Anything Poker

[firlandsfarm]

Has anyone tried using AI to play poker? Just wonder if it would fair better than the commercial bots such as Holdem Bot and if the playing style would be less distinct than such and maybe less likely to be ID'd as such and lessen the risk of funds being confiscated.

[Fugazi]

Thought about it myself. Any decent AI poker bot out there wont be sold to the public, or if it is will be extremely expensive

[wearthefoxhat]

I haven't used A.I. to play on-line poker, but, some years ago I thought it might be an idea to formulate a basic computer strategy that could run alongside the computer to help make quicker decisions. I didn't make one in the end, even though I thought it wasn't cheating as many others were doing similar things. Nowadays, it's considered "cheating" and if programmed into the web server and detected, you'd get banned from the site.

F.Y.I. I looked at the following. (Basic)

Own hand dealt, any flop information.
Personal Chip counts, and chip counts of the other players.
Number of players at the table, and number of players left in the tournament.
My position near to the button.

The excel sheet would need a lookup condition for all the poker hands and best % on the flop/turn/river. I'm not talking about modern day solvers (GTO) as this would be considered "cheating" but hard to prove if off-line. Note WSOP are changing their rules this year to stop devices being used at the table and in the crowd to assist other players.

Most sites also provide a HUD (Heads Up Display) alongside each players name. The sites don't consider this as "cheating" as it's based on whats gone before. If used properly, it can exploit a players pattern of player. ie: How many pots they play, how aggressive they have been, how often they fold their blinds to a raise...etc

https://pokercopilot.com/poker-huds#whatisapokerhud

HUD.png

If I were to re-visit it, I would use a HUD (if available) alongside the basic strategy decision maker (off-line excel sheet). The inputs would take up any time you have. so there would have to be a balance of using the time in between hands to update the information and when you fold. If you get a playable hand, then it would only be the hand itself that's input and any flop information.

[jamesedwards]

How would AI be any better than a standard poker bot? Surely the optimal action is always already known for every situation so how would it 'learn' to get better?

[Archery1969]

There are 1,326 possible combinations preflop and 19,600 combinations post flop.

My Bot and many others out there already account for these and will Bet, Raise, Call based on the Pot Odds.

Therefore, can’t see the point in using AI.

The only thing I can think of is it being used when bluffing. I use random bluffing depending on table and player play but again only if the Pot Odds are worth the risk.

[firlandsfarm]

It was never like this for Doc Holliday!

[Archery1969]

It was never like this for Doc Holliday!

[Fugazi]

There are 1,326 possible combinations preflop and 19,600 combinations post flop.

My Bot and many others out there already account for these and will Bet, Raise, Call based on the Pot Odds.

Therefore, can’t see the point in using AI.

The only thing I can think of is it being used when bluffing. I use random bluffing depending on table and player play but again only if the Pot Odds are worth the risk.

AI could take the Hud data and diverge from GTO to exploit the player further and continually self improve its tactics

But yeah, a GTO bot is enough. The computing power to use GTO AND then some would be hard to manage in inder 10 seconds

[Archery1969]

There are 1,326 possible combinations preflop and 19,600 combinations post flop.

My Bot and many others out there already account for these and will Bet, Raise, Call based on the Pot Odds.

Therefore, can’t see the point in using AI.

The only thing I can think of is it being used when bluffing. I use random bluffing depending on table and player play but again only if the Pot Odds are worth the risk.

AI could take the Hud data and diverge from GTO to exploit the player further and continually self improve its tactics

But yeah, a GTO bot is enough. The computing power to use GTO AND then some would be hard to manage in inder 10 seconds

My bot makes a random decision in 1 to 6 seconds. Apart from preflop and postflop rules. It’s never been detected across 8 poker sites in 4 years.

[Fugazi]

There are 1,326 possible combinations preflop and 19,600 combinations post flop.

My Bot and many others out there already account for these and will Bet, Raise, Call based on the Pot Odds.

Therefore, can’t see the point in using AI.

The only thing I can think of is it being used when bluffing. I use random bluffing depending on table and player play but again only if the Pot Odds are worth the risk.

AI could take the Hud data and diverge from GTO to exploit the player further and continually self improve its tactics

But yeah, a GTO bot is enough. The computing power to use GTO AND then some would be hard to manage in inder 10 seconds

My bot makes a random decision in 1 to 6 seconds. Apart from preflop and postflop rules. It’s never been detected across 8 poker sites in 4 years.

That's fast. No idea how you managed to build its decisions post flop adequately. Did you input them manually? Or use very wide broad moves. Eg. Monotone flop = 1/4 pot bet always

[Archery1969]

AI could take the Hud data and diverge from GTO to exploit the player further and continually self improve its tactics

But yeah, a GTO bot is enough. The computing power to use GTO AND then some would be hard to manage in inder 10 seconds

My bot makes a random decision in 1 to 6 seconds. Apart from preflop and postflop rules. It’s never been detected across 8 poker sites in 4 years.

That's fast. No idea how you managed to build its decisions post flop adequately. Did you input them manually? Or use very wide broad moves. Eg. Monotone flop = 1/4 pot bet always

Poker, manually or automated is based on Pot Odds, over 1,200 rules preflop and 19,000 rules post flop. My bot can make a decision in less than 1 second but I randomise it from 1 to 6 seconds which above other things keeps the poker rooms guessing as to am I a real person or Bot.

[Fugazi]

My bot makes a random decision in 1 to 6 seconds. Apart from preflop and postflop rules. It’s never been detected across 8 poker sites in 4 years.

That's fast. No idea how you managed to build its decisions post flop adequately. Did you input them manually? Or use very wide broad moves. Eg. Monotone flop = 1/4 pot bet always

Poker, manually or automated is based on Pot Odds, over 1,200 rules preflop and 19,000 rules post flop. My bot can make a decision in less than 1 second but I randomise it from 1 to 6 seconds which above other things keeps the poker rooms guessing as to am I a real person or Bot.

Interesting.

I was a profitable player but think i was only managing about £2 an hour online. Insanely better live though.

I have the poker know how. May start this as a side project its the programming side that will be a struggle knowing how to get it to respond to the poker table software

[wearthefoxhat]

Some of the RTA (real time assistance) bots out there can read the computer screen, no input required, indicate bet sizing, perform randomisation with slight delays, to mix up any timing tells. They are off-line and undetectable to the poker sites.

No wonder when these guys, when they appear on the live poker table, find it hard to adapt to the slower pace and stare downs that often occur during a game. The upside is that they've played millions of hands more than the average live tournmment player.

Sounds like Archery has got something that working and evading the web police.

[Archery1969]

That's fast. No idea how you managed to build its decisions post flop adequately. Did you input them manually? Or use very wide broad moves. Eg. Monotone flop = 1/4 pot bet always

Poker, manually or automated is based on Pot Odds, over 1,200 rules preflop and 19,000 rules post flop. My bot can make a decision in less than 1 second but I randomise it from 1 to 6 seconds which above other things keeps the poker rooms guessing as to am I a real person or Bot.

Interesting.

I was a profitable player but think i was only managing about £2 an hour online. Insanely better live though.

I have the poker know how. May start this as a side project its the programming side that will be a struggle knowing how to get it to respond to the poker table software

You need constantly updated table maps per poker site as they regularly update them if you go the full automated route.

Example fore Preflop play:

When (hand = AA or hand = KK or hand = QQ or hand = JJ or hand = AK SUITED)
When Raises = 0 RaiseMin Force Delay 6
When Raises = 1 RaisePot Force Delay 6
When Raises > 1 RaiseMax Force Delay 6
When Others Call Force Delay 6

The Delay 6 means it waits from 0 to 6 seconds randomly before making an action. That keeps the players at the table and poker rooms administrators guessing as to if you are real or a Bot.

[wearthefoxhat]

How would AI be any better than a standard poker bot? Surely the optimal action is always already known for every situation so how would it 'learn' to get better?

Overall, I'd say not much difference, assuming, the bot the player has programmed hasn't poorly structured decision trees.

If it's a neural learning bot that can adapt to a players style (normal raises/3bets/4bets/limps) there are ways to counter attack/defend against them, especially if a player keeps changing their pace.

Even the GTO stuff can be interpreted/misinterpreted in different ways. You can be right and get it wrong and vice-versa, a bit like trading I suppose.