Games: Checkers, Chess, and Go

The chapter starts off with a discussion of knowledge in an AI program. Pieces of knowledge, or heuristics, are represented as numbers and then combined in a formula. This formula is usually referred to as an evaluation function. Given multiple options, the program will typically choose the option that yields the highest formula score. A detailed example illustrates how an AI program processes knowledge to end up with a decision. Heuristics tend to capture simple human concepts; it can get extremely difficult trying to numerically express more involved concepts. There is a discussion of the challenges of acquiring, representing, and using knowledge. Hence, AI turns to alternatives. Three case studies are shown to illustrate how an AI program works. The first is chess, where specially designed computer hardware allows Deep Blue to analyze 200 million chess positions per second, compared to a human’s two. The second is checkers, where Chinook precomputes the win, loss, draw value for trillions of positions, dwarfing anything that the human brain could memorize. Finally, for the game of Go, AlphaGo plays against itself millions of times, each time slightly modifying itself to avoid repeating mistakes. No human could possibly commit to play and learn from so many games. AI takes advantage of the computer's incredible computational capabilities to help create the illusion of intelligence.