The State Dinner of Doom
" The danger is not data centers, but the belief in data centers " W hen most people hear “AI is dangerous,” the image that pops into their head was storyboarded in 1984 by James Cameron. Skynet becomes self-aware, decides humanity is a threat, and launches the nukes. Cue Arnold Schwarzenegger in biker leathers, cue Sarah Connor learning to do pull-ups, cue thirty years of sequels. If you are as old as I am, your envisioning probably drifts back to Isaac Asimov’s I, Robot in 1950. As a model of how AI might actually go wrong, these stories are almost useless — and this matters, because it has quietly become the default mental furniture most people use when they try to think about AI risk. Terminator or I, Robot scenarios require a machine that wants things in the way a person wants things: dominance, survival, revenge. It requires malice, or at least a very human sense of self-preservation. The way Anthropic speaks of Claude as if it were a person feeds this sense. R eal m...
