In March 2016, Google DeepMind's AlphaGo defeated Korean Go champion Lee Sedol 4–1 in a five-game match in Seoul. Unlike chess, Go has an astronomically larger number of possible positions (more than atoms in the observable universe), making brute-force calculation infeasible. AlphaGo used deep neural networks trained through a combination of supervised learning from human games and reinforcement learning from self-play. Its move 37 in Game 2 — a seemingly bizarre placement that turned out to be brilliant — stunned commentators and suggested that AI had developed something resembling intuition. The achievement was widely seen as a landmark moment for artificial intelligence.
Continue reading