How Innovation Works: And Why It Flourishes in Freedom

by Matt Ridley

Go is too complex a game for the brute-force techniques that helped Deep Blue, and the crucial ingredient of AlphaGo is its ability to learn. It was not taught the rules of Go, but intuited them from examples of games using neural networks (the latest version of the program does not consult human games at all). Thus the human beings who programmed AlphaGo have no idea why it chose the moves it did. Move 37 in Game 2 was described by an expert as ‘creative’ and ‘unique’, because it broke all the normal rules and seemed to be stupid. Lee Sedol took an unusually long time to respond, and although his counter was also unusually clever, he eventually lost the game following a series of equally brilliant moves by AlphaGo at moves 151, 157 and 159.