OpenAI Puts Up $1M for Experiments in AI Democracy

OpenAI is trying to answer one of the thorniest questions in technology: who gets to decide what an AI system should and shouldn’t say? The company’s nonprofit arm is putting $1 million on the table, spread across ten $100,000 grants, to fund teams building experimental democratic processes that could set behavioral ground rules for artificial intelligence. The idea isn’t to immediately dictate binding policy for models like ChatGPT, but to prototype systems that balance diverse public interests on hot-button issues.

Consider a scenario where users ask an AI about a polarizing public figure. Should the model issue a neutral description, a sharp critique, or reflect a specific cultural consensus? OpenAI argues no single company, let alone a lone individual, should dictate that answer. “The governance of the most powerful systems, as well as decisions regarding their deployment, must have strong public oversight,” the organization stated in its announcement. The program explicitly calls for proof-of-concepts that go beyond basic polling, citing inspiration from deliberative platforms like Pol.is, Twitter Community Notes, and the Collective Intelligence Project, with whom OpenAI is already partnering on alignment assemblies.

The real challenge isn’t just gathering opinions but preventing the process from being hijacked. The announcement flags risks like participation washing, manipulation by special interests, and the drowning out of minority viewpoints. Teams applying for the cash will need to show they are keenly aware of these failure modes. They’re being asked to design a process that tackles at least one specific policy question from a list provided by OpenAI, running a transparent decision-making engine that involves a broadly representative group.

There’s a healthy dose of skepticism baked into the timeline. The June 24, 2023 deadline means these initial experiments won’t be binding on OpenAI’s product decisions in the short term. They are explicitly framed as stepping stones toward something more ambitious and global. While governments draft the laws, OpenAI is betting that a parallel track of public deliberation can handle the nuanced, adaptive guidelines that legal code can’t reach. Whether $100,000 per team is enough to solve a problem that has vexed political philosophers for centuries is an open question, but the results will be released freely, potentially giving the entire industry a new blueprint for public oversight.