Poker AI

[M.B.E.]

There were a couple of interesting threads at the Zoo speculating about bots in the 1/2 games at Pokerroom:

http://forumserver.twoplustwo.com/sh...202#Post401202

http://forumserver.twoplustwo.com/sh...;Number=413073

[esknights]

Let's say this whole Poker Bot thing starts happening for real. Can sites such as UB and Party stop bots or will they even know if Bots are being used?

[jdarwin]

One possible way in stopping bots is by popping up messages that confuse or mess up a bot's current status. This would be annoying but it is still a preventable measure against bots.

However, I could certainly program part of my code that allows me to search the screen and detect anomolies. If found, it would autoclick the region to get rid of the box, to continue playing -- thus it's not foolproof.

Other methods including measuring how much a person plays online - for example I've run my bot on Party for atleast 2 days straight... but no one has complained or emailed me about it. Unusual ... yes...

But it's not that wierd - If the operators of such sites see that players are playing for weeks on end without disconnecting one could say that a bot could be employed...

However, I could still get around this by allowing my bot to play on certain times of the day (b/c the night is juicier than the day) and I could disconnnect my bot for several hours of the day and then reconnect.

overall there isn't really a way to stop a bot... b/c if there's a will there's a way = the hacker mentality.

-jdarwin

[JellyFishy]

It would be foolish for the cardroom to devote energy to reduce it's own revenue. Bot's are effectively players, bad bots = bad players, ok bots = ok players, good bots = good players except they play much more often and help fill the tables.

A cardroom can still maintain its integrity and say that they don't use bots to enhance the efficiency of their cardroom, but that doesn't mean they can't benefit if there are artificial players induced by a third party.

Besides, it's difficult enough to become a winning human player, much much more difficult to become a winning bot player.

Don't reply to me, I won't post again.

[JellyFishy]

Take away the ability of the human to use intuition, psychology judgements, and feel to play. Worst than the unimaginative rock.

Then this task becomes much much more difficult than to be a winning player yourself.

Don't reply to me, I'm outta here.

[jdarwin]

of course i'll reply [img]/images/graemlins/smile.gif[/img]

Poker is not an intuitive as people make it out to be.

If a neural network can stimulate and figure out that a player will Fold, Call/Check, or Raise -- 85% of the time what does that say?

That says that bots can win in the long run... coupled with the fact that it can't go on tilt, it doesn't give a [censored] about losing to good cards, it can play "forever", and it practically can't be detected.

Instituting random play isn't hard, opponent modelling has already been done, and now the programming and bots are coming.

The papers have been written folks. All that's left to do is to take them and turn them into working poker bots -- which I guarantee there are out there.

I'm practically 75% done with mine... all that needs to be done is add opponent modelling and bluffing and bam... money in my pocket.

go figure.

-jdarwin

[karlson]

I call BS.

The fact that the NN has 85% accuracy does not mean that it's going to be a winning player. I'm guessing here, but I wouldn't be surprised if you could get something like 70-80% by just folding when you have nothing, raising with a big hand, and calling with the rest (i.e. most decisions are easy). And this is almost certainly not enough to beat most games.

Post some results of your bot, maybe even some hands that it played. That will be interesting to see.

I'm sure your bot can beat the really soft games. I think Poki will probably beat the really soft games (I don't know if those kinds of results have been published). But I have yet to see something that can beat a game where the players use their brains more than occasionally.

[JellyFishy]

Email me if you dare:

NA2888@yahoo.ca

I'll tell you privately why your ideas don't make sense.

[Sixth_Rule]

I do agree that a program could be made to play agaisnt and beat low limit hold'em BUT i don't beleive that it could ever beat any skilled player or group of skilled players, especially short handed. The reason is this.

I jump on to an 10 player empty table and someone else and a player joins me, i quickly notice the player will check call anything and only bet if he/she has somthing, I lose 10 BB learning this info and then start to exploit it praying that the table stays heads up. I win my 10BB quickly but then 2 other players join the table. watch a couple of hands and started playing. 1 played all hands preflop, rarly raised but would fold post flop with nothing and raise with top pair. the other was tight passive and could be bluffed.

Now the trick is not to look if you put a bot in my place but see if you could predict my play as any of these players before i adjust.

for 7 hands i played like a calling station heads up just to see what my opponent was betting with. and even reraisd with nothing to see his reaction.

afterwards i am pushed out by any bet of my opponent and bet anytime i have anything and my oponent checks or calls.

when two other players join i bet no matter what vs 1 of them and call and raise alot of preflop situations that will put me heads up with the guy who can't play preflop.

NO AI COULD FIGURE OUT WTF I WAS DOING WUSING THE NUMBERS.

i might play a hand with 72o from UTG one time and show it just to scew evryone up.

so there is AI will have one of these two problems vs any specific human player that dosn't play like a bot.
If it collects data long term and uses it vs a skill player it will not notice any short term varience in the players game and usually that will be enogh of an edge for any skilled player. IF I NOTICE THAT THIS BOT OR PERSON ALWAYS CALLS MY bluffs i will bluff less and make it pay, and soon and if it has recorded 20 000 hands that i have bluffed less often and it won't stop calling my good hands until the statistical numbers tell it too wich could take forever.

2. if it uses only sessional data it won't have enough info to make any good assumptions on my play

And the one problem AI will always have is it is a BOT
if it always folds Top pair top kicker to 3 bets you better belive a skilled player will figure it out quickly. the bot may eventually, after a number of losses, figure out that it needs to start calling it down but the first time it does the player may adjust and make it call him down when it really is facing a set or better.

you could make a machine that makes these human dicisions quickly but it is not trivial and the machine would have two little info to make them correctly by the numbers

[brianmarc]

Why would anyone go to the trouble of building a bot and then play it in unfavorable situations (SH and against experts)? There are many more weak than strong players, esp. in low-mid limit games. No reason to go anywhere elsa. BTW: Same applies to humans; why not look for the easier games and feast there?