There’s an easy joke to be made, perhaps, about machines having perfect poker faces, given how they have no actual “faces” to read any tells on. However, programmers have actually had a harder time engineering A.I. for this game in comparison to others, like chess and checkers, because of its uncertainties. A computer has a hard time even integrating the concept of bluffing, for instance, and that’s obviously essential to poker.
Inspired, or emboldened, by the victory of the AlphaGo A.I. against world champion Go player Lee Sedol last year, a professor and PhD student from Carnegie Mellon took another stab at mastering a game described as “the last frontier of game solving” for artificial intelligence. They created two programs, Libratus and Bridges, which would work in tandem against four pro Head’s Up, No-Limit Texas Hold ‘Em players in a “Brains vs. A.I. Rematch” at Pittsburgh’s Rivers Casino.
Yes, the machines ended up winning.
Though, to be fair, it might be more accurate to characterize these matches as two-on-one instead of one-on-one. As noted, Libratus reviews past games, identifying the strategies of its opponents, while Bridges perform real-time calculations that inform the calls for each hand. Either way, their tactics worked. After 120,000 hands were played, Libratus and Bridges scored a $1,766,250 lead against the four pros who took them on.
To frame that win in greater context, Carnegie Mellon staged a similar challenge a little under two years ago (presumably the “match” which this series is a “rematch” for) and the humans beat the A.I. that time with a $732,713 lead. While a two and half million advancement in about two years is impressive, the techies are more excited about the breakthrough in “general” intelligence. Basically, Libratus has demonstrated a capability to learn, and to apply what it has learned, as opposed to just operating within a game’s specific parameters. Alas, it does seem its win has definitively demonstrated that even the gut instincts of poker can be understood with numbers and probability.
Do you think you could psych Libratus out at the poker table? Share your thoughts in the comments.
Featured Image Credit: Carnegie Mellon University