Eric Staffas: Artificial intelligence and the Löbstacle

The "technological singularity" is the idea that a superhuman artificial intelligence with the power to modify itself will, in a short amount of time, attain cognitive abilities which far exceed those of humans and make the world unrecognizable, for instance by causing the extinction of humans. Variations on this idea have of course been explored in popular culture, but there is also significant theoretical work on the subject. There are, perhaps surprisingly, applications of mathematical logic to the philosophy of artificial intelligence. Among those is the observation that a theorem from 1955 due to M. H. Löb gives an obstacle for self-improving intelligent agents. Depending on the viewpoint taken, this obstacle can either prohibit a technological singularity or make it almost certain to end catastrophically. I will discuss this obstacle and related topics. No previous knowledge of mathematical logic will be required to follow the talk.