The latest developments in Artificial intelligence (AI) have reached another new landmark after Libratus, a powerful poker bot developed by Tuomas Sandholm, a machine-learning professor at Carnegie Mellon and his PhD student, Noam Brown, defeated a number of world-class professional poker players in games of No Limit Texas Hold‘Em at a January tournament. Despite success in poker, the odds for AI or any machine replacing jobs soon are still quite slim.

In the online tournament hosted by Rivers Casino in Pittsburgh, Pennsylvania, Libratus won $1,776,250 over 120,000 hands reportedly, using an approximation approach that is similar to that of human instinct. If this is the case, it signifies a huge achievement for game-playing AI, as noted by Michael Wellman, a University of Michigan professor specializing in Game Theory and AI who described the win as "a major milestone".

For decades, AI research in the gaming industry has focused on games such as Othello, chess, Go, Jeopardy! and checkers. The general outcome was that the best AI was able to surpass the best human brain. Poker however, particularly No Limit Texas Hold ‘Em, remained difficult for AI researchers to crack. Prior to this tournament, no AI had been able to beat the best professionals in this particular game. However, non-peer reviewed research conducted by a team responsible for an AI-based poker-playing software program began breaking new ground in 2015.

Developed by a team led by Michael Bowling of the University of Alberta, DeepStack learned to play poker by playing itself over several hands. After each game, it went through the process of revisiting and refining its strategy, resulting in an optimized approach to playing. Due to its complexity, No Limit Texas Hold ‘Em, which has 10,160 possible plays in each hand, was too difficult for machines to play at an expert level for a long time. But the team behind DeepStack overcame this by integrating a fast approximation technique into the development of the software, refined by feeding previous outcomes into a deep-learning algorithm. According to the research, DeepStack has successfully played over 40,000 hands against general poker players with clear wins in each.

Although the exact details of how Libratus approaches a poker game haven’t been disclosed, Brown says that it basically attempts to calculate each possible outcome early in the game in much the same way that DeepStack does. Bowling and his team have compared this approximation technique to those gut instincts human players feel when an opponent is bluffing or holding a winning hand. He calls it “DeepStack’s intuition" - describing it is an instinctive feeling for the total value of "any possible private cards in any possible poker situation.”

DeepStack’s advancements and Libratus’ victory have ignited some fears about the future of online poker. However, fears about losing to bots online, or bots being used to fuel extortion, are a little exaggerated. Because online poker is played in a multi-player environment, bots will have a much harder time trying to solve the game. The software mentioned in this article has won against players in head to head battles, not against a table of five or more players. Libratus’ creators believe that it actually adds a new depth to the game, and the algorithms used in the bot may help pave the way for further advancements in general AI.