Join 3,442 readers in helping fund MetaFilter (Hide)


I, Raise
September 21, 2004 10:53 AM   Subscribe

Are bluffing bots outplaying human poker players online? They might not be as strong as Vex Bot, but they could spell trouble for the booming online-poker industry.
posted by mrgrimm (28 comments total)

 
Part of the problem with playing poker online is that you can't "read" your opponent. Everyone's got a perfect stone face online. Online poker may be decent for learning the game and getting the lingo down but its nothing like real, live poker.

Now I've gotta consider if the other player is even real? Bleh, I'm going to back to my first internet love, surfing for pr0n!
posted by fenriq at 11:10 AM on September 21, 2004


There are two issues of concern, at least as I can tell. The first is that there exists the possibility that these bots are in the employ of casino's in order to increase rake. The second is that these are used by players to make money for themselves.

In both cases, I simply don't think it is the case that bots can outperform humans at a significant level of play. In low limit poker, a bot almost definitely could play profitably, but that's not saying much. Additionally, the rake generated by a bot playing low limit poker would not justify the risking all the negative attention a poker site would receive once word got out that they had employed bots on their sites.

At the high limits, the bots simply aren't good enough. Casinos wouldn't make enough rake to cover how much the bots would lose, and anyone wishing to use a bot for personal gain would face the same problems.

So, while it may be the case that bots are being used, right now, by private individuals for personal gain at low limits, I don't really see this as a bad thing. I'm a low limit player, but I don't think I'd be afraid of facing up against a table full of bots. The MSNBC article touts a bot's skill over 7,000 hands, which isn't a large enough sample for a clear conclusion in the first place, regardless of the limits (which were not mentioned).

The point the MSNBC article got best was that poker is very, very different from chess. There is a very limited number of moves, and only ~220 different starting hand combinations - a computer's computational advantage is not valuable in this context. What makes good players good is their ability perceive how the other players are playing and adopt the most profitable counter-strategy. Good players often "mix up" their play for this reason. Without expert input from professionals detailing how do interpret moves - Is this player who is bluffing to much a loose player, or are they a good player playing loosely, or do they have a read on another player? - I simply don't see a world-class poker bot in the near future...
posted by cohappy at 11:26 AM on September 21, 2004


You should be able to tell if its a bot if it continues to beat you with the railgun before you even see the flop.
posted by Stan Chin at 11:46 AM on September 21, 2004


I'm sorry, but I simply cannot fathom what the attraction is to playing poker online. As I understand it, there are three main joys of poker-playing: The drinking/smoking/bullshitting social event; "reading" the other guys' faces and trying to make your own unreadable; and winning money. Online automatically removes the first two, and the third is always dubious anyway. What gives?
posted by soyjoy at 11:59 AM on September 21, 2004


soyjoy, i don't really get it either, but friends tell me that it was quite easy to make a lot of money for a while. i'm not much of a gambler any more, and like you say, i prefer the drinking/toking/bs'ing social aspects of the game as much as anything else.

however, for those who actually dig playing Hold'Em for umpteen hands on end, i can see the appeal of playing multiple games at the same time.

it's kinda like asking why would anyone play Risk online? while you can't to do a victory dance in front of your opponent after you crush Ukraine, it's a heck of a lot faster and you can do it in your skivvies.
posted by mrgrimm at 12:08 PM on September 21, 2004


Online casinos have no need to develop poker-playing bots since they already rake 5% of each pot. Research into increasing the number of high-pot hands played per hour would be much more profitable.
posted by mischief at 12:13 PM on September 21, 2004


Addendum: The biggest problem with online poker is cheating via IM.
posted by mischief at 12:15 PM on September 21, 2004


There are many reasons why I prefer online poker to casino poker. Casinos are several hours away from where I live. Additionally Casinos average 30-35 hands per hour. Online tables average 50 hands per hour. I play 3 tables at a time, effectively meaning that I'm seeing almost 5x more hands per hour online than at a casino - given that I am a better player than average at my level, I'll make more money faster online. At a casino I would be expected to tip the dealer after every pot I win. I would also have a harder time changing tables. Online I can change tables in < 5 seconds.

But this is from the perspective of someone who would play at a casino. If you play poker for the "home game" scene, then none of this will probably matter.
posted by cohappy at 12:25 PM on September 21, 2004


Tangential: RL casinos do use bots to model blackjack players' strategies and try to detect card counters. I don't know how effective these are, but I've seen the bots advertised.
posted by sonofsamiam at 12:32 PM on September 21, 2004


"RL casinos do use bots"

Just wait until cards and chips are electronically tagged! ;-P
posted by mischief at 12:35 PM on September 21, 2004


Well, OK, I wasn't really counting in the casino comparison - I've never been to one, and always imagined that it was a place you went knowing that you were about to lose a bunch of money but enjoyed because the high-rolling, glamorous "maybe this time" atmosphere was so intoxicating. I can't imagine online casino play is anything like that. Anyway, thanks for answering.
posted by soyjoy at 1:33 PM on September 21, 2004


Part of the problem with playing poker online is that you can't "read" your opponent.

Au contraire, bonjour. You just have to be a lot better at reading patterns of play rather than tells. It is absolutely possible to read players to find out what type of player they are. It just takes a lot more work and concentration.

Online automatically removes the first two, and the third is always dubious anyway.

Making money online, if you've spent the time learning the game, is entirely likely (. Most players online are tourists; they've read their Barnes and Noble book on how to play poker and played a few games with people they know. Then they play the play money table and make money because that's ridiculously easy. So they deposit 20 or 40 dollars into a real account and it gets taken by those who know how to actually play the game. They might continue or they might not, it doesn't matter. This happens day in, day out.

I know of a bunch of players who just play online poker for a living and do quite well. I do it to make some side cash, but I have taken a week off work and made more playing poker than I did at my job in half the time.
posted by dflemingdotorg at 1:49 PM on September 21, 2004


Wow; I'm currently living in a room that was previously occupied by one of the guys working on the poker bot at the U of Alberta. He headed up there about two months ago, just as the two-player Hold 'Em was being solved. They're working on getting the four-player variant now. He got sick of whoring out for the thinly-veiled corporate workhouse of a doctoral program that he was in, so they sent him up North to solve poker... Hear it's a helluva good time. I'll ask him for a summary and maybe post it here later.
posted by kaibutsu at 2:05 PM on September 21, 2004


I'm a winning middle-limit poker player (routinely play in live 20/40 & 40/80 limit games, and 2/5 & 5/5 no-limit games) so I feel qualified to rebut a claim a few of you have made.

Part of the problem with playing poker online is that you can't "read" your opponent.

This simply isn't true.

Once a player get past the 'complete clown' stage of their game development, their face and body become extremely secondary sources of information. 95% of hand reading comes from watching the pattern of how the hand played, and using logic and previous observation to put the player on an ever-narrowing range of holdings as the hand progresses.

I've deleted a few paragraphs about the sort of things I'm thinking about during a hand, as it was way too long, but I'll post it if anybody expresses interest.
posted by mosch at 2:17 PM on September 21, 2004


I'd be interested, mosch.
posted by kenko at 2:20 PM on September 21, 2004


Oh, and my standard retort to those who fear collusion or bots online. If there are people cheating online, they aren't doing it very well, because it isn't stopping me from winning.
posted by mosch at 2:21 PM on September 21, 2004


Here's the UofA Games Group page... Apparently they're working on a lot more than poker.
posted by kaibutsu at 2:29 PM on September 21, 2004


If there are people cheating online, they aren't doing it very well, because it isn't stopping me from winning.

Ditto.
posted by dflemingdotorg at 2:31 PM on September 21, 2004


As a decent $4-8 and $10-20 player in real life, I've wondered about the online realm. My question to those who play is this. Since it is questionably illegal on the federal level and definitely illegal in my state, do you worry about potential consequences?

For instance, a player I play with at a local casino had all of his credit card accounts frozen and the cards themselves cancelled because he tried to use them online. Anyone else have something similar happen?
posted by karmaville at 3:00 PM on September 21, 2004


A Look At A Hand Of Holdem

before I even look at my cards. I'm thinking about the general characteristic of this particular game.
How many players are usually seeing the flop, the turn and the river?
How many bets does it tend to be on each street, and what sorts of bets are they?
Are they semi-bluffs, where players are raising draws to try to knock out other players?
Is it simply value betting, with most hands being won by showing down the best hand rather than getting other players to fold?

Things I want to know about my opponents:
Are they sensitive to position? (if you're first to act in every round, it's very difficult to play drawing hands and borderline hands profitably. If you're last to act... it's much easier)
Are they sensitive to other players? (do they make sure to have better hands against tight players? do they go for extra bets with marginal hands against loose players?)
What types of hands will they open-raise? (from early/middle/late position)
What types of hands will then open-limp? (from early/middle/late position)
What do they have when they cold-call a raise?
How often do they limp, if somebody has limped in, in front of them?
How often do they limp, with multiple players limping in in front of them?
What sorts of hands do they raise with in large, multi-way pots?
What sorts of hands do they raise with in small, two or three person pots?
How do they play their big draws in big pots? In small pots?
How do they play more average draws in those situations?
How do they play their big made hands in big pots? In small pots?
If they bluff on the flop in a small pot, do they continue it on the turn?
Can they make a big fold pre-flop? on the flop? the turn? the river? How big of a fold?
If they're shown weakness, are they likely to try to bluff the pot?

Now, I start the example hand. It's a 40/80 limit holdem game. The game is extremely tight and an open-raise wins the blinds about 40% of the time. It's 9 handed, and I'm 2 off the button.

Everybody folds to me and I have Th 9h (ten nine of hearts). It's not a great hand, but it's better than average, and with half the table having folded, I decide to raise it, and try to steal the blinds.

The button calls me, and the big blind re-raises. The button has been calling one raise with any hand that he wanted to play, and has been playing any pair, any ace, any two cards above ten, and any two cards that are connected. The big blind is most likely raising with a real hand. He knows that the two other players have to call his raise, so he's raising for value, not a an attempt to fold anybody. Additionally, since he will be first to act in all future rounds of betting, his hand must be very strong, because he won't know what we're going to do until he's already put his money in the pot. As such and based on the solid play I've witnessed, I think he most likely has a big pair from JJ through AA, or Ace King. Myself and the loose/passive player call, and the pot stands at $380.

Flop: Tc 9s 2d

The big blind is first to act, and he checks. This player has usually bet out if they had the best hand in a pot with 3 or more players, so I now think that AK is more likely than AA, KK, QQ or JJ. The only real danger now is if he has TT, but that is unlikely.

I bet my top two pair, and the loose and passive player to my left raises. This indicates that he most likely has either a pair higher than the board, two pair, or top pair. There is a chance that he has three of a kind, but again... that's unlikely and a risk Ihave to take.

The pot is now $500 even, and it's $80 to the big blind. He calls it cold, and I decide that by far, his most likely hand is Ace King. If he had a big pair, he would've three-bet to face me with two bets and try to fold me out and get the loose player heads-up. I call the additional bet. The pot now stands at $620.

Turn: Kd

The big blind springs to life with a bet, just as he previously has with top-pair, top kicker. I have trouble thinking of anything else he is likely to have. I raise my two pair, tens and nines. The loose but passive player just calls. This rules out the possibility that he has three of a kind, or that he just made a straight to the king. The best hand I can realistically put him on is Ten Nine, and it's likely his hand is significantly worse than that. Other likely possibilities are JJ, QQ, JT, QT and AT. Most other hands would've resulted in a fold or a raise. The big blind just calls my raise. Clearly he has some concern that he is beat. The pot stands at $1100.

River: 3c

The big blind checks, which with this player in this situation, definitely means weakness. I bet, and the loose player to my left calls, and the big blind mutters about how he knows he got "sucked out" and calls. Final pot is $1340.

The big blind has Ace King, for a pair of kings, ace kicker.
I have Ten Nine, for two pair, tens and nines.
The calling station has Ten Nine, for two pair, tens and nines. we chop the pot.

This hand contains two major misplays. The first is the terrible player who called a raise cold with Ten Nine Offsuit. That's just chip suicide.

The more interesting mistake is by the player with Ace King, on the flop. He bet the flop, which is a fine move. He's getting 9.5:1 on his bet, and both of his opponents are likely to fold more than 1 in 9 times, so it's a good bet. The problem occurs when I raise it, AND the loose player calls.

At this point, the player with Ace King fails to look at what hands I might have. He knows I'm aggressive, and likely to steal the blinds when possible, so I was likely raising weak, and could, before the flop, should have put me on any hand from AA through 88, suited AK through AT, unsuited AK through AJ, and suited KQ/QJ/JT/T9/98.

When I bet the flop, after having been three-bet pre-flop, it is unlikely that I am bluffing, so he should have put me on a made hand. This would narrow the possibilities to AA-88, AT, JT, or T9. When the calling station raised, that indicated significant strength, and he should have put the calling station on AA-88, or top pair, decent kicker. With both of us having hands like these, it becomes unlikely that he can win the hand. He is a 6.5:1 dog to make top pair on the next card, which seems reasonable, but if his hand-reading skills were up to par, he'd realize that he is very likely to lose when he only makes one pair here. As such, he had a clear fold on the flop.

And that concludes my random babbling about a single hand of poker (note: that was an actual hand, that I played about three weeks ago)

blinds: two people who are forced to make bets pre-flop, to start the action.
button: last person to act during the hand during each round of betting.
cold-call: to simply call (rather than raise or fold) directly after somebody as raised.
connected: when your hole cards are sequentially next to each other: e.g. Jack Ten
heads-up: only two players in the pot.
open-limp: to be the first person to voluntarily pot money into the pot, and to merely call.
open-raise: to be the first person to voluntarily pot money into the pot, and do so for a raise.
suited: when your hole cards are both of the same suit
posted by mosch at 3:24 PM on September 21, 2004


Since it is questionably illegal on the federal level and definitely illegal in my state, do you worry about potential consequences?

To put it succinctly, no.

I've yet to read of a single conviction of an online player, despite their extremely high profile nature in the world series of poker and the world poker tour. Nor have I read any accounts of issues regarding credit cards or checking account, despite frequenting online-poker related discussion forums.
posted by mosch at 3:27 PM on September 21, 2004


Anyone else have something similar happen?

At least in Canada, most banks just block you from playing. I happened to have one that doesn't and I've been playing for a long while with no troubles to date.
posted by dflemingdotorg at 3:27 PM on September 21, 2004


mosch, thanks for the thought process. It was informative and good to keep in mind.
posted by karmaville at 3:30 PM on September 21, 2004


The biggest problem with playing in a casino is that the rake is much, much higher then in online play. The other thing is that you get to play more hands.

Anyone who can program a high-quality poker-bot deserves the money, IMO. Because it's not easy. This is not a solved problem and it will probably never be solved. It's something top-teir AI researches are looking into, and havn't solved.
posted by delmoi at 3:32 PM on September 21, 2004


I note that mosch did not introduce the concept of 'slow play', how to play it and, more importantly, how to defend against it.
posted by mischief at 4:30 PM on September 21, 2004


Mosch, since you only have to show your cards if you're the aggressor, how do you learn so much about what kind of cards the other players are likely to bet with? Just from knowing that they didn't win a particular hand?
posted by kenko at 5:04 PM on September 21, 2004


mischief: the slow play simply didn't apply to that situation.

Slowplaying is only a good move in a few limited circumstances where:
1) your opponents are going to bail if you don't do it
2) your opponents are unlikely to draw to a better hand.
3) you're willing to lose the occasional pot in order to get a more frequent extra bet.

In this case nobody was going anywhere, so the first condition didn't apply.

The range of hands I put my opponents on left them with a wide variety of ways to beat me if I let them draw cheaply, so that eliminated the second reason.

And I was going to get plenty of bets out of them without doing anything tricky, which ruled out the third.

The loose player who also had ten nine could've slowplayed, but he probably would've popped the turn, and he just wasn't a tricky sort of player so it would've been extremely surprising.

The big blind could've had KK, but it was so unlikely that again, that's just a risk that has to be accepted.

Slowplaying is something that should generally only be done when your hand is so good that you have to let other people catch up a bit. An example of this would be if you have KK in your pocket, and the flop comes K82 rainbow. Odds are very good in that situation that you're the only person with a hand, so you have to try to let your opponents catch up a little.
posted by mosch at 6:16 PM on September 21, 2004


kenko: it's a mix.

The first thing you look at is the percentage of hands a player plays in a situation. If you see them playing 40% of their hands after a few limpers, they're probably adding a ton of suited cards to the mix. If they're playing 60% after four limpers, they're playing almost any two suited, any two connected, and any two broadway. (yes, this is common to see in middle limit games these days)

You then get a better read on what exactly is going on based on what they show down on the occasions that they show, and how the hands that they don't show play out.
posted by mosch at 6:19 PM on September 21, 2004


« Older Five Days in Hell...  |  U.S. Geological Survey before ... Newer »


This thread has been archived and is closed to new comments