Humans have now accepted that they will never be as good at chess as robots, but now even robots must accept that they will never be as good as other robots!
A new artificial intelligence platform, known as AlphaZero, can learn to play the games Go, chess and shogi, without any human intervention. Using deep neural networks, AlphaZero quickly learned how to play each game “to become the best player in history.”.
AlphaZero was revealed by DeepMind Technologies in a study published in Science on November 6. DeepMind, a British artificial intelligence subsidiary of Alphabet, Google’s parent company, has been working on Go AI for many years. In 2017, DeepMind retired former AI champion AlphaGo, but has continued to work on AI. With AlphaZero, DeepMind’s research has reached its zenith.
The program faced the world's best AIs in three board games:
- Stockfish, world chess champion AI
- elmo, winner of the 27th annual World Computer Shogi Championship in 2017
- AlphaGo Zero, DeepMind's Go AI, the strongest Go player in history
In any case, AlphaZero was only given knowledge of the basic rules of the game. Before competing with the other AIs, it played millions of games against itself, starting with random tactics and trying to win, but slowly learning which strategies worked best through a trial-and-error process called “reinforcement learning.”.
The training and learning process took nine hours for chess, 12 hours for shogi, and 13 days for Go, using 5,000 TPUs (Tensor Processing Units). A single TPU can process over 100 million photos per day in Google Photos. Once it finished training, AlphaZero competed with the rest. And crushed them!
What's unique about the study is that the algorithm was combined with a "search method" called Monte Carlo tree search (MCTS). This is how Go AI programs determine what move to make next. The DeepMind team used the same system for chess and shogi, showing for the first time that it could be adapted to other complex games.
Perhaps more interesting to chess players is the fact that AlphaZero, without human hands, can construct its own knowledge, implement strategies and innovative ideas that have never been seen before. Its aggressive style and extremely dynamic play surprised chess Grandmaster Matthew Sadler, who spoke about it on the DeepMind blog.
These unique strategies and abilities make AlphaZero a great teaching tool for chess players.
The dominance of artificial intelligence over humans is quite common in the world of gaming, with robots beating us in board games, complex video games like Dota 2, and of course Go.
Does this mean that AI is poised to beat us at literally every competitive game ever invented? Fortunately not. Although the three games DeepMind uses are extremely complex, they do provide some advantages for AI, as they involve two players and all the information necessary for the next move is always visible.
So while they've certainly taken on the role of champion in games like chess, robots will probably never beat us at Texas Hold 'Em.
