Inclusive AI Development: Americans Draft an AI Constitution
In an effort to democratize the development of AI tools, Anthropic, the creator of Claude AI, partnered with the non-profit Collective Intelligence Project (CIP) to explore a new approach. They surveyed 1,000 Americans to draft an AI constitution, aiming to involve more people in shaping the future of AI.
The Mission of the Collective Intelligence Project
Founded by Divya Siddarth and Saffron Huang, the CIP promotes collective intelligence for collective progress. They teamed up with Anthropic for the Collective Constitutional AI initiative to develop an AI model that incorporates outside inputs and examines how democratic processes can influence AI development.
A Collaborative Process
The project asked 1,000 members of the American public to help Anthropic choose rules for its AI chatbot. Participants could either select from a list of rules or create their own. The sample aimed to represent diverse demographics in terms of age, gender, income, and location.
Key Findings from the Survey
The participants contributed 1,117 statements and cast over 38,000 votes on various questions related to AI’s behavior. While there were differing opinions, there was generally a high degree of consensus on most statements. Notably, the survey revealed a desire for the AI model to avoid discrimination based on race, sexual preference, and disability.
Expanding Beyond Current Limitations
Siddarth acknowledged that future experiments should reach beyond tech-savvy and affluent Americans to include a more diverse range of voices. The goal is to ensure that values from different communities are adequately represented in shaping AI development.
Collaboration with OpenAI and Future Prospects
The CIP has also collaborated with OpenAI on other democratic AI projects. While the AI Constitution project exclusively used Anthropic’s Claude AI models, Siddarth highlighted the importance of comparing models trained on public input with those developed by researchers.
Hot Take: The Future of Inclusive AI Development
Despite the limitations of the Constitutional AI experiment, Siddarth remains optimistic about the Collective Intelligence Project’s future. The aim is to continue developing processes that involve civil society groups, policymakers, and various companies to ensure more inclusive and responsible AI development.
Sources:
Collective Intelligence Project,
Collective Intelligence Project – About,
Collective Constitutional AI Initiative
By
By
By
By