Slashdot Mirror
Humans Dominating Poker Super Computer
New submitter IoTdude writes: The Claudico super computer uses an algorithm to account for gargantuan amounts of complexity by representing the number of possible Heads-Up No-limit Texas Hold'em decisions. Claudico also updates its strategy as it goes along, but its basic approach to the game involves getting into every hand by calling bets. And it's not working out so far. Halfway through the competition, the four human pros had a cumulative lead of 626,892 chips. Though much could change in the week remaining, a lead of around 600,000 chips is considered statistically significant.
How are we supposed to have any idea what a cumulative lead of 626,892 chips means without knowing how many total chips there are? If there are 650,000 chips then the game is almost over, but if there are 1,000,000,000 chips then there hasn't been any movement at all.
This is some pretty poor journalism.
-- All that is necessary for the triumph of evil is that good men do nothing. -- Edmund Burke
So let me get this straight. At the start of the game the computer calls all bets and then it lets the other players train it and change their strategy to take advantage of that training? And they thought this would beat a seasoned professional poker player?
This is basically a beginning poker player (fresh blood) but who is more consistent. A pro will absolutely clobber it.
Windows is a bonfire, Linux is the sun. Linux only looks smaller if you lack perspective.
I don't know about the technology or the algorithm(s), but the linked article is certainly nonsense.
“You could use the same basic framework to do robust decision making like trying to come up with insulin and glucose monitoring plans [for diabetes patients],” says Neil Burch, a computer scientist at the University of Alberta who helped design a poker-playing AI earlier this year. “You get regular snapshots of glucose levels, and you have to decide how much insulin you should take, and how often.”
Look, I get it. Nobody wants to admit that they're spending their grant money this way because it's fun to get a computer to play Hold 'em. But that's got to be the dumbest justification I've ever read. Human metabolism is complex, but the pancreas doesn't bluff.
~Idarubicin
Lt. Commander Data struggled with the intricacies of poker as well.
Gamertag: WyleType
We got ourselves a victim, get 'em boys!
"If you've been playing poker for half an hour and you still don't know who the patsy is, you're the patsy." -- Warren Buffet
Playing poker is a lot more about reading your opponent. figuring out if they are buffing or if they really have something. No computer can do that. Not at least at this point in time.
I almost did a double-take with this story; a few months ago I read about computers having solved heads-up _limit_ Texas hold’em: http://arstechnica.com/science...
Well, it looks like the computer can win when there is a limit, but humans can still win when there is no limit.
I guess that's not too surprising: the limit really cuts down the number of choices, making a brute-force calculations more practical, and brute-force calculations are what computers do best. Without the restrictions of a limit, the AI needs to be a lot more clever. I wonder how long it'll be until computers win at this.
There's no way the computer is going to win at Strip Poker.
I read the internet for the articles.
A person can read the "tells", or body language and mannerisms, of another player. The computer cannot. Presumably the computer has no "tells" of its own, but an experienced player should usually win - because poker isn't a game of math, it's a game of psychology. Or rather, it's only PARTLY a game of mathematics and probability.
And the computer here is a patsy called a calling station.
From the article: "but its basic approach to the game involves getting into every hand by calling bets".
I'm pretty sure Deep Blue lost the first few thousand chess matches, real or simulated.
And what facial expressions would the computer betray?
That's not what arbiter1 meant.
"Reading your opponent" doesn't rely on facial expressions or tells when you are playing online poker.
You brought her; you poke her.
Yeah starting out by calling all the time is STUPID. I think this game has four players, so it'll have losing cards 75%. In the first-order analysis, you should therefore fold 75% of the time. Sometimes you should raise, so you should only call about 10%-15% of the time. With further analysis it gets (much) more complicated, but those percentages are in the ballpark.
Their LEARNING algorithm might be good, it might be very good, but they're starting out with a strategy that sucks. Really sucks. They'd probably be much better off starting with a reasonably good strategy, then learning from there.
I made a simple computer poker bot which worked reasonably well, and should work much better than theirs with learning added to it. I used two phases to come up with a set of starting strategies, then played them against each other to determine the winner. First, I analyzed a few hundred thousand hands of actual online poker, ranking each player's cards for strength. I could then see through simple statistics that one should always raise preflop if your hole cards are XY-s or better, for example. That was stage 1.
In stage 2, I modified that base to create a virtual Phil Helmuth (slow plays, etc) and a few others based on actual champion players, just teeaking the basic statistical strategy to play more like the champ it's emulating. Then play the virtual champions against each other. The two virtual champs who come out on top are the seeds for a genetic algorithm to create the strategy you debut against human players.
First of all, this is the link that the story should have included. It includes updates of the scoreboard, etc. On it you will see that even though the brains are collectively beating Claudico, the computer is actually over $100,000 ahead against Jason Les, a feat that almost no human could match. Yes, Claudico is down against the other three, but these are the top players in the world, and most human pros would get clobbered much worse by these guys. Are we really so hard to impress? This is the first time that something like this has been tried, and already, the computer is performing on a level that most poker pros would love to reach.
Ummm, wrong. It's the other way round. You don't put shoes on by keeping your feet warm, do you?
At the bottom of the
if I understand, it is playing 4 heads up matches simultaneously. It is losing in 3 of the matches, and winning in 1. To even be winning in 1 is pretty damn amazing though, as I'm sure they don't lose to anyone who isn't damn good. It also looks like a second match could swing back its way. Nothing certain as they are only half way done, but this computer is doing damn well. I'm sure if I played 20 top level players in heads up, no limit, I would lose against all 20 by a pretty damn large margin.
Humans were designed over a long, intense period of selection to be able to perform deep, self-referential thinking, ie to be able to know that the other guy knows that they are bluffing/lying/have some particular knowledge, and make plans accordingly. It is probably the single task that humans are best at. I'm not saying that having a machine beat humans at poker will mean the Singularity has arrived, but when they go one step farther and start beating us at politics (which requires the same skillset as poker but with more complexity, plus the ability to integrate other types of knowledge), then it's all over.
I must say that I am looking forward to it. Humans are much better at beating others in the race to gain power than they are at actually ruling and making decisions for other people.
or did anybody else picture Data playing poker with the command crew of the Enterprise? Thanks for another smile Mr. Roddenberry :)
"A mind reader? That sounds like sci fi." "Honey, we live on a space ship"
A naive strategy that would beat most non pros would be as follows:
At each round of betting, evaluate how many stronger hands there are than your cards and how many weaker hands there are. If there are N other players at the table, you should bet if the number of stronger hands divided by the number of possible hands is less than 1/N because you should assume that whoever has the strongest cards is still in the game, so the strongest opponent is the strongest hand of N random hands.
> I'm sure if I played 20 top level players in heads up, no limit, I would lose against all 20 by a pretty damn large margin.
Perhaps. Actual tournaments aren't generally heads up, they are 8 to a table, with as many tables as needed. In these actual tournaments, you can easily "beat" over half of the players by simply folding anything but the strongest hands. Source - I've actually done this in WSOP events. You can get "in the money" (win prize money) by simply playing super, super tight, folding 90% of the time or more. Winning first place requires switching to a strategy of taking far more risks. Not unlike investing, actually- most millionaires get there by investing in low risk, broad based mutual funds.
In chess, you CAN beat half the grandmasters while playing heads up against each of them. Simply copy the your first opponent's move against your second and vice versa. They end up playing each other, with you as the messenger.
So, one Claudico and four humans. The odds are already stacked against Claudico 20%-80% that a human will come out tops.
I wonder if the outcome will be different had there been more than one Claudico at the table.
Yeah that's a restatement of my first-order analysis. At a table of four, you'll have the best hole cards 25% of the time, so you should fold if your cards aren't in the top 25% of possible pairs. There are patterns you can learn to know approximately how many hands beat yours.
Further analysis brings out the fact that sometimes you can call cheap, get a good flop, then have someone bet large into your strong hand. So if you're playing against a table who bets small preflop and large postflop, you might call more often. On the other hand, if there are several potential raisers behind you, calling the one bet might not let you see the flop without calling more, so you should fold more often rather than getting stuck between multiple raises. It just depends on who you're playing against and your position relative to the button.
Comment removed based on user account deletion
Comment removed based on user account deletion
Comment removed based on user account deletion
Comment removed based on user account deletion
Comment removed based on user account deletion
Comment removed based on user account deletion
I was speaking preflop. Post flop is a whole other ball of wax. Post, you have more information not just about the cards, but from betting. So you can then analyze what the nut is, and who may have it (from their betting), and what it might cost you to showdown.