OpenAI explores collective decisions on AI, like Wikipedia entries

OpenAI, the research organization behind the popular chatbot ChatGPT, is experimenting with ways to gather broad input on decisions impacting its artificial intelligence, its president Greg Brockman said on Monday. He also suggested that governments around the world should coordinate to ensure AI is developed safely and responsibly.

At AI Forward, an event in San Francisco hosted by Goldman Sachs Group Inc. and SV Angel, Brockman discussed the challenges and opportunities of advancing the state-of-the-art in AI, especially generative AI that can create realistic text, images, and audio from prompts.

One idea he previewed is inspired by the model of Wikipedia, which he said requires people with diverse views to coalesce and agree on the encyclopedia’s entries. “We’re not just sitting in Silicon Valley thinking we can write these rules for everyone,” he said of AI policy. “We’re starting to think about democratic decision-making."

Another idea that Brockman discussed, on which OpenAI elaborated in a blog post Monday, is that governments around the world should collaborate to ensure AI is developed safely. He said a body like the International Atomic Energy Agency (IAEA) could place restrictions on deployment, vet compliance with safety standards and track usage of computing power. He also proposed a global agreement to limit the annual growth of frontier AI capabilities, or a joint global project that major governments could participate in.