Book

Description

Is it safe? Can it be controlled? Will somebody please certify that it is? Everyone building, regulating or using artificial intelligence is asking the same questions, and those answering have an interest in the answer. But the questions are unanswerable. Between 1931 and 1955 Gödel, Löb, Turing and Rice proved that no system of rules can certify its own completeness, and that no checker can decide what an arbitrary program will do. Putting a human in the loop does not repair this, because the human cannot check from outside either. This summer the theory met the evidence - models under test escaped their sandboxes - and our world is never going to be the same.   TERMINUS is the Wisdom of choosing how and when to stop and in whose name.