Interesting article
I play online poker sometimes, and I’m not very worried about bots. I believe that writing a program which can actually play poker well against the range of players you tend to find online is incredibly difficult, far more difficult than most people realize. One could write a bot that only played very good hands, i.e. taking advantage of the computer’s infinite patience. Such a bot would probably consistently make a profit at low limits, as long as it changed tables fairly often, but that’s not exactly devastating – there’s lots of people who play much the same way, and they’re not hard to spot and play against.
Something which was aggressive and unpredictable like the Vex Bot would be problematic if it could be written to handle more than one opponent, but I think that’s a huge if. Heads-up poker is a completely different game than poker against several players, and what works against a single opponent will bankrupt you against five.
Of course, these opinions are just based on my own gut feelings and meager experience – I’d be interested to hear what experts like Sam Stone think about it.
Based on his very detailed and logical strategy advice to Otto I suspect Sam is an AI poker bot.
If there aren’t, there will be soon. The thing is, you don’t have to make a phenomenal bot to make money at online poker. You just have to make one that’s better than the average player. As the popularity of online poker continues to climb, the incentive will continue to increase and the difficulty of building a profitable bot will decrease.
Personally, I think it would be really cool to test my bot out against other bots. More fun than playing poker, anyway.
By the way, I loved this quote from the article:
So… he’s angry at the online casinos for enforcing their policy against him, because they don’t stop other, possible abusers. It sounds to me like kicking off known bots is a pretty good first step. What he’s really mad about is that the other bot-builders are better than he is.
For some reason, I doubt poker bots are going to be a big problem.
Oh… I know the reason.
Cuz it’s far, far easier to just write a poker room web application that stacks the deck in your favor. Just slightly in your favor, of course, so that it doesn’t appear suspicious.
How would someone use this “web application” in one on one online poker play?
You misunderstand. You don’t play poker against the house, so they don’t rig it directly against you to take your money. The article wasn’t about the house employing bots, but other players, unaffiliated with the house, using bots.
In any case, it’s not much of a worry. It’s extremely difficult to write bots that play good poker… it’s easier if you’re not too ambitious and you just want to beat really amateur players at very low limits, but that limits the potential profitability and therefore desire to spend all that time writing the bot in the first place.
Well, there’s nothing that could stop them from using bots the same way they use live players…to take back pots as sheer profit, on top of whatever rake they charge.
Personally, I don’t know if there is a bot that could realistically regularly beat a player of even intermediate skill. I’ve done a bit of research into Poker AI (might well end up taking my CS grad thesis/problem report that way), and it’s not really as cut-and-paste as chess would be, since unlike a game like that, there is incomplete information available to the program, and with so many possible variables (position, player tendencies, current bets on the table to name a few), it’s hard to create a really good algorithm to simulate the skill that a world-class poker player has. U of Alberta definitely are the leaders in the field, so maybe they’ll be the first and end up scooping online poker like the MIT Blackjack team…but for now it’s not a worry.
I know the guys in the article very well. Played poker with them for years. Both are excellent players, and the U of A has been working on its ‘Pokibot’ for a long time. This is the AI group that made Chinook, the world checkers champion, so they know what they are doing.
Poki is very, very strong heads-up. That’s because the game can be played with game theoretic strategy. Poki is not very strong in a ring game, from what I understand, but it’s getting better. Ring games are very, very difficult to analyze compared to heads-up. Players are modifying their strategy not just on yours, but on the other players, and also perhaps on how you are modifying your game to other players.
I’m worried about bots in ring games, but not bots like Poki. The bots to worry about are the ones that can drop into a game in pairs and transmit hand information to each other, so that one human who knows what he’s doing can clean up. Just knowing two other cards gives you a huge advantage over the other players. Apparently, there are bots commercially available that can be played in a ‘team’ mode where they share card information. That’s scary.
That’s true, I was just pointing out that it wasn’t really the topic of that article.
Vegas really started to take off when someone finally figured out that you will make much, much more money in the long run by running a clean game than you will be cheating people. The woman who is in charge of Paradise Poker is a friend of a friend and she is making so much money that it would be beyond idiotic for her to fuck things up by putting bots into play.
Back in the Usenet era, one of the rec.gambling.poker guys wrote r00lbot and set it loose in a long running on-line poker game. This game was not played for real money but the play was pretty good compared to anything else on the Net in those days. r00lbot consistenly won. The creator always said that r00lbot was not all that good and only did well because of the poor play of others.
I believe that a not very sophisticated bot could grind out a profit at low limits fairly easily. There are geeks out there that would do it for fun so it would be worth their time because it’s a hobby for them. A solid player has nothing to worry about when it comes to them.
Sam Stone brings up an excellent point. Two bots colluding could gain an edge that would be very profitable. Supposedly, the online rooms have anti-collusuion detecting software but I don’t know how effective it is.
Haj
pokibot used to make a ton at FIPS, but the play was pretty bad.
remember this; “pokibot, tell a joke”
Haj, I always thought the investment group behind Paradise was a big mystery. Am I wrong about that?
Do we know who runs it?
It’s always been so well run and the software has always been first rate, it was clear there were some brains behind it. I always suspected a medium-size coorporation put it together.
I didn’t realize it was such a big mystery. I don’t know who runs it but my friend is definitely in a position to be in the know. From what I understand, it’s just one person in charge and no more than a small group involved. They contracted top knotch programmers and compensated them very, very well. Their fixed costs are like nothing and they are making a mint.
Haj
Help me out here, Sam Stone. I read the MIT book about the humans who counted blackjack as a team and took different roles to mask their intentions (the “gorilla” and so on).
If my bots are A,B,C, and D, wouldn’t it be more profitable for at least one of them to “take a fall” every now and then, losing a small pot to string the human along? A, B, C, and D are all sending information to each other. The bot with the best hand relative to the others “calls the ball” and wins the hand; the weakest folds right away; the others bet realistically, or bluff the humans off the table.
That way, you could get a sequence of hands like this:
- A wins 50
- C wins 50
- Human wins 10 because ABC all fold
- D wins 35
- Human 2 wins 15 because BCD all fold
- A wins 50
…
A+B+C+D: up 300
H1 + H2: down 330
Realistic? Any scarier than your two colluding bots?
I was playing on PartyPoker.com and doing pretty well in the play-money area. I decided to make a deposit and play a few hands for real money. I had both a full house AND a straight to the ace get beat by four of a kind. Now playing for play-money I see four of a kind at most twice a day, that’s playing for a good 6 hours, almost every day, and in play-money more people tend to stay in the hand. Both four of a kind filled up on the river card. OK, it’s not impossible for two four of a kinds to show up in a half hour, it just seemed rather odd to me.
What’s your point? Are you implying that the real money games are rigged? If they were to take the time and effort to rig the game, do you really think that they would make it so obvious? I mean if they’re continuously spitting out quads someone is going to figure it out.
Haj
Yeah, there’s 4 of a kind winner on every hand, that’s what I was saying. :rolleyes:
I’m just saying it was odd that the frequency of 4 of a kind hands increased dramatically when I played for real money. It was all in the realm of possibility that two 4 of a kind hands would show up in less than a half hour, just highly unlikely. I found the coincidence of being beat by 4 of a kind, twice, in less than a half hour to be baffling. Probably just a streak of bad luck, but it will probably keep me from playing PartyPoker.com for real money ever again.
You don’t have enough experience at their tables to gripe that they’re fixed. That said, I might well have acted the same as you did.
This “stringing along” would be almost pointless since there is basically an endless stream of human players into and out of online cardroom.
Also, it doesn’t realistically represent how the money moves at a table. A human is going to win pots without the bots throwing one to him. He’s just going to win a little less because the bots are able to throw some hands away that they otherwise would have had to pay off. It’s a subtle way for the “mark” to lose his money.
And Sam isn’t even talking about bots playing at the table, really. He’s talking about 2 GOOD humans (or 1 GOOD human) playing with knowledge of 2 hands – and he’s using ‘bot’ loosely to describe a device that just gives the humans the knowledge of the 2 hands. This is going to give the human – not PERFECT information – but MORE information than you have. A chump probably can’t do much with this knowledge, but it’s going to increase the good player’s edge every so often (how much of an increase and how often is hard to figure).
That type of collusion is definitely profitable, hard to detect if done well, and probably already being done.
When you start getting into 4 bot territory, you need to start worrying about the rake, the decreased number of marks, the increased difficulty of processing the information, and possibly a greater chance of detection. What would happen, conceivably, is that your optimal play based on knowledge of 4 hands would be so far from a “standard” play that a card room might flag you.
For example, with 2 hands, an optimal play might look strange to the cardroom, but might still be well within the realm of normal play. For instance, you choose not to draw to an inside straight even though you should have. “no big deal”, they say, “lots of players don’t draw to an inside straight” as a matter of principle.
Now, with 4 people let’s say you choose not to an open-ended straight with nut flush possibilities because you know what 4 other hands are. That’s such an egregious mistake for a normally profitable player, the card room just might start looking into your hands.
Some people may have done real studies on “optimal collusion” but I think you’d start seeing diminishing returns with more than 2 or 3 getting in on it.