As AI becomes increasingly advanced, it promises many benefits but also comes with risks. How can we mitigate these risks while preserving scientific inquiry and openness? Who is responsible for anticipating the impacts of AI research, and how can they do so effectively? What changes, if any, need to be made to the peer review process? In this talk, we’ll explore these tensions and how they are playing out right now in the AI community. AI is not the first high-stakes, ‘dual-use’ field to face these questions. Taking inspiration from fields like cybersecurity and biosecurity, we’ll look at possible approaches to responsible publication, their strengths and limitations, and how they might be used in practice for AI.

Rosie Campbell leads the Safety-Critical AI program at the Partnership on AI, a multistakeholder nonprofit shaping the future of responsible AI. Her main focus is on responsible publication and deployment practices for increasingly advanced AI. Previously, she was Assistant Director of the Center for Human-Compatible AI at UC Berkeley, a Research Engineer at BBC R&D, and cofounder of Manchester Futurists. Her academic background spans physics, philosophy, and computer science. Rosie is also a productivity nerd and enjoys thinking about how to optimize systems, and how to use reason and evidence to improve the world.