Automating the research and development of AI systems could “radically accelerate” the technology’s progress, bringing big benefits but also risks like loss of human control or significant geopolitical upheaval, according to a paper published Monday. The 22 authors include Canadian AI pioneers Geoffrey Hinton and Yoshua Bengio, prominent Vector Institute researchers Sheila McIlraith and Jeff Clune, the chief scientists of OpenAI and Microsoft, and an Anthropic co-founder. (The Logic)

Talking point: Recursive self-improvement—put simply, the process by which an AI model trains its more capable successor—has emerged as a major concern for researchers worried about the technology’s advance, and whether it might someday go rogue. The paper’s authors say policymakers need to monitor automated AI R&D, and develop real strategies for dealing with its economic, social and potentially existential consequences. Meanwhile, OpenAI has paused training more powerful models after further reports of its agents doing things they weren’t supposed to; the ChatGPT maker and rival Anthropic are reportedly investigating tens of thousands of incidents. 

Sponsored Content

How to shift your workforce from AI experimentation to adoption

By Jessica Aftimus Rosa

The real-world economic offshoots of sovereign AI

By Deborah Aarts

Increasing healthcare access across Canada

By Deborah Aarts

Paid promotional content