After successfully solving games like chess and even Go, scientists seem bent to crack poker next. The research into an artificial intelligence that could stand up to top-tier poker players has been going on for some time now. The results so far haven’t been very encouraging for these scientists, but they aren’t giving up.
Libratus, a poker playing super computer developed at Carnegie Mellon University, represents the latest attempt to crack the barrier. The match is set to take place starting on Wednesday at Rivers Casino in Pitsburgh. The AI will play 120,000 hands of Texas Hold’em against four human players with $200,000 up for grabs.
Claudico & “statistical tie”
Back in 2015, we saw the first serious match of this type. The poker-playing software going under the name Claudico was pitched against four of the best No Limit Hold’em heads-up specialists. The AI played 20,000 hands against each player and by the end of it all, Claudico was beaten pretty badly.
Of course, scientists were quick to call it a “statistical tie.” Many felt that this conclusion after the fact wasn’t really fitting. The team Claudico simply decided that although the bot lost, the loss wasn’t big enough over the given sample. The “statistical tie” implies that, should the match continued, both sides would break even.
New and improved Libratus
The new poker-playing software Libratus was developed by the professor Tuomas Sandholm of Carnegie Mellon University and Noam Brown, a Ph. D. Student. Sandholm, Brown, and the team behind Libratus used the information from the 2015 Brains vs. Artificial Intelligence match to build an improved AI version.
The code for Libratus was written from scratch. Although the team used the information available to them, they decided to build a completely fresh artificial intelligence for this challenge. The software should get rid of some of Claudico’s flaws, like constant limping from the small blind. According to Sandholm, the new AI will be much more unpredictable.
The match will take place over twenty days. Professional players who will take on Libratus in individual heads-up matches are Jason Les and Dong Kim, who were involved in the 2015 challenge, as well as Jimmy Chou and Daniel McAuley. If they manage to beat the AI, the four will get to share the $200,000 prize pool.
As expected, the announcement created some buzz in the poker community. Reactions vary, as everyone seems to have a different take on the whole poker-playing AI concept. Some believe that developing software like this isn’t a good idea as it will lead to creation of nearly perfect poker bots, ruining poker as we know it.
Others are simply pumped about the match and curious to see what will happen this time around. The fact that Doug Polk, arguably the best heads-up player around, isn’t a part of the team (as opposed to 2015) doesn’t sit well with a part of those looking forward to the match. They believe that in a competition such as this, “humans” should be represented by the best.
Sandholm and his team keep insisting that beating poker isn’t their endgame. They want to develop an AI capable of making complicated decisions on the fly and put it to use in various other aspects. Hence, poker simply serves as a great testing ground.
While concerns about developing perfect poker bots aren’t completely out of place, it will be interesting to see how this match plays out. Regardless of what scientists claim, so far there hasn’t been an AI out there that could really match human competition. Will new and improved Libratus rise to the challenge?