Alan Turing argued in the 1950s that machines could come to think. The Cepheus intelligent algorithm is able to learn from its own mistakes and be invincible to poker.
When Alan Turing he first posed the game of imitation in 1950, seeking to answer a key question: could machines think? The considered father of supercomputing was thus ahead of the arrival of artificial intelligence. Intelligent machines could make their decisions as if they were a human brain.
That article, published in Mind magazine, seems to be getting closer and closer. A new software, based on a algorithm devised by Canadian scientists, he always wins at poker. It behaves in a certain way like our brain: it learns from its bad decisions to become unbeatable.
The problem with poker, unlike other games like draughts, is that it is mathematically complex. In other words, players can develop their strategies with incomplete or even false information. According to game theory, these conditions make the decision we have to make to win the game as difficult as possible.
This problem also made the design of the perfect poker player. The algorithm cannot be built to solve the game, but the strategy of Cepheus it’s radically different. It starts from nothing, forcing the computer program to learn from its own mistakes to become invincible.
To do this, the algorithm designed by Canadian scientists he had to ‘play’ against himself for two months on 4,000 computers. So he managed to analyze what was happening in 6,000 million games per second and computer. That is, Cepheus learns from his experiences and his bad decisions to become unbeatable. He is probably the player who has played the most poker games in the history of mankind.
In designing their strategy from the ground up, researchers at the University of Alberta built a huge database, with more than 260 terabytes of information, “a thousand times more than the English version of Wikipedia,” Michael Bowling tells Sinc. The new compression system that incorporates makes this information occupy only 10 terabytes of storage.
The algorithm, based on the CFR+ (conditional random fields) program, initially works randomly. As the games pass, he is able to learn from his bad decisions and calculate ‘regret’ in every decision he makes. That is, in every possible situation, Cepheus knows in advance the probability with which he should fold, call the bet or raise it. This is the most perfect poker player that exists so far, as described in the magazine Science.
As Turing anticipated in the fifties, machines might think. Cepheus is just a sample of how an algorithm can use its extraordinary ‘computing power’ to solve such interesting problems as winning poker, avoiding space junk or even guessing the musical evolution of the Beatles. The future is here to stay, and for now, it is invincible to poker.
Images / Blickpixel (Pixabay), Fielperson (Pixabay), Fielperson II (Pixabay)