How Poker Still Confounds Artificial Intelligence

It has already been hailed as the death-knell of chess, go and yes, even poker, but the rise in AI - Artificial Intelligence - still has a long way to go before it solves the games we love completely and replaces humans at the top of the ‘clever tree’ according to one distinguished academic.

Noam Brown, a Ph.D. student and AI researcher at Carnegie Mellon University, and one of the main forces behind the Libratus AI program which defeated a quartet of the strongest heads-up human players in the world earlier this year, has calculated that there are 10^163 (that’s 10 followed by 163 zeroes!) different game situations in no-limit hold’em.

According to MatchPoker, ‘assuming starting stacks of 20,000 chips’, this makes it impossible for even the strongest computers to calculate all possible situations!’

In matching up the AI problems with what we see in actual poker, the focus of the AI researchers and the problems they face becomes very clear…

Match Poker itself was devised by the International Federation of Poker to reduce the element of luck inherent in traditional forms of poker, ensuring that it is a contest based on players’ skill and conforms to the accepted definition of a sport.

Their format mirrors that of the Libratus match against the team of Jason Les, Dong Kim, Daniel McAuley and Jimmy Chou which lost horrendously - Libratus taking the best part of $1.76million from the humans over 120,000 hands and leaving everyone scratching their heads at its almost complete dominance!

This, of course, is far from seeing AI solving poker of course – the numbers Brown is bandying about being so immense that it could be decades before computers are strong enough – if they ever are - to solve poker, which as we can see in the diagram above is a game of incomplete information, or ‘imperfect knowledge’ as they term it.

In chess, by contrast, all the variables are out there for all to see, although that’s easier said than done. Humans – including World Champion Magnus Carlsen - can’t get anywhere near the best machines nowadays, but with the number of all possible games of 40 moves standing at 10^120 (far greater than the number of electrons in the observable universe!) the best computers can get nowhere near to solving the game.

This means that poker, a huge order of magnitude more difficult in the solving sense, has a long way to go before AI puts it out of business. And of course, we still have chess tournaments played all around the world – the only real difference being that the biggest events see signal jammers, airport-style scanners and random checks to ensure that nobody is carrying some AI on their person during play!

As Noam Brown’s professor of computer science at Carnegie Mellon, Tuomas Sandholm, said before the Libratus match which stunned the poker world:

"Since the earliest days of AI research, beating top human players has been a powerful measure of progress in the field,” referencing computer wins against top chess Grandmasters and the world’s leading Go experts.

He added that:

"Poker poses a far more difficult challenge than these games, as it requires a machine to make extremely complicated decisions based on incomplete information while contending with bluffs, slow play and other ploys.”

For how much longer we don’t know, but try writing out 10^163 and see how you get on!