AI Risk Prioritization: Public Input for OpenAI

In June 2023, The Collective Intelligence Project ran an Alignment Assembly with OpenAI as a committed audience, supported by public-participation expertise from The GovLab and Citizens Foundation's All Our Ideas platform. The practical question was direct: if AI labs are deciding how to measure and mitigate risks from large language models, what does the public want them to prioritize?

The engagement used All Our Ideas as the wiki-survey infrastructure. Participants compared pairs of statements completing the prompt: "When it comes to making AI safe for the public, I want to make sure..." They could also add their own concerns, allowing the process to combine structured ranking with open public input.

From public concern to ranked priorities

Over two weeks, 1,000 demographically representative Americans took part. After moderation and deduplication, the voting pool included 136 statements, including 44 added by participants. The top five statements were all user-submitted, showing why open-ended input matters: the public did not simply validate an expert agenda; it expanded it.

The Collective Intelligence Project's report, Participatory AI Risk Prioritization, found four clear signals. Participants wanted regulation rather than a "Wild West" approach. They cared more about good governance than any single technical risk. They were especially concerned about overreliance on systems people do not fully understand. And they worried about misuse of large language models, including misinformation, hate speech, and enabling violence.

Six participants then joined a follow-up roundtable with OpenAI to discuss the results in more depth. OpenAI's response in the report says it participated by sending the poll to a subset of ChatGPT users, attending the roundtable, and considering the findings as part of its broader work on public input into AI deployment.

Why this mattered

The recommendations were concrete: monitor post-deployment effects, create evaluations for overreliance, show that acceptable-use policies are enforced, share data on real-world use cases, invest in literacy and accessibility, and create stronger forums for public input into AI.

For Citizens Foundation, this was an early example of All Our Ideas being used in high-stakes AI governance. The process did not ask a small group of experts to guess what the public might care about. It gave a representative public a fast way to compare concerns, add missing ones, and produce a defensible ranking that AI labs and governance bodies could use.

That is the role of All Our Ideas at its best: turning open public judgment into a clear signal decision-makers can act on.