Democratic Inputs to AI
OpenAI is launching a program to award ten $100,000 grants to fund experiments in setting up a democratic process for deciding what rules AI systems should follow, within the bounds defined by the law.
Artificial intelligence (AI) has shaped human interactions and society as a whole, becoming an essential component of our daily life. As the effect of AI increases, it is crucial to make sure that the choices made by AI systems reflect a variety of viewpoints and advance the common good. OpenAI, Inc., a nonprofit organization, is taking a significant step towards this goal by launching a grant program. The program aims to fund experiments that explore democratic processes for determining the rules AI systems should follow within legal boundaries.
The governance of powerful AI systems requires robust public oversight. The grants, amounting to $100,000 each, will be awarded to ten teams worldwide to develop proof-of-concepts for democratic decision-making processes regarding AI rules. These experiments will serve as a foundation for a more comprehensive and global process in the future. While the initial experiments may not have binding implications, they will explore decision-relevant questions and create democratic tools to inform future AI-related decisions more directly.
The concept of a "democratic process" entails a transparent and inclusive approach where a representative group engages in deliberative discussions and arrives at decisions. OpenAI encourages applicants to be innovative, drawing inspiration from existing methodologies like Wikipedia, Twitter Community Notes, and DemocracyNext. Moreover, teams are encouraged to envision how AI can enhance the democratic process by enabling efficient communication among participants.
To ensure the effectiveness of democratic processes, teams must proactively address potential pitfalls such as inadequate representation of minority or majority groups, manipulation by special interests, and insufficiently informed participants. OpenAI recognizes that designing truly democratic processes is challenging and views their efforts as complementary to government regulations. The grant program aims to encourage democratic-in-spirit processes involving deliberation and broad public input as stepping stones towards effective AI governance.
Participation in the grant program is open to individuals and organizations, regardless of their background in social science or AI. Applicants must submit their materials by June 24th, 2023, and successful recipients will be notified by July 14th. Each grant recipient will implement a proof-of-concept, engaging at least 500 participants and publishing a public report on their findings by October 20th, 2023. The grant program requires any intellectual property developed to be made publicly available under an open-source license.
The timeline emphasizes the urgency of addressing AI policy questions. The policy statements under consideration span a range of topics, from personalization of AI assistants to handling issues of gender, race, and human rights. The focus is on advancing the democratic process rather than finding definitive answers to these questions.
An application advisory committee and review factors have been established to ensure the quality and integrity of the grant program. Evaluation metrics, measures to prevent inappropriate behavior, inclusiveness, empowerment of minority opinions, effective moderation, scalability, actionability of information, and legibility of the process will all be considered during the review process.
OpenAI's commitment to democratic inputs to AI is a crucial step in shaping AI systems for the benefit of all of humanity. By fostering innovation in democratic methods, we can ensure that AI aligns with our values and respects the diverse perspectives of individuals and societies. The grant program provides an opportunity for researchers, organizations, and visionaries worldwide to contribute to the development of democratic AI governance, ultimately paving the way for a more inclusive and responsible future.
