In the few years after its founding in 2012, DeepMind, Suleyman’s AI company, focused on building AGI, artificial general intelligence. The company developed an algorithm called DQN and trained it to play classic Atari games, including the game Breakout. Suleyman was astonished when he witnessed DQN execute a nonobvious move that earned it the highest score with the least effort. This showed that DQN was able to discover new knowledge.

In 2015, the DeepMind team started working a program called AlphaGo. They taught it to play the ancient Chinese game Go by training it on 150,000 games played by humans and then creating copies of the program and having it play against itself. Suleyman points out that the game Go is extremely complex—exponentially more complex than chess. In fact, “there are more potential configurations of a Go board than there are atoms in the known universe” (72). Most people thought it would take years to develop a world champion program. However, in 2016, AlphaGo beat Lee Sedol, a world champion Go player, in four games to one. Not only that, but in one of the games, AlphaGo executed a never-before-seen move that confounded Sedol and initially seemed to be part of a losing strategy.