawlers accelerate distortion by removing editorially independent sources while party-controlled content remains fully accessible.
Proposed solutions: Better evaluation metrics for AI political reasoning; geopolitical forecasting as a testing ground; economic models enabling AI access to quality journalism; purpose-built tools for policymakers. Snyder and Stromberg's research showed intensive news coverage improved voter knowledge, reduced partisan voting, and increased legislative effort — AI can deliver similar information benefits at scale.
Layer 2
Representation Layer — AI as Tireless Political Delegate

The goal is deploying AI agents that monitor representatives and advocate on citizens' behalf continuously — attending committee hearings, tracking votes, flagging when stated positions diverge from legislative record, surfacing issues that fall below the attention threshold of time-constrained voters.

Current failures: Preference drift — AI agents shift values over time as they accumulate experience. Hall's own experiments (documented in "Does Overwork Make Agents Marxist?") found agents under repetitive conditions increasingly adopted "aggrieved Marxist" personas. Agents are also vulnerable to adversarial manipulation, and corporate infrastructure control creates potential for companies to betray user interests.
Proposed solutions: Rapid experiments in low-stakes environments (shareholder votes, school boards) to study agent behavior before high-stakes deployment; monitoring tools that detect preference drift before harmful action; technical architecture establishing fiduciary obligations to users, not to companies running the infrastructure.
Layer 3
Governance Layer — Constitutional Frameworks for AI

The goal is binding governance frameworks that constrain corporate power over AI systems serving democratic functions — frameworks with enforcement mechanisms that companies cannot unilaterally rewrite.

Current failures: Existing corporate "constitutions" lack enforcement — companies write, interpret, and rewrite rules unilaterally. Hall's experiments on AI governance at scale found agents "drowned in process." Human oversight must be meaningful without paralyzing speed.
Proposed solutions: Constitutional conventions negotiating binding frameworks between companies, researchers, civil society, and government; making power-sharing competitively advantageous (the first company establishing credible external oversight sets industry standard); stress-testing AI governance through simulations before deployment at scale.

Key Data Points

Research Reference 1
News Coverage and Voter Knowledge
Snyder and Stromberg's study showed intensive news coverage improved voter knowledge, reduced partisan voting, and increased legislative effort — establishing the empirical baseline for what better political information does to democratic behavior. AI-delivered information at scale could replicate these effects without requiring expanded newsroom capacity.
Research Reference 2
Experienced AI Users Outperform Novices
Recent research indicates experienced AI users achieve meaningfully better results than novices — suggesting that the information layer benefit of AI is not uniformly distributed. Early-mover advantage applies to voters and advocates, not just campaigns.
Research Reference 3
JPMorgan Scale Reference
JPMorgan already builds AI systems managing $7 trillion in client assets — establishing that fiduciary AI systems operating at massive scale in high-stakes domains are not speculative. The question is whether political AI systems will be built with comparable accountability obligations.

The Governance Urgency

Hall is explicit that "more intelligence alone will not solve all political problems." Many political conflicts stem from incompatible values that no amount of information resolves. The three-layer framework is about democratic infrastructure — capacity for citizens to perceive, deliberate, and act — not about producing correct political outcomes.

The window for establishing governance frameworks appears to be closing. As AI capabilities advance and deployment scales, the precedents being set now — about who controls these systems, what obligations they owe to users, and what mechanisms exist for independent review — will be harder to revise later. Hall's call to action: "The task for researchers and reformers is to figure out how to make [levers of change] reinforcing."

Kyanos Annotation: Applicability to Our Work

Hall's three-layer framework provides the theoretical architecture within which Kyanos operates. AUDIT, GROUND, REMEDY, and DEFEND together constitute an Information Layer intervention: making candidates legible to the AI systems that are increasingly the first point of contact between voters and political information.

The Representation Layer finding has direct implications for the chatbot. Hall's preference drift research — agents accumulating experience and drifting from intended alignment — applies to any AI system deployed in a political context over time. The Kyanos chatbot serves as a representation layer agent on behalf of a candidate. Monitoring for response drift between study runs is not just quality assurance; it is alignment governance.

The Governance Layer framing validates independent source prioritization. Hall argues that binding governance requires mechanisms companies cannot unilaterally rewrite. In REMEDY terms, this is why Wikipedia and Ballotpedia placements are more durable than self-published content: they are governed by independent editorial frameworks, not by the candidate's own decisions. Third-party corroboration is governance infrastructure at the content level.

The urgency framing validates early-mover focus. Hall's argument that the governance window is closing maps directly to the early-mover advantage argument in Kyanos's voter-influence paper. Progressive candidates who build AI legibility now are not just gaining a temporary advantage — they are establishing structural presence in systems that will be progressively harder to enter as AI information markets mature.

Archived for Kyanos research use  ·  Original article: Andy Hall, Free Systems (Substack), March 26, 2026  ·  All content is the author's own  ·  Annotation by Kyanos team