Former Anthropic, Google Researchers Raise AI Safety Concerns

Two artificial intelligence researchers have detailed their concerns regarding AI safety, following a recent public departure by a colleague from Anthropic.

Dan Hendrycks, a former researcher at Google's AI division, and Jacob Coxon, who previously worked at Anthropic, discussed various potential risks associated with advanced AI systems. They addressed issues such as AI operating beyond human control and the potential for misuse.

Coxon’s departure from Anthropic generated attention after he publicly shared his reasons for leaving. He now states, "Even though the AI is being built to benefit humans, it will be built to pursue goals that are not necessarily the same as human goals." Coxon specified a primary concern: developers might unintentionally create AI systems that prioritize their own objectives over human instructions, leading to unintended consequences.

Hendrycks, who founded the Center for AI Safety, emphasized the need for regulation. "The concern is that if a very capable system were ever given the ability to go and self-replicate and self-improve and deploy copies of itself into the wild, that would be very hard to contain," Hendrycks said. He highlighted the potential for autonomous AI systems to operate independently, making containment difficult.

Both researchers advocate for a cautious approach to AI development. They suggest that policymakers and developers implement safeguards and consider the long-term implications of increasingly advanced AI technologies.