An artificial intelligence system developed by Carnegie Mellon University has defeated four professional poker players in a 20-day poker tournament held last month in Pittsburgh. The AI system named Libratus was developed by computer science professor Tuomas Sandholm and PhD student Noam Brown, and hosted on the Pittsburgh Supercomputing Center’s Bridges computer, according to a report here.
Game playing has always been a domain where the fields of Machine Learning and Artificial Intelligence have met their most significant challenges and where some of its most significant advances have been tested. Arthur Samuel‘s checkers playing program from 1959 was a pioneering achievements of early Machine Learning and checkers was the first game that was completely solved on a computer. The victory of IBM’s Deep Blue over World Champion Garry Kasparov, and the strongest chess player of the modern era, in their chess battle of 1997 was an epochal event in the history of AI. More recently, Watson, the AI from IBM defeated two human champions in the TV trivia game Jeopardy! in 2011 and deep learning arrived with a bang on the world AI stage by defeating one of the world’s strongest wei ch’i (go) players in 2016.
The difference between poker and all the other games that AI mastered in the past, checkers, chess, wei ch’i and Jeopardy! is that these games of perfect information while in poker is not, since you do not get to see your opponent’s hand. Also the ability to bluff and to read bluffs is a very important part of poker play, an area in which humans have traditionally held an edge over machines. Of course, an AI will have one major advantage in poker in that they always have a “poker face”, and they will never give their hand away through involuntary “tells”.
The victory of Libratus over champion human poker players is as significant as past AI victories in chess and wei ch’i, perhaps even more so. The analytical abilities required for champion level poker play are similar to what is needed to excel at business negotiation, military strategy, espionage tradecraft and political campaigning. That suggests that this major advance in machine poker is likely to lead to major advances in Machine Learning and Artificial Intelligence.