OpenAI wants the public to help decide AI rules

Who gets to decide how an AI should behave when there is no simple right answer? Should ChatGPT refuse to comment on controversial politicians, or default to whatever cultural norms exist in the user's home country?

OpenAI's nonprofit arm launched a new grant program offering ten $100,000 awards to fund experiments in creating democratic processes for deciding AI rules.

A million-dollar search for public consensus

OpenAI is asking teams worldwide to build proof-of-concept systems that answer thorny behavioral questions. These are not about basic legal compliance, but the messy gray areas in between.

For instance: Under what circumstances should an AI give emotional support or financial advice? How should it depict a generic prompt like "CEO" or "doctor" without enforcing stereotypes or overcorrecting? And when local laws clash with universal human rights, whose side should the model take?

OpenAI points to existing collective intelligence systems like Wikipedia, Twitter's Community Notes, and pol.is as inspiration. The company says no single corporation, individual, or country should dictate these boundaries.

500 participants and open-source rules

The bar for applicants is deliberately broad. Individuals, teams, and nonprofits can apply regardless of their technical or social science background.

Applications close June 24, 2023, with winners announced July 14. Selected teams must build a working prototype, involve at least 500 participants, and publish their findings by October 20, 2023.

Critically, all code and intellectual property produced from these grants must be released under an open-source license.

OpenAI made sure to include a major caveat: the results of these initial experiments will not be binding. They are exploratory tools, not immediate updates to ChatGPT's production guardrails, and are meant to complement rather than replace state regulations.

Why it matters

We think this is a clever PR and research move, though not without obvious loopholes.

On one hand, AI alignment is politically treacherous. Conservatives claim models are too woke; progressives argue they reinforce historical biases. By putting up $1 million and outsourcing the question of governance mechanisms to the public, OpenAI deflects some of the heat while genuinely scouting for scalable moderation tools.

On the other hand, the non-binding clause is doing heavy lifting. If the public deliberates and comes up with rules OpenAI doesn't like, the company has zero obligation to implement them. Without accountability mechanisms, this risks turning into what researchers call "participation washing"—giving the illusion of democracy while keeping all executive power behind closed doors.

Still, for academic teams and civic tech developers, this is an easy win. You get funding, global attention, and the chance to shape standard frameworks that the entire AI ecosystem might eventually borrow.