Research initiative developing methods for democratic input on AI development — working to give affected communities a formal voice in how AI systems are designed and deployed.
What It Is
The Collective Intelligence Project (CIP) develops and tests mechanisms for incorporating public input into AI development decisions. Founded in 2022 by researchers with backgrounds at OpenAI, Anthropic, and academia, CIP bridges the gap between AI labs and democratic processes.
The core premise: AI systems will reshape society, but the public has no formal role in shaping them. Corporate executives and small teams of engineers make decisions that affect billions. CIP argues this isn't just a legitimacy problem — it's a quality problem. Systems built without diverse input fail in predictable ways.
CIP's flagship work centers on "alignment assemblies" — structured processes that bring together representative groups to deliberate on AI development questions. These aren't focus groups or user surveys. They're designed to produce considered public judgment on complex technical and ethical trade-offs, drawing on traditions of deliberative democracy (citizens' assemblies, deliberative polling) adapted for AI governance.
Sustainability Model
CIP operates as a nonprofit research institute, funded primarily by foundations. This positioning matters: foundation funding allows independence from both government agendas and corporate interests. The trade-off is scale — CIP can prototype and test governance mechanisms, but scaling them requires institutional adoption.
The model resembles academic research institutes but with an explicit focus on translation — moving from "interesting finding" to "usable governance tool."
Why It Matters for Redistribution
AI governance currently happens in corporate boardrooms and engineering meetings. The people most affected by AI systems — workers subject to automated management, communities surveilled by predictive policing, patients diagnosed by clinical algorithms — have no seat at the table. CIP works to change that.
Who benefits? Publics gain mechanisms for meaningful input on AI development. AI labs get better signal on what society actually wants (not just what Twitter complains about). Policymakers get tested models for participatory governance.
Who pays? Foundation grants fund the work. AI labs invest time and resources in running assemblies. The opportunity cost is efficiency — participatory processes are slower than executive decisions.
What changes? CIP's most significant contribution may be demonstrating that participatory AI governance is possible, not just theoretical. The Collective Constitutional AI project with Anthropic showed that public assemblies could generate usable input for model training — a concrete proof point that shifts what's considered feasible.
This creates a precedent. If Anthropic can incorporate public input into constitutional AI, the question becomes why other labs don't. The existence of working examples makes "we can't do participatory governance" a harder position to defend.
CIP's work connects to broader redistribution currents. When The Frontier Lab Oligopoly makes AI development decisions behind closed doors, CIP provides a mechanism for opening them. The project takes seriously what Langdon Winner argued in "Do Artifacts Have Politics?" (1980) — that technical design choices embed political values — and asks: what if those values were chosen democratically rather than corporately? In the landscape of Democratic AI Governance, CIP represents the experimental wing, testing whether participation can be more than rhetoric.
Current Status
Active and expanding. CIP has completed several alignment assembly pilots, including a large-scale collaboration with Anthropic on Collective Constitutional AI (2023). That project convened a representative sample of Americans to deliberate on principles for AI assistant behavior — principles that were then incorporated into Claude's training.
Ongoing work includes experiments with international assemblies, methods for ongoing (not just one-shot) public input, and research on how to make AI systems more legible to non-experts participating in governance.
The challenge is institutionalization. Pilots demonstrate possibility; the question is whether participatory processes become standard practice or remain occasional experiments.
How to Get Involved
- Read the research: CIP Publications — detailed reports on alignment assemblies and participatory methods
- Participate: CIP recruits participants for assemblies — sign up for updates to be notified of opportunities
- For AI labs: CIP works with organizations to design and run participatory processes
- Follow: Website and Newsletter for updates on new projects and findings
Related
- AI Now Institute
- DAIR Institute
- Democratic AI Governance
- AI Cooperatives
- The Frontier Lab Oligopoly
- Public AI
Sources
- Collective Intelligence Project. "About CIP." Website
- Anthropic. "Collective Constitutional AI: Aligning a Language Model with Public Input." 2023. Blog
- Huang, Saffron et al. "Collective Constitutional AI: Aligning a Language Model with Public Input." Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency. Paper
- Ovadya, Aviv. "Reimagining the Foundations of AI Governance." CIP Working Paper, 2023.