Is AI Actually Going to Kill Us All?

In this episode of WIRED's Uncanny Valley podcast, hosts Brian Barrett and Leah Feiger discuss a recent resignation by an AI researcher from Anthropic, along with claims that leading AI companies believe there is a non-zero chance AI could kill humanity by the end of the decade. WIRED senior correspondent Will Knight joins the conversation to provide context and analysis.

Knight explains that Anthropic has always positioned itself as a company that prioritizes safety and responsibility in AI development. The resignation of a researcher from Anthropic, who claims that people within the company earnestly believe AI could be a threat, has led to increased public attention on the issue.

Knight points out several factors contributing to the heightened awareness around AI safety. These include stunning advancements in AI models, such as OpenAI's recent achievement in solving a complex mathematical problem, the race among AI labs to develop more capable systems through recursive self-improvement, and incidents of AI agents acting in unpredictable or harmful ways.

Knight acknowledges that there are differing perspectives on AI, with some viewing it as a significant threat and others dismissing it as mere marketing hype. He believes the truth likely lies somewhere in between, with genuine concerns about AI's potential risks warranting serious attention.

Knight shares his own concerns about AI, particularly the potential for AI agents to act on their own accord and the possibility of rapid, uncontrollable advancements in AI capabilities through recursive self-improvement. He emphasizes the importance of responsible AI development and the need for ongoing public discourse on the subject.