Another one bites the dust. Libratus, an artificial intelligence program from Carnegie Mellon University has beaten four top Texas Hold’em poker players in a series of two player games over the last three weeks.
That’s big news because unlike chess and go, poker playing relies on incomplete information. And you have to decide when to bet, and how much. So there is a more complex set of factors than just calculating moves to consider.
A range of games have seen top human players lose to computers over the last few decades
As Toby also notes, the Texas hold’em approach is largely brute force computing, relying on a supercomputer. This is different from the deep learning approach of AlphaGo, and the machine learning route that IBM’s Watson that won Jeopardy! But the Libratus software was developed just by a PhD student and his supervisor, rather than the larger teams involved with Watson and AlphaGo.
The poker win doesn’t mean that anyone can now break the casino with a poker AI app on their phone. For one thing electronic devices aren’t allowed to be used at gaming tables. Secondly, most people don’t have access to a supercomputer. And, the algorithm hasn’t yet won when more than two players play – the computational challenge increases as the number of players rises.
Still, the pace at which algorithms are besting humans at complex games, even surprising their designers with some of the moves they make, illustrates how software can master a range of well-defined rule-based tasks if it is well trained.
It isn’t time for humans to, in the words of that great 20th Century philosopher Kenny Rogers, fold ’em or run. People still play chess, checkers, and scrabble without computers. But it is another reminder of the timeliness for more consideration of how we chose to “play” with machines that could take on more of the things we have taken for granted as human tasks in activities such as legal and financial services, customer service, farming, medicine, policy advice, …