Poker Bots on the Rise: A Guest Blog

Ian Ayres is an economist and lawyer at Yale and the author of Super Crunchers, which we excerpted here. He has agreed to write occasional guest posts on our blog, which delights us, since he has a lot of compelling interests and insights.

Ian is not the only notable guest blogger who will turn up on this site in the future — although, in keeping with the casual and random nature of this blog, I don’t want to make any promises since we tend to play things by ear. Suffice it to say that we will continue to strive to bring you the best in arcane, trivial, and time-wasting blogification by the likes of ourselves, Ian Ayres, and sundry others. — SJD

It is not an opportune time to start an online gambling site for checkers.

In July, researchers at University of Alberta “solved the game” using brute computer force. As such, their computer knew the best strategy to play in any of the possible 50 billion checker positions. Humans should now be very scared to bet money against any virtual opponent, for fear that they are really facing the Alberta computer or its clone. (You also wouldn’t want to play a money game against a computer in Connect Four or Othello, or even backgammon.)

What’s true for checkers is also becoming true for poker. The same group at Alberta has just shown that their Polaris program is on the verge of taking down poker professionals. In the recent “Man vs. Machine” Poker Championship, Phil “The Unabomber” Laak and Ali Eslami barely beat Polaris two sessions to one (with one a virtual draw). After the tournament, Laak candidly acknowledged that “the bots are closing in.”

People have expressed worry that computerized competition and game-theoretic solutions could psychologically depress player enthusiasm. What’s the point of playing checkers now that we know it should always end in a draw? What’s the thrill of playing chess, when we humans know that we can never be the best?

But the rise of gambling bots may soon depress online poker participation for a very different reason. In the very near future, online poker may become a suckers’ game that humans won’t have a chance to win. Bots are quite scale-able and it will be virtually impossible to prohibit computer or computer-assisted online playing.

Poker sites are trying to assure customers that they will kick bots off their site and seize their assets. But unlike the statistical trail left by crude poker cheats at Absolute Poker, it is possible for bots to randomize their strategies and even hire individual humans to run them.

Ultimately the Albertus Polaris program and its offspring could be more effective than any Justice Department indictment in crippling the growth of online gambling. Indeed, our government might even think about subsidizing the development and use of these bots. Imagine a DARPA-like competition for creating a bot that can beat the average law breaker. Constructing a bot that can consistently win (and then publicizing this fact) is a sure step toward virtual temperance. (By the way, I’m agnostic about whether online poker should be illegal.)

Poker enthusiasts have argued for online legalization, saying that poker is a game of skill. And of course, it is (just like chess and checkers). But ironically, it’s because poker is a game of skill that humans’ chance of winning are undermined. Unlike checkers, the key to poker is to predict whether other players are bluffing. On the Internet (without the possibility of visual cues), computers are probably better at predicting a rival’s hand from his or her past play. But computers are much better at confounding the expectations of their human opponents. Computers can play randomized strategies much better than we can. Our brains are so hardwired to see patterns, it’s devilishly hard for most of us to generate random behavior.

Indeed, take a minute and try to write down a random sequence of 200 heads or tails. If you actually flip a coin that many times, there’s a very large chance (98%) that there will be a run of at least 6 heads or 6 tails in a row. But very few people can bring themselves to produce such runs in trying to be random. Your iPod’s shuffle function isn’t broken when it plays songs from the same artist two or three times in a row. By chance, runs do happen.

When I play poker, I use my watch as a crude random number generator. Before my first bet, I look down and bluff if the second hand on my watch is between :00 and :06. Unlike machines, people have a hard time ignoring the past. Our biggest tells aren’t facial tics but that we just can’t stop ourselves from playing non-randomly. With training, we can get better, but we should fool ourselves. The handwriting is on the wall. High quality bots are an online gambler’s worst nightmare.

Bots won’t kill poker. They’ll just drive it off line. Old fashioned “humans-only” competitions will still thrive. But this is one Darwinian struggle where the unaided human mind is definitely not the fittest.



That is the argument used by the sites when questioned about the extreme numbers of high hands.

I remind them as I will you that I tracked 10,000 hands live and 10,000 online. Ten thousand is ten thousand reguardless of where you play them. I am not comparing hours of play. If you compared the play by the hands per hour the only difference there should be, is how much faster you could see 10,000 hands played online.

Horia Margarit

It is true that a computer program can best the human brain at mathematical computation. A poker playing program may have the necessary algorithms to compute all the possible permutations of any given hand. Of course, it will not have the system resources [processing power, RAM, swap space] to actually compute said permutations. But that is not my point.

Computers do not have the capacity for emotion. They do not have the capacity to compute emotion. If they did, Artificial Intelligence [most notably the learning and linguistics sub-divisions] would be decades ahead of where they are now. If our computers could simulate emotion, Cognitive Science researchers would be experimenting with Androids. Yet we know that these endeavors, while fascinating to think about, are not presently undertaken.

I. How then does a poker robot compute the behavior of a human player? It finds patterns in our pseudo-random behavior.

Let us briefly examine this answer.

On the one hand we have that 1) human behavior is only somewhat random, meaning that there is an underlying pattern, and that 2) machines are very good at generating random events, and at finding patterns in somewhat random events.

On the other hand we have that 1) computer programs are syntactic expressions of algorithms which, in the case of programs such as poker robots, are mathematical algorithms, and that 2) mathematics and algorithms themselves are human abstractions, created by and IMPLEMENTED by humans.

Now we are confronted with the error of our assessment in proposition I.

A computer program which is designed and implemented by humans, using the science and mathematics discovered by humans, cannot best humans in what they have created. To say this is to go down the road of HAL, the Terminator, Robocop, and Marvin [from Hitchhiker's Guide To The Galaxy].

As previously mentioned, while fascinating to think about, these endeavors are not presently undertaken.


Sam W

Hey there,

I have been fascinated by poker bots for quite some time now and have tried most of them out.

From personal experience I have to say that works best.

Best of Luck!

Rachel Rosenberg

So.. do you think that our site should start stocking up on checker supplies instead of poker chips?

Jason McSard

i love poker


Great read... I believe that the reason you see higher card hands online is the fact that the players are not as good as the live game players and it is way easier to play crappy cards with a press of a button than actually putting 100 chips on the table!!! Good research though.. could be true!!


I still want to know how poker wound up in the sports pages of several newspapers.


Poker bots can easily clear a small edge at lower limit tables, but the best opponents know how to change thier strategy often and take advantage of a computers inability to adapt.


"even hire individual humans to run them"

If the poker bots are hiring people, I think we may have more pressing issues to deal with than poker.

"Before my first bet, I look down and bluff if the second hand on my watch is between :00 and :06."

How about a game? I'm not very good.


Well, at least now I know why I was so much better at computer games as a kid than I am now! I haven't gotten dumber; the computers have gotten smarter... although I do feel dumber for not having figured this out sooner.


The Polaris match and others like it are a little misleading. The poker bots that can compete well with humans generally only play heads-up, limit Hold Em. That kind of game, which only has two players and very limitted betting options, is much easier for a computer to "brute force" than a full table game of No Limit Hold Em.

Doing well at a full table, No Limit game takes a lot different kind of thinking than what computers are good at. The most important part in those situations is being able to mentally model the actions of your oponents, which computers are very bad at right now. I think computers will need to progress to the point that they can understand the meaning of what humans are saying in a conversation before they will be able to understand what's going on in the average No Limit Hold Em game.

Tim Harford

Welcome, Ian!
This does seem likely, but some poker-watchers think that a combination of sophisticated virtual reality and reputational mechanisms may maintain trust in some form of the online game. We'll see.

Fans of this post may also be interested in a feature article I wrote about a year ago, covering "Poker Machine" Chris Ferguson, the pokerbots - including those from Alberta - and the economics behind it all. I come to a similar conclusion to Ian Ayres.

The article is available at:


CT, Look at the people who play 20 tables at the same time of No Limit Hold'em. They are basically bots who just occasionally have to "think/deviate from their standard strategy". Everything else is memorized. The "thinking" part will be here before we know it (it already is to some extent). And like Ian said, machines are much better randomizers than us.


CT (#5):

Bots aren't "good" or "bad" at any particular kind of strategical thinking - they're only more or less sophisticated; i.e. they can only account for certain factors. That's not an inherent limitation, but a technological one that can be overpassed simply by intelligently adding complexity to existing methods.

In short, there are different tells and strategies at play in your no limit, full-table scenario, which the Polaris doesn't yet incorporate in its methodology.

But that doesn't mean it can't.


The stronger bots are in heads up (HU) matches.

However, they definitely exist at lower limits. The problem isn't that the good players can't beat them - the problem is that they suck money out of the really bad players.

I'd much rather sit at a table with 5 bad players than 3 bad players and 2 OK bots.


Mr. Ayres,

I look forward to your posts. I read your Super Crunchers book because of a recommendation from this Freakonomics blog.

I would be curious to find out if the bots ever take unreasonable chances based on the potential bluffing by an opponent. And if so, is there a threshold hand it would require before calling such a bet, or any bet for that matter.


Gee, I guess you read "The Professor, the Banker, and the Suicide King" too. You'll remember that the Banker got smoked when the pros realized how he was "randomizing" his play with his watch. Good luck.

Holdem Caulfield

Here's a link (see below) to an interesting discussion showing that a decision support system designed for human players can beat the top bots like Polaris. The results aren't surprising. The DSS approach roughly equates humans and bots on mathematical knowledge of the game thus making it a battle between human intelligence vs. bot intelligence. Personally, I'd love to play against all bots--bots can be beaten.


Bots are not nearly as dangerous as Ian makes them out to be. It is in the best interest of the poker sites to keep their sites free of bots so that people will feel safe playing there. If bots become a bigger problem, sites will do more to stop them. With the amount of information available to the sites, it would be very easy for them to detect bots. They can already track your mouse movements, processes running, timing of your clicks, and how you play all your hands. Also they could change the interface frequently to thwart bots.

It would become a game of cat and mouse, but with one major asymmetry. The sites have millions of dollars to gain by keeping ahead of the botters while the botters only have tens of thousands to gain by running their bots.

Alan Bostick

I, for one, welcome our new cybernetic masters....

There's something irrational about the fear of poker bots. Why is it a bad thing when a computer is set up to beat the poker games in cyberspace, when there are thousands of skilled flesh-and-blood player doing the same thing?

Online poker pros don't like bots because they take money from bad players out of the game. But that is exactly what the flesh-and-blood poker pros are doing, often with high-tech aids that begin to rival what bots are capable of.

The online poker boom has been a bonanza for experienced players. Now a new generation of experienced players has emerged, and the games are getting tougher. People who have been playing poker in brick-and-mortar cardrooms for decades have seen this cycle before: after the legalization of Texas Hold'em in California cardrooms; when poker was legalized in Atlantic City casinos; when Foxwoods opened its poker room; and now, with the double impact of televised poker tournaments and the explosive growth of Internet poker sites. After every boom, there was a contraction as some losing players dropped out and those who remained got smarter. But the busts have been less extreme than the booms that preceded them, and always ended up with more money for the skilled players -- automated or otherwise -- to win after the bust than before its preceding boom.

Today's poker player has a fairer shot at winning money in an honestly run online game, whether or not the other players are bots, than she did fifty years ago in the old Las Vegas "snatch games," in which the casino owners encouraged the dealers to take as much money out of the pot as the players would tolerate.