University of Alberta Scientists Reach Milestone in AI Poker Research
Dawn of the new age in poker as we know it is upon us.
Computer scientists from the University of Alberta have developed a computer program called DeepStack which is beginning to display poker skills many of professional poker players could be jealous of.
Following 20 years of thorough research on programs that try to ‘solve’ poker problems, a huge milestone has been reached most recently.
In a breakthrough which marks an important milestone for artificial intelligence (AI), the program represents an important step in mimicking aspects of human brain and ways of how it acquires expertise.
Games – and poker most specifically – provide valuable test ground for AI as they offer different aspects of programming to be evaluated and compared. The University of Alberta scientists have gone with Heads-Up No-Limit Texas Hold’em for their research as it presented a different kind of a challenge in which the program is required to deal with incomplete knowledge since cards, unlike in other games, are hidden from view.
“The essence of poker is being able to make decisions when you don’t have all of the information that you need”, leader of the University of Alberta’s computer poker research group, Michael Bowling, said.
Machine Thinking Like Human
DeepStack was developed in cooperation with colleagues from Czech Technical University in Prague and it is based on advanced processes that not only understand the strength of its own hand, but it also makes a good guess at its opponent’s hand, while at the same time making an assessment whether the opponent is thinking about bluffing.
It is a deep-thinking and deep-learning program which dives within a winning strategy of the aforementioned kind of poker to understand aspects of the game that are highly human, psychological even.
Comprehensive testing of the DeepStack system was performed which involved 33 poker professionals from 17 countries which completed 3,000 games against the computer in four weeks back in 2016.
Only a third of the human players were able to beat the machine.