Scientists Design Unbeatable Limit Hold’em Software?

We had this training phase where the program started off playing uniform random against itself," meaning that "it had no idea what it was doing other than following the rules of the game." www.theverge.com;

Michael Johanson, one of the co-authors of the study, explained that the program works in such a fashion that it searches through a huge database of pre-computed situations to find the right move for any possible situation. As one can imagine, building such a base was a big undertaking. The computer played against itself and got better with time.

If it decides to play randomly, and loses money, then it goes back and computes how much money it would have won if it had raised instead." www.theverge.com;

The training phase took 70 days to complete and 200 computers equipped with 32 GB of RAM, and 24 CPUs each were used. Over the course of that period, the machine played itself and learned from its mistakes, making a note (‘regret value’) every time it could have made a better play.

After the seventy-day period, they stopped the process, as the machine’s game approached such perfection that it was impossible to tell it apart from the actual perfect. In fact, the creators believe that anything the program could learn from this point on would bring barely any value to its overall game.

Michael Bowling, another co-author of the study, and a team of scientists started working on developing heads-up limit software capable of defeating human players back in 2003. They were partially successful when ‘Polaris’ was able to win three out of six games against some of the top pros in 2008.

After that, the next logical step was aiming to create software that plays perfect poker. Although deterred at first by the size of the project, the team decided to go for it and although it took 520 terabytes of data, they came up with the solution.

Bowling says that the next challenge will most likely be trying to solve No Limit Hold’em, but he is aware that it could be nearly impossible due to the game’s complexities. However, he does believe that they could devise a piece of software capable of beating the best human players.

We’ll see if they will be successful and how long it will take them, but in the meantime, you can check out ‘Cepheus’ at the University of Alberta pages.

Did you like this article?

Tweet +0