The AlphaZero AI program learns on its own to beat the best machines in chess, go and shogi.
AlphaZero’s technical prowess has two relevant components in the field of artificial intelligence. The program of DeepMind, a subsidiary of Alphabet, has achieved master multiple games at once. But he has also managed to learn to do it on his own.
The overcoming of humans has been the constant that has moved the invention and development of machines. Before, physical strength was sought, the repetition of tasks. But now intelligence is the main challenge when designing a machine. And a good way to test these cognitive faculties is by pitting programs against complex board games. Here you need combine logic, intuition and strategy to beat the opponent.
It’s been more than 20 years since Deep Blue beat Gary Kasparov in an unprecedented milestone. And only in 2016 was an artificial intelligence developed that won the best player in the go world. This has been a new milestone for the creative component of vesta disciplina. The IBM machine played chess by means of a gross calculation of possibilities. However, this method did not work for go, which requires a high level of intuition.
DeepMind managed to overcome this barrier. And the step he has now confirmed is a new leap forward. The company had already shown that its software was able to beat go, chess and shogi (a type of Japanese chess). But now the merit of the AlphaZero platform has been recognized by the journal Science.
A champion for all games
AlphaZero has beaten the main machines prepared to be the best in each game individually. Program Stockfish and Elmo, both considered the most advanced in chess and Shogi, respectively, have been the main rivals to beat. In go, AlphaZero has faced his predecessor Alphagowho will win the best human player.
Specialists say that AlphaZero has not only learned to play these three disciplines. (Their opponents, on the other hand, have been trained specifically for a single game.) Platform has shown a genuine style of play. It wasn’t just to avoid making mistakes and raise the level of the game. The program would have demonstrated what in a person we would call intuition.
DeepMind’s artificial intelligence i didn’t need to evaluate such a high number of possibilities before you move tile. To make a decision, it was enough to make a massive calculation – far superior in options to that of a human-but considerably smaller than that of his rivals. Its system favours a more in-depth analysis over an exhaustive review of the situation.
The result is that the neural networks of which AlphaZero is composed have behaved as a human being in a state of grace. Even more, of course, but with the computing power that these systems allow. The fact that he has learned all three games on his own is another notable aspect that brings us closer to a –still very distant– general artificial intelligence.
Images: kimfaires, AlphaZero
