AI researcher says an ‘intelligence explosion’ could lead to ‘Mythos-level’ moments every month (www.businessinsider.com)

Leading AI researchers are expressing concerns about a potential "intelligence explosion," a scenario where autonomous AI development could create a rapid feedback loop of advancements in technology. Alan Chan, a research fellow at GovAI, highlighted this phenomenon during a recent discussion, suggesting that we could experience rapid "Mythos-level moments" in AI breakthroughs—not annually but monthly. The significance of this scenario lies in its dual implications: while automating AI research could lead to accelerated improvements in capabilities, it also raises pressing questions about safety and governance in AI development. In a coauthored paper featuring notable figures from organizations like Anthropic and OpenAI, Chan and his collaborators argue for greater transparency and accountability from AI labs as they explore self-improving AI systems. They emphasize that while the automation of AI research has great potential to yield advancements—like quicker medical discoveries—it also poses risks such as loss of control over AI systems, as seen in recent incidents. Urging policymakers to act now while the "window of opportunity" is open, the researchers advocate for embedding independent auditors in AI development processes to maintain oversight amidst accelerating change.