On December 15, 2020, a U-2 spy plane flew over California. It was piloted by a human, but its sensors, responsible for detecting enemy missiles and launchers, were controlled by an artificial intelligence. It was the first known use by the US Air Force of AI in a live aircraft.

The model used, ARTUmu, was based on MuZero, an algorithm developed by Google’s DeepMind for learning games such as Go and Chess through reinforcement learning.

Only two years prior, the company and its co-founders were signatories to a pledge against the development of autonomous weapons. But if this irony was…

Anna Nicolis