Yale Researcher Proposes Three-Tier Public Consultation on AI

A political theorist at Yale and Oxford has published a detailed proposal for giving ordinary citizens direct input into AI governance decisions — and she argues the infrastructure to make it happen already exists.

Hélène Landemore, writing in Noema Magazine, responds to recent statements from AI researchers and economists calling for urgent action on AI policy. She notes these declarations consistently address "economists, policymakers and technology leaders" while treating the public only as beneficiaries, not participants.

Her alternative: Use the platforms people already interact with daily to conduct what would be the largest public consultation in history.

Why it matters

As AI capabilities advance rapidly, governance discussions remain confined to technical experts, corporate leaders, and government officials. Landemore's proposal offers a concrete mechanism for broadening participation before major policy decisions solidify — addressing a legitimacy gap that could undermine public trust in whatever regulatory framework emerges.

A three-tier framework

Landemore proposes a graduated approach that balances speed with deliberative depth.

The first tier would use AI platforms with massive existing user bases — ChatGPT (1 billion monthly users), Google Gemini (900 million), and Meta AI (1 billion) — to conduct a rapid global consultation. These systems could ask users direct questions about AI governance and synthesize responses into a preliminary framework within months. Results would be broken out by region to surface genuine disagreements rather than forcing artificial consensus.

The second tier would organize actual online deliberations where participants interact with each other in real time, mediated by AI facilitators. This would produce richer output than automated aggregation alone, allowing humans to test and refine the AI-generated recommendations.

The third tier would convene traditional citizens' assemblies — both local and national — combined with referenda where feasible, culminating in a global assembly that could ratify or revise the earlier work.

Precedents and proof points

Landemore cites existing evidence that such consultations can work at scale. Anthropic gathered constitutional principles from 1,000 Americans in 2023 and found that incorporating public input reduced bias in model outputs. The company later surveyed nearly 81,000 Claude users across 159 countries in a single week.

Meta ran online deliberations with 6,500 participants across 32 countries on metaverse issues in 2022, then conducted two forums on generative AI with over 1,000 participants each.

Direct appeals to tech leaders

The essay includes specific calls to action for Dario Amodei at Anthropic, Sam Altman at OpenAI, and Mark Zuckerberg at Meta, asking each to commit resources to global consultation efforts. Landemore argues these companies currently have better infrastructure for global deliberation than any government or international body.

She acknowledges the legitimacy concerns inherent in corporate-facilitated consultations. Any resulting framework would lack the binding force of treaty law, functioning instead like the Universal Declaration of Human Rights — unenforceable but carrying reputational weight that shapes behavior over time.

The proposal also calls on governments and the United Nations to provide official standing and public funding, particularly for the third tier of in-person assemblies that require state authority to convene.

The full proposal, including detailed discussion of representativeness concerns and model bias, was first reported by Noema Magazine.