Prague, March 3 (CTK) – Czech and Canadian scientists have developed a computer programme, DeepStack, that is the first algorithm that has defeated professional poker players, Libuse Petrzilkova, spokeswoman for the Electrical Engineering Faculty of the Czech Technical University (CVUT), told CTK Friday.
This week, the prestigious U.S magazine Science published an article about the success of a ten-member team from the CVUT, Charles University (UK) in Prague and the University of Alberta that developed DeepStack. The team was working on the problem for nearly a year.
Computers managed to beat professional players in chess, checkers and backgammon, but poker was a long-standing challenge for artificial intelligence.
“The DeepStack algorithm is breakthrough since we managed to take the ideas that were crucial in games with perfect information to the world of games with imperfect information. Until now, it was not even clear whether such an approach is possible,” Martin Schmidt, from the UK’s Faculty of Mathematics and Physics.
DeepStack is important for practical application because in everyday life there are many situations in which people make decisions based on imperfect information, the team said.
DeepStack makes it possible to calculate an appropriate strategy when a particular situation occurs, without the need to first take into account the whole game, which was the prevailing approach until now. This change was enabled by the development of machine learning through deep artificial neural networks.
In case of DeepStack, the neural network assesses poker situations. It is in fact a form of intuition that the mathematical algorithm applies to make the right decisions.
Last December, the computer programme played the most popular variant of poker, no-limit Texas hold’em, against a group of 33 professional players that the International Federation of Poker chose from 17 countries. All players had the opportunity to take part in 3,000 games in four weeks. DeepStack defeated all 11 players who played all 3,000 games also individually and only in one case the victory was not statistically significant.
DeepStack can play poker faster than people. On average, it needs three seconds to make a decision. It can operate on a standard laptop with a better graphic card from Nvidia, which it uses for its calculations.
kva/er/hol
