The Collective Intelligence Project Proposes Democratic Governance for AI Companies

Summary

The Collective Intelligence Project argues that the governance of companies developing and deploying frontier AI models will shape whether AI concentrates power or broadly benefits society. Its note says conventional venture-backed corporate structures can drift from public-benefit missions toward financial self-interest, while social trust and accountable oversight are especially important for transformative AI. The authors define democratization across four dimensions: access to use, participation in development, distribution of benefits, and shared governance over trade-offs such as safety and innovation. They recommend giving employees, including engineers, quality-assurance staff, and support workers, formal channels to influence decisions, while recognizing that employee interests do not always match the public interest. They also propose independent stakeholder councils, public assemblies, and evaluation juries to bring wider social input and judgments about ambiguous impacts into company governance. For preserving organizational values, the note discusses nonprofit parents, public-benefit corporations, perpetual purpose trusts, and external organizations holding a golden share, but it cautions that these structures can be difficult to finance, especially before a frontier lab reaches positive cash flow. Finally, it calls for composable technical and legal infrastructure with distributed ownership and cooperative models, drawing examples from Visa, the Associated Press, Linux, Debian, Ubuntu, and Bluesky. The authors present these ideas as options for experimentation rather than a single prescribed model.