杏吧原创

IMAGINE being up against a poker player who can calculate the exact odds of a hand being a winner, play it with a straight face, and if necessary bluff with the best of them. Such a player exists, but you won鈥檛 find him wearing a Stetson or hiding behind a pair of dark glasses. This player lurks within a computer, created by a pair of academics who have succeeded in making a software agent that can bluff just like a human player can.

Poker-playing computer agents or 鈥渂ots鈥� are nothing new in themselves. Indeed, many online poker players, especially those who habitually lose, believe that poker sites are riddled with such bots 鈥� virtual players that can outwit even the best humans. 鈥淚 hear this conspiracy theory about poker bots all the time,鈥� says Cl茅ment Sire, a physicist and avid poker player at the CNRS, the French national research organisation, in Toulouse 鈥� but he doesn鈥檛 believe they are that good. 鈥淕iven the current state of poker bots, if you are losing to them you should be ashamed,鈥� he says.

That鈥檚 because the bots now playing poker can鈥檛 bluff convincingly. 鈥淐omputers are programmed to perform the best strategy, but bluffing is based on unexpected, illogical actions,鈥� says Evan Hurwitz, a computer scientist at the University of the Witwatersrand in South Africa.

Now Hurwitz and Tshilidzi Marwala, also at Witwatersrand, have developed a virtual player that has taught itself to bluff at a card game called lerpa. Their artificial intelligence bot, named Aiden, is based on a neural network algorithm that usually forecasts stock market fluctuations.

Crucially, Aiden was not pre-programmed with the rules of lerpa. Instead, Hurwitz and Marwala allowed Aiden to play against three 鈥渄umb鈥� virtual players that made choices entirely at random. Aiden was dealt his cards and told which of these could legally be played for each hand. At first he was almost too smart for the task. For the first 40 hands he wouldn鈥檛 play, then he tried one hand and lost. This proved so much of a setback that he refused to play again.

Hurwitz then changed tactics, giving Aiden no choice but to play the first 200 hands. Aiden then began to infer the rules of lerpa by treating his cards, his opponents鈥� actions and his own win-lose history as parameters to learn from. At this stage, though, he still wouldn鈥檛 bluff.

Then the researchers decided to play Aiden against three other similarly trained bots to see what would happen. 鈥淭hey began to develop their own personalities 鈥� either aggressive or conservative 鈥� depending on their past successes,鈥� Hurwitz says. After a streak of being dealt bad hands and consistently folding, one of the more aggressive players, Randy, suddenly changed tactics and began to play even when he had poor cards 鈥� he began to bluff. Aiden, a more cautious player, responded by tending to fold even when he held a relatively strong hand ().

鈥淩andy suddenly changed tactics and played poor cards. He began to bluff鈥�

鈥淭his demonstrates that computers can learn this peculiarly human behaviour,鈥� says Philippe de Wilde, a computer scientist at Heriot-Watt University in Edinburgh, UK. 鈥淭hey generate the strategy from play, which is a very human way of learning.鈥�