Research map / Large language models / AI and society

Position: Democratic AI is Possible. The Democracy Levels Framework Shows How It Might Work.

ICML 2025 · Large language models › AI and society · approach: Human-centered trust and agency

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Position paper arguing democratic governance and alignment of AI is feasible and valuable, and proposing the Democracy Levels framework (L0–L5), its dimensions, and two tools (Levels Decision Tool, Democratic System Card) to guide and evaluate democratic AI decisions.

Problem

Decisions about AI with systemic societal impacts lack democratic legitimacy, and there is no shared language or evaluation standard for how democratic AI governance and alignment processes are.

Approach

Defines discrete levels (L0–L5) describing which roles (informing, specifying, binding, initiation, metagovernance) are transferred from a unilateral authority to democratic systems, paired with process-quality, delegation, and trust dimensions and practical tools for planning and evaluation.

Key results

• Proposes six Democracy Levels (L0 unilateral to L5 metagovernance) defined by which of five roles (informing, specifying, binding, initiation, metagovernance) democratic systems perform.
• Organizes evaluation into three dimension categories (process quality, delegation, trust) with 12 dimensions and a Democratic System Card.
• Classifies Anthropic's Collective Constitutional AI and Meta's Community Forums as L1 (informing) examples, and Meta's Oversight Board content decisions as L4.

Surprising findings

• Argues Meta's Oversight Board is binding per-post (L4) but was not designed to be democratically representative, showing power transfer and democratic quality can diverge.

Limitations

• Framework is conceptual and not empirically validated on real democratic AI processes.
• Explicitly agnostic about which decisions should be democratic and about which theory of democracy applies.
• Dimension assessments are context-sensitive and not operationalized into quantitative metrics.

Experimental setup

Builds on: Collective Constitutional AI Constitutional AI RLHF self-driving autonomy levels model cards

Closest papers by meaning

• A Vision for Reinventing Credible Elections with Artificial Intelligence (AAAI 2025 · similarity 0.85)
• Position: The Right to AI (ICML 2025 · similarity 0.82)
• Position: Humanity Faces Existential Risk from Gradual Disempowerment (ICML 2025 · similarity 0.81)
• Position: Social Environment Design Should be Further Developed for AI-based Policy-Making (ICML 2024 · similarity 0.80)
• Towards a More Burkean Approach to Computational Social Choice (AAAI 2024 · similarity 0.80)
• Political Neutrality as Balanced Approval: A Large-Scale Human Evaluation of AI Responses (NeurIPS 2026 · similarity 0.80)