Perfect Poker Bots

[Larry_Csonka]

Seeing how it seems limit poker is a very mathematical game. I beleive it relies mainly on 3 virtues.
Patience
Math Skill
And the ability to rate your opponents.

All 3 of these skills a computer would be the master at. Using a system like poker tracker has to rate the opponents and adjust play accordingly.
So, My question is, does anyone else beleive it would be possible to make a computer bot that can play limit hold 'em as good if not better then any real life player?


PS. I may start trying to program something like that soon, if you interested in helping PM me.

[kasey2004]

I dont think u could make a "Perfect" bot better than any player.
I don't doubt it would be possible to make a program like this using player info and odds for limit poker and that it could win money ( just not better than any player "Perfect").

[img]/images/graemlins/spade.gif[/img] Kasey [img]/images/graemlins/spade.gif[/img]

[Amerretto]

Agree totally, in a heads up at the moment the perfect player would win every time.

In a ring game though, I reckon a good bot could make a lot more money especially in low limits, basically capitalising on players tendency to misplay hands.

Only concern is that for a bot to get a good read on a player he would need quite a lot of hands against him, and sitting on a ring game, losing players don't stick around long. Tie it in with pokertracker stats, and you may have a winner, especially if you can tie it in along with table selection and buddy lists.

[troy5168]

along those same lines I believe you can teach a chimp to play NL just as well as most internet players. They already play like a bunch of monkeys, lol

[sully4321]

[ QUOTE ]
Agree totally, in a heads up at the moment the perfect player would win every time.

[/ QUOTE ]

perfect player would not win everytime.

ex.: if you go all in with 72o and he calls with AA -- mathematically knowing he has the best hand -- is it possible that the board could come 3456X? or 777XX? or 77XXX? or 22XXX? or 89TJX? or..? (i think you get the point). there is something called Luck in poker, so no possible way to be a stagnant winner. you can win more than you lose, but nobody is "undefeated" in poker.

[PhantomeX]

[ QUOTE ]

perfect player would not win everytime.

ex.: if you go all in with 72o and he calls with AA -- mathematically knowing he has the best hand -- is it possible that the board could come 3456X? or 777XX? or 77XXX? or 22XXX? or 89TJX? or..? (i think you get the point). there is something called Luck in poker, so no possible way to be a stagnant winner. you can win more than you lose, but nobody is "undefeated" in poker.

[/ QUOTE ]

he original post was speficially postulating a limit poker situation. Anywhere that the player can control the pot odds to a nearly unlimited degree (NL and PL games) the bot would be easily beat out. And in structured limit games there would be opportunities to beat it out by manipulating pot odds but it wouldn't be enough to offset the "perfect" play of the machine.

By the way anyone who thinks that a computer cannot perform mathematically perfect play has no understanding of how computers work. Anyone who wants to seriously discuss the possibility that mathematically perfect play would be below average in terms of $ won per hand average over a statistically meaningful period I would be willing to explore that possibility my "math gut" tells me that it would be signifigantly above average considering the "typical" online players it would be facing, but I haven't actually done the functions up to say that for sure.

Regardless I can be absolutely certain that a pro that knew he was playing a bot would in short order be able to walk all over it, although if you added in some variance to "perfect play" via a hardware RNG and some serious player tracking it could probably hold it's own against even some of the very strongest players at times, it would most likely go back and forth with the strongest players pulling ahead eventually. Now where it would run into some problems is if it ran into an entire table of people who knew it was a bot and worked together to take it down (not actual collusion though that would obviously work also) in the same way that the "crazy better" that bets at every opportunity might be able to walk all over an average to weak player in headsup play (in limit hold'em) he will always loose to a table of solid players.

[Amerretto]

[ QUOTE ]

perfect player would not win everytime.


[/ QUOTE ]

Apologies, didn't explain myself properly, of course a bot can get sucked out on, but at low limits, if the bot can play mathematically correct, then it will beat the low stakes game in the long run, capitalising on LL propensity (myself included) to call to far. Agree though with later posts, if a decent player knew they were playing a bot then, sure the bot would get beaten.

[evain]

I believe winning bots are possible (perhaps even reality) but a perfect bot? No.

[primetime32]

i think a bot could be made that would at an incredible rate. however, it can never be perfect since it would be gathering information and basing its decisions on imperfect players that are inconsistent.

What if it takes 300 hands from a player that was on tilt that day and then basis its decisions on those 300 hands that arent reprensentitive of that players usual play?

So the answer is definitely not.

[pokerjoker]

Ive been thinking about a decent ammount recently. They say poker is about making decisions with imperfect odds. I think what makes someone a good player is they play with more "perfect" odds.

hypothetical example. You know from poker tracker that when someone raises preflop 6x the big blind that
25% they have kk
25% they have aa
30% they have aks
20% they have two suited connectors below q

you have qq

the bot would figure out the implied odds for this situation based on previous plays from opponent and determine if calling would +EV. and would probable depend on how far the opponent will go against a set of q's but if you knew the percentages for everything to occur this could be an exact calculation. You would need a HUGE sample set for a decent opponent however and the opponents sample would have to be in sync with his current playing.

The difficult part to program in my opinion would be two things

1. Making moves...this could be done based on opponents previous loses to say check raises vs just betting out but would be very difficult to program to the level of intelligence of a good player.

2. THE BIG PROBLEM....this might not matter in lower limit games but the bot has to account for the opponents perception of it. Has it been playing tight recently, loose? Is the opponent paying attention to this?

I could see using random numbers to vary play slightly but it would be VERY difficult to make a bot unpredictable enough to win money against decent opponents.

Could you just make 100 of them and set them loose at $25 tables? Maybe.....