For the first time, a computer algorithm is much better at playing Texas Hold’em Poker than any human being. The development of this algorithm, which has been a complete success so far, is not going to be limited to the field of games of chance, as it can also help develop strategies that maximize performance in a commercial negotiation or minimize the risk of attacks terrorists, even if an adversary knows the defense strategy.
The programming of this algorithm, and its result in the tests that have been carried out, has been a big step towards understanding games that are closer to real-world problems. In fact, it is based on the same technology that created the supercomputer that defeated Russian chess champion Garry Kasparov in 1997.
Every game is perfect
The algorithm is the first to address a game based on imperfect information that is constantly changing, and in which the participants do not have complete knowledge of all the previous moves. Just as poker players must act without knowing what cards their opponent is holding, researchers want algorithms that can make effective decisions based on robust but incomplete data sets.
For decades, scientists have tried to create algorithms capable of defeating humans by combining the computational prowess of computation with the decision-making strategies of game theory. This theory belongs to the branch of mathematics that calculates optimal strategies in competitive matches. In 2007, a group of researchers created an unbeatable algorithm for the game of checkers, simulating the actions of players who made the perfect move every turn. It was an impressive achievement, but the algorithm had a relatively simple task: For each turn, it would review the positions of all the pieces, evaluate its options, and then choose the best option.
Everything is based on the data
In any game of chance, humans are very limited when it comes to storing the data of each of the plays or knowing what cards are left in the deck taking into account those that have been dealt. Another problem is that the players, in this case of poker, are unpredictable. In checkers, the best move in a given situation is the same for all games. But in poker it’s wise to bluff from time to time to throw your opponent off the scent. And this is where the human has some advantage, as most algorithms can’t deal with those kinds of in-game uncertainties.
By generating an optimal set of probabilities for each possible scenario in a hand, the researchers ensured that the algorithm would never lose money in the long run despite losing a few hands. Beyond poker, scientists can apply this unprecedented blend of artificial intelligence and game theory to important real-world problems.
REFERENCES:
Science (article) / IBM Thomas J. Watson Research Center in New York / University of Alberta / Computer Poker Research Group.
