How CPRG’s AI programs approximate real-world poker tactics
Poker’s move into the digital realm has been very beneficial for the gaming industry. Even with the card game’s popularity already reaching new heights due to the exposure provided by the 2003 World Series of Poker TV broadcast, online poker was able to further push that popularity up to 11 thanks to its wider reach and more accessible nature.
A lot of poker pros actually got their start by plying their skills first in virtual poker rooms. In fact, established superstars like Patrik Antonius and Phil Ivey still make their respective presences felt in online matches. These indicate that just like their real-world counterparts, online poker games provide just as rich a tactical gameplay experience. Online gaming operator bwin.party’s PartyPoker brand even provides strategies on its blog page that cater specifically to digital matches. Furthermore, its YouTube channel has additional online-specific tutorials hosted by Kara Scott, herself a pro player.
The question, of course, is how the AI of virtual poker compares to good old-fashioned human thinking; and if the University of Alberta’s Computer Poker Research Group (CPRG) has anything to say about it, the current state of poker AI is highly comparable, if not 100% emulated. And the best part is, it’s constantly evolving.
Most video gamers know the CPRG as the team behind Poki AI, the highly acclaimed artificial intelligence program used to power the video game Stacked with Daniel Negreanu. The program’s claim to fame is its ability to switch tactics on the fly instead of sticking to one pre-programmed set of instructions, allowing for more dynamic gameplay akin to actual poker sessions.
As amazing as Poki is, a more notable achievement for the CPRG is its series of AI collectively called Hyperborean. These AI codes have proven their mettle in the Annual Computer Poker Competition. As CPRG researcher Richard Gibson explains, the Hyperborean programs operate on a system of tactics that take Nash equilibrium into heavy consideration.
To better illustrate this, Gibson differentiates Hyperborean from the chess AI Deep Blue. The latter, he says, relies primarily on a technique called the alpha-beta search, where the aim is to go for that one move statistically proven to have the best outcome given the specific arrangement of the remaining pieces.
On the other hand, since poker has a lot more unknown variables to consider, Hyperborean sees the best strategy as that of a more defensive bent, where the assumption is no one player (human or otherwise) has anything to gain without considering the strategy of the others. Therefore, until a game goes into its later stages when more constants are introduced, the best bet (pun intended) is to hold back until more cards are dealt.
That, of course, is a highly simplified explanation of the obviously more complex programming coded into CPRG’s poker AIs. The point remains, though, that the way these programs approximate human thinking is getting more and more fully realized.