A really good article/op-ed piece on Pokerati about poker bots Tim Chilcote. Tim sat down with Darse Billings of the University of Alberta Computer Poker Research Group, the folks who created the poker bot Polaris that played Phil Laak and Ali Eslami, to get some insight into what sort of threat a poker bot would be in the online poker environment.
One of the more interesting points from the article is:
Bots sound dangerous, and it would be easy to infer that their skill is only going to grow and that their dominance of the poker world is a forgone conclusion. But for online players who fear losing their bankroll to a robot opponent, Billings wholeheartedly dismissed their concern as ludicrous. “The amount of time and knowledge that would go into making even a mediocre bot,” he said, “makes them nearly impossible to run online.”
Billings also offered that bots are completely beatable, even suggesting that players should welcome bots. Part of Billings’ initial research for the Polaris project was to build a rock-paper-scissors bot that would consistently break even. The Polaris poker bot was an extension of this rock-paper-scissors design, and the bot plays poker exclusively to break even, only profiting when human players succumb to their own weaknesses. If you’re still worried, it’s worth mentioning that, despite fears, bots are not out fishing in multi-table tournaments; they’re designed to play heads-up limit games. So a strong limit player could, in theory, profit by playing bots.
I have written about this in the past. In fact, in this post “Poker Bots Can Bluff” there is a quote from a French researcher who has created a poker bot that can bluff effectively saying “Given the current state of poker bots, if you are losing to them you should be ashamed”.
I’m not in favor or poker bots. I find them interesting from a purely intellectual perspective. However, it’s pretty clear that some of the people who probably know as much about poker bots as anyone don’t view them as a credible threat at this stage in their development. The University of Alberta Computer Poker Research Group has been working on this with some of the best minds out there since 1995 and if they think there’s still a long way to go then I have my doubts about whether some guy is going to sell you the ultimate winning poker bot for $79.95.
The biggest problem with creating a realistic poker bot that can beat the game is that it’s expensive to develop. If your goal was simply to make money it would be far more profitable to set up a very basic, non-thinking, break-even bot and collect the rakeback. Then you set up 50 poker room accounts and let them run wild at relatively low-stakes tables. That’s basically what a lot of bot software vendors sell today (though most are crap). You tell it how to play under a variety of conditions, how aggressive or passive you want to be, etc and then let it go play some break-even poker.
But anything that’s programmed to play by a set of rules, rather than to think for itself, will be pretty easy to figure out if you’re even a little observant. If you notice that the suspected bot never 3-bets pre-flop without a certain range of hands then when you get 3-bet you know exactly what range of hands you’re up against.
Certainly some level of complexity could be introduced by the bot author(s) but you still have a bot that doesn’t really think. It’s playing by some set of instructions that tell it what to do. You can try to randomize things as much as possible but too much randomization of how it plays would cause it to lose.
When you look at some of the recent bot busts that have been in the news they typically employ very crude tactics. Many even seem unable to beat the low-stakes games and thus collude in order to win.
The biggest problem that poker rooms have today is that they aren’t doing a good enough job keeping even these very obvious bots off their sites. Very few sites devote any real level of money or resources for identifying and combating bots. Many would be picked up just by improving the automated collusion detection on the site.