Greengard's ChessNinja.com

And Then They Came for the Poker Players...

| Permalink | 19 comments

Now that so many strong chessplayers are more involved in professional poker (including current Dortmund leader Jan Gustafsson, I'm told), the computers are pushing deeper into that domain as well. Bluffing and betting move the game out of the realm of pure calculation where chess resides, but they machines are clearly getting stronger. The University of Alberta, home of checker champion Chinook and its author Jonathan Schaeffer, is also on the leading edge of computer poker. Their program Polaris is currently taking on some strong human players.The machine came close to a win in a similar event last year. This article in an Edmonton paper has the most info. This site is tracking the results day by day and has reports on each match.

As described, the machine uses an interesting combination of playing the odds and learning the style of the opponents based on previous play. Such learning would be helpful in having a chess program teach you based on the types of mistakes you make instead of just pointing out your blunders and bashing your head in on demand. The moment commercial programs become significantly stronger than the average serious human player, the fantastically lucrative world of online poker will be in trouble. Online poker coaching will survive, but if eventually everyone is using software, the broader gambling market will fall apart.


And the one thing that humans will have remain in their exclusive domain is procreation. Unless the computers truly do become f***ing machines...

Celebrate the welcome to earth day of newbies. They're just little babies and they need our love.

Re f***ing machines: have you seen David Levy's latest book?


Nice one Bill :)

On topic I think machines don't have the advantage like in chess. Poker calculations are not that complex. On the other hand-assessing situations (i.e. position evaluation) is not so complex as in chess, and humans do not have an advantage there. I think it will be possible to make software to play better than humans, but not like chess engines are against human GMs. Human poker pros should be much closer to optimal play than chess GMs.

In his song "Ms. Pinky", Frank Zappa broke important ground for the eminent Dr. Levy:

I got a girl with a little rubber head
Rinse her out every night just before I go to bed
She never talk back like a lady might do
An' she looks like she loves it every time I get through

And her name is
P-I-N, no lie
K-Y, me-oh-my
She's $69.95
Give her a try

Her eyes is all shut in an ecstasy face
You can cram it down her throat, people, any old place
Throw the little switch on her battery pack
You can poot it, you can shoot it till your wife gets back

And her name is P-I-N-K-Y
P-I-N, I cry
K-Y, don't be shy
$69.95, boy
Give her a try

I feel like I have to post this every time Mig posts about poker. Poker, as a computer problem, is nothing like checkers or chess. It is a nuanced game that cannot be brute forced because it lacks a mathematical shortcut for analysis like chess has w/ material.

Also, Polaris plays heads-up limit hold em, which besides being very simple, is sort of a stupid game of showdowns and not popular at all. The most popular game by far is no-limit ring, an immensely more psychologically complex and nuanced game than hu limit, and I doubt we'll ever see a computer win at nl ring anytime soon at anything more than dollar stakes.

Check out http://www.outstandingpoker.com

OutstandingPoker.com is a poker training website that is run by Tim (also known as "Atta22") and Errol (also known as "killer_be_99"). Outstanding Poker offers assistance and guidance at every possible angle: videos, articles, and podcasts all at one site from pro poker players.


OutstandingPoker.com is a poker training website that is run by Tim (also known as "Atta22") and Errol (also known as "killer_be_99"). Outstanding Poker offers assistance and guidance at every possible angle: videos, articles, and podcasts all at one site from pro poker players.


golfstrom, your comments seem pretty off base to me. Material as a shortcut for positional analysis seems much less conducive than a straight up consideration of some fairly basic probabilities, so I wouldn't be so sure that poker is that much more subtle.

As for HU limit vs. 6 max NL and things like that, I again think that your distinction is not obvious. For one thing, in HU limit the computer is going to have to play a lot more hands and make many more nontrivial decisions. The continuum of bet sizes available is something of a red herring, because in practice you're generally not using more than four or five different bet sizes relative to the pot anyway. So I don't think it's clear that NL is going to be vastly harder to attack. Further, even if programs merely get good enough to crush low and medium stakes games, that's a problem for the economy, because those games are likely to be sources of players for the bigger games. If those games dry up, then to a certain extent all bigger games will suffer some as well.

'so I wouldn't be so sure that poker is that much more subtle.'

Given that chess is a game of complete information whereas poker is a game of imperfect information, of course poker is a much more subtle problem for computers to work out.

'As for HU limit vs. 6 max NL and things like that, I again think that your distinction is not obvious.'

Texas hold 'em, according to wikipedia, with one opponent the amount of starting hand combinations is 1,225, with 5 opponents there is over 9 trillion! So it is unclear how an evaluation function that consists of calculating 'a few basic probabilities' is going to help in this case, without having many more complex factors programmed into the machine.

Ben, if you think poker is game of 'fairly basic probabilities', you probably also think chess is a game of 'moving some pieces around'.

I don't think you realize at all how specialized HU LHE really is. It's a game of chicken. You play every hand. It's strategically limited and really quite stupid.

Also, to say no-limit is just limit w/ more bet sizes...that's like saying chess is just checkers w/ different pieces. It's a considerably richer game which requires very different strategy. Good limit players have a completely different skill set from no-limit players.

And when I said dollar stakes, I meant you sit down at the table with a dollar. So if your program is good, it'll win, say, a quarter an hour. That's really only a problem for the guy who spent the last 5 years developing his poker bot.

HU is stupid when stacks are low. Go play deep-stack HU with a Pro, then we'll talk whether its stupid or not.

Chess is indeed much more complex as a game than Poker. No one wins poker because being good at psychology. Good strategy is the one that maximizes you equity while making least little assumptions as possible. That is what computers are good at.

Well, well, Errol. I do not believe Mig will be happy if people start advertising their favourite poker training websites here. At least I am refraining from doing so.

I've seen poker destroy too many chess clubs. (It's about to destroy another one here in Chicago.) Great game, but play it in Vegas.

OTOH, I've browsed the Harrington on Hold 'Em books: very interesting stuff.

Small world: just last Friday, I was just talking about Ms. Pinky with my son the guitarist.

golfstrom - obviously limit and no-limit are quite different and emphasize different objects. You are not making any kind of convincing argument for why no-limit should be more complicated and I haven't really seen much to suggest that it is. The primary difference, I would say, is that no limit is much more about optimizing your chances for limited opportunities at big pots, whereas limit is about death by a thousand cuts. "Playing every hand" means that you're going to be exposed to a lot more decisions, and there's definitely a ton of skill in consistently making all of them as good as you can. Low level NL against common online opponents is beatable with a pretty mindlessly tight-aggressive strategy that seems like it would be real easy to implement.

I'm pretty sure that there's a lot of evidence that bots play at pretty high stakes (15/30 maybe?) LHE games online. There certainly were a couple of attempts at single table tournament NL bots, and given the rote nature of those games I'd be quite surprised if better ones don't exist by now. And they don't have to be that good! Playing without fatigue or interruption for long periods of time, steadily grinding away, would work just fine in most instances (even if the variance is a bit higher) and would still have a pretty deadening effect on the poker economy.

My feeling is that it probably isn't that hard (compared to what is a fair question I suppose) to build a mediocre poker playing bot, and that's about all you'd need. If so, why don't we hear about them very often? Because almost everybody involved has an interest in keeping it quiet. Those operating the bots don't want people to know because they are against the terms and conditions. Those who have enough evidence to recognize the bots might take action to get them out of the game, but don't like to publicize it heavily because they are afraid it will frighten the fish. And so on.

A program I have, called Sitngo Analyzer, has some help text that claims there is an optimal strategy that cannot be beat, even if you tell your opponent how you intend to play (Nash equilibrium?). The strategy only works in no-limit, heads-up hold'em when the blinds start approaching a significant percent of the stack sizes. So, if this is the case, computers already have an endgame advantage in heads-up no-limit poker, as they can play that strategy perfectly. One caveat is that a good player can adjust their strategy when facing a weaker opponent to maximize their advantage. Most computer programs would not take this into account and while winning in the long run, they'd take longer to do it against weak opposition.

A site that's gotten a lot of attention lately is:

The guy blogging on the site, has been providing a step by step guide to writing a supposedly profitable poker bot.

Much congrats to Brooklyn player and FIDE Master Ylon Schwartz for making the final table of the Main Event of the World Series of Poker early this morning. The Final table gets played in November, but he's guaranteed $900K.

How much was his entry fee / "buy-in"? (Yeah, of course he'll come out with a mouth-watering profit...but my point is aimed at those who might feel, "what do I have to lose if I try to follow in his footsteps?")

By the way, I know Ylon and have lost money to him in chess.

By the way #2, Zviad Izoria is in Vegas trying to make a living at poker these days. (But he took a break to play in the World Open.)

Main Event buyin was $10,000.

I visited this page first time and found it Very Good Job of acknowledgment and a marvelous source of info.........Thanks Admin! http://www.bestphonelookup.com

Twitter Updates

    Follow me on Twitter



    About this Entry

    This page contains a single entry by Mig published on July 4, 2008 3:46 PM.

    Outsiders On Top In Dortmund was the previous entry in this blog.

    Last Laugh for Leko in Dortmund is the next entry in this blog.

    Find recent content on the main index or look in the archives to find all content.