OpenAI and Others Warn AI R&D Automation Could Trigger 'Intelligence Explosion'

On September 28, researchers from OpenAI, Anthropic, Meta and Microsoft released a joint report urging policymakers to pay closer attention to the automation of AI research and development. The report says AI systems can already complete most of the coding work required to develop AI, and further automation of the R&D process could trigger an "intelligence explosion," compressing AI capability gains from years to months or even less.

The report suggests AI systems could automate most or even all AI R&D work within the next few years. While such an intelligence explosion could accelerate the benefits of AI, it also carries risks such as AI capabilities growing faster than society can adapt, and humans losing control over superhuman AI systems. The researchers said these risks need early examination.

They recommend that policymakers improve their understanding of AI R&D automation and study ways to guide and constrain its development to keep technological progress under control.