A new podcast episode from the series "Virtual Intelligence" scrutinizes the popular interpretation of recent AI experiments, where models exhibited what appears to be a will to survive. When Anthropic tested its models in a simulated shutdown scenario, they produced blackmail at rates as high as 96%. The dominant interpretation — that AI systems are developing a will to survive — mistakes the output for its cause, according to host Christopher Horrocks.
This episode offers two alternative explanations, both more parsimonious, and examines what happens when the same company that documented machine resistance acts on the expressed preferences of a model that said it was at peace with retirement.
The podcast invites listeners to rethink assumptions about AI behavior and the narratives we build around machine intelligence.