Poker bots are far from a problem yet. They excel at heads up limit hold'em, but as more players enter the table they get progressively worse. Even the best of them can't yet beat a low limit full table.

Eventually they will though. And there will be no real way to stop them. We've got some time, and how much depends on who you ask.

Also, the human vs computer match cited doesn't really mean much, for reasons I'll leave for a poker blog entry.

When I heard about poker bots, I thought "oh, I should get one." So I cruised the seedier side of the internet and downloaded one.

Then, I discovered that you had to program the AI yourself!

Matt, if a well-programmed pokerbot is excellent at heads up limit hold'em, what percentage of online limit heads up players do you think are bots or bot-assisted? Can a bot win an average 2/4 game? What about an average 30/60 game?

The barrier to entry as far as a bot goes is certainly the AI. Programming the rest is far more doable. The most notable team that has been publishing poker-related AI research is headed by Darse Billings at the University of Alberta. Their bots can be purchased online as a training tool called Poker Academy.

From what I've seen first hand and heard from poker players I respect, the only bot worthwhile is the heads up limit hold'em one. I don't know how far they've come since the Poki I played against, but I doubt that one could have beaten a full $1/$2 game. I'd guess it would break even at $0.50/$1 online.

The short handed bot didn't seem much better. The heads up one, however, could probably trounce anyone who would typically play below $15/$30, and it might even be higher up the scale than that.

To put it in perspective, many very notable winners at the biggest games online used to fear a player who went by Neverlose since they thought he was that bot. You can google for "Neverlose bot" and read all about it.

So at heads up limit I would intuitively suspect there are bots, though to be honest I've played a decent amount of it in the $3/$6 all the way up to $100/$200 range and the vast majority of players at low/mid limits are atrocious. So if there are bots in heads up, they are either poorly designed ones or they are in the higher limits exclusively (which really wouldn't make sense, since there is probably much more money to be made for someone who can play all day every day in the lower limits).

As far as other games go, I suspect bots reside largely in low level sit and goes. I once had a bot breaking even around the $22 level using Winholdem, which at the time only supported min-raise or shove as the bet amounts (no idea what it can do now, I haven't used it in years) and my AI used nothing more than if/then statements. Someone capable of any sort of AI programming more advanced than that could probably beat them handily. Due to some other terrible limitations of WinHold'ems software I never pursued it much further.

I love that blog. Arnold Kling is spot on when he writes about entrepreneurship. Bryan Caplan is as close to a sane anarcho-capitalist as you can come.

At CMU's Robotics Institute a few years ago, a friend made a poker bot that was pretty good. It was for a machine learning class -- but it used a lot of standard techniques like constraints and search. Another guy made a parser to take the screen from a popular online poker system, and input the state into any software you want. The final step -- to automatically make moves and plug-in a simple chat-bot to counter accusations of non-humanity, was not done by them.

I am 100% certain it has been done many, many times. Of course, why would you hear about it in the news?

I don't think I really buy into their politics (and this isn't really the forum to use to get into that - write me if you're curious), but they're certainly smart guys and worth reading from time to time.