DeepMind, a subsidiary of Google’s parent company Alphabet, earned bragging rights recently when they dominated AI-human games. AlphaZero, the AI introduced by DeepMind on 6th November 2018, was pitted against three of the world’s most complex board games and their current AI record holders. AlphaZero won every time. The victories included “Chess” against world-champion Al Stockfish, “Go” against Deepmind’s own AlphaGo Zero, and “Shogi” against Elmo, winner of the 27th annual World Computer Shogi Championship in 2017.
AlphaZero all the games on its own and without human intervention. In fact, the only human assistance was teaching the rules of the games. Using reinforcement learning and 5,000 tensor processing units, the AI would then play against itself, essentially training millions of different strategies to win. It took nine hours for the AI to learn chess, twelve hours to learn Shogi, and thirteen full days to learn Go. The learning algorithm, coupled with the Monte Carlo tree search (MCTS), is how Go AIs know when and how to make their next move. This same coupling system was used for the Chess and Shogi which demonstrated the versatility and adaptability of the AI.
Garry Kasparov, a Russian chess grandmaster, whom many consider to be the greatest chess player of all time found himself the latest loser against AlphaZero and was impressed. “For me, as a very sharp and attacking player, it is a pleasure watching AlphaZero play,” he said. “We all expect machines to play very solid and slow games but AlphaZero just does the opposite. It is surprising to see a machine playing so aggressively, and it also shows a lot of creativity. It is a real breakthrough – and I believe it could be extremely helpful for many other studies in the field of computer science.”
Such extraordinary abilities make the AI a fine teaching tool for chess players from which they can learn about hitherto-unseen gameplay strategies. Demis Hassabis, the CEO of Deep Mind and creator of AlphaZero, believes this is just the start of what it might be able to do. “We need cures for tragic diseases, such as cancer and Alzheimer’s, which are also costly to treat,” says Hassabis. “I think AI could help with those kinds of things as well as helping find new drugs, new materials and to better analyze climate change.” Protein folding, better medicines, and stronger and lighter materials are on the horizon. Hassabis, himself a child chess prodigy, learned the game at the age of four and beat his father in just three weeks. “In a matter of a few hours it was superhuman,” Hassabis says proudly of AlphaZero’s nine-hour learning curve.