I would be surprised if the machine is beatable long term. One of the bots folds, but gets replaced by an advanced bot after so many hands. Only way to have a chance would be to adjust quickly to each bot. One bot we can call FOLD BOT and another we can call BET BOT. I know fold bot is beatable with aggression, but Bet bot you would need to calculate your equity on every street and still its going to be thin. Oh yes, Mixed in with Fold and bet bot is a bot I'm going to name tricky bot. Tricky will mix up everything rotating a small number of hands where fold or bet will play the same style for a while. You will have to know the equity of every hand you play (especially the weak ones) against current bots range in every situation. Again I apologize for the grammar. Good luck

I'm new to this great game. Can a bot be designed to play equilibrium in every situation?

It's on the horizon. Take a look at this thread. The stuff directly relating to your question begins at post 62.

Relating somewhat to a point Phil makes above, there's a thought I've had more than once that I haven't yet seen discussed in any of these threads:

It seems to me that it's possible that there are strategies that many of us would characterize as "bad" that actually would perform better than the strategies most of us would normally employ. I'm talking specifically about the case of a bot that makes no adjustments to its opponent, whether it's a GTO approximation or not.

For example, think about a TAGish player. I think most of us who post here regularly can beat this guy pretty easily, but we have the luxury of adjusting to the fact that his postflop ranges are more narrow. So a player who's ranges are appropriately wide could in theory have a harder time against the bot than a TAG player.

Obv it depends on exactly what his ranges are as the extra bets he wins postflop will be balanced out by the loss he takes by sacrificing too much equity on the small-bet streets, and I'm not necessarily suggesting that a TAG strat would be +EV against the IGT machine or any other bot. What I am saying is that this TAGfish player might lose less than a better player, precisely because the bot's play essentially presumes a better opponent with wider ranges.

I think u are confusing different concepts here. What makes any player with TAG stats easily beatable? You could argue that they fold too much preflop or dont open enough, which will be decent sized mistakes depending on the situation.
But contrary to popular belief (i guess) a smart TAG is no easier to exploit postflop than a smart LAG. Whatever range u take to the flop can be perfectly balanced both for every turn/river to come and for any corresponding action of our opponent. Having a wider range will only give us more combos of hands to pad each part of the decision tree with.
True that it will also make it easier to fine-tune balance each decision point but at that point we are splitting hairs pretty much.



and also, just to add.. I think most humans today have a very long way to go until they are even close to an unexploitable playing style. And that a lot of the "weird" stuff bots do only look weird to us humans because we have evolved into playing LHE in a way that takes a ton of shortcuts. Which makes total sense for us to do btw with our brain having so limited computing power. Meaning simplification even if they come at a cost will make sense to do a lot of the time.

Yeah, we're speaking in such generalized terms here but basically I'm saying if a guy folds too much pf against a good player, there's nothing he can do postflop to make up for it. I'd also argue that the fact that a player makes such pf errors argues against the "smart TAG" argument you provide below, i.e., typically the same guy is going to fold too many flops, etc.

The thing to remember is that we're discussing a bot that never has the option of analyzing the fact that a particular opponent is folding too much pf. The bot in no way seeks to exploit this behavior and in fact what I'm arguing is that its inability to adjust works in the TAG player's favor.

The good human player however will be able to make such adjustments, so he might outperform the bot in terms of wr, at least against certain opponent types. This isn't ofc my main point but it might serve to better illustrate what I'm saying above.

Maybe a different way of making the same point: The University of Alberta has an algorithm to come up with a perfect exploitative strategy to beat each of their bots. By their own admission, the resulting strat is likely useless outside of that context, i.e., even though it beats one particular bot for the maximum, it doesn't rate to do very well against a wider range of opponents.

ok, yea think i agree with that
Although its not completely black and white here. Smaller mistakes preflop like folding marginally profitable hands could easily be made up postflop a lot of the time. But yea against expert players then it becomes tough, true.

Yes definitely! A smart human player will always outperform a GTO bot vs less good players. As a GTO bot per definition doesnt care about his opponents leaks. While a smart human will and then exploits accordingly.

And yes u could also then say that any player that has a leak (as in deviating from a perfectly balanced style) will then be spared a lot more by a GTO bot than a player looking to exploit. He will still have a negative winrate, but significantly less than vs a player that managed to take max advantage of every imbalance.

So then going back to my larger point, the coup de grace is that if we were presented with a strategy that succeeds in exploiting the IGT machine in one or more ways, it may have traits that those of us who have never played the thing would find suspicious, either because we would recognize the strategy as being generally exploitable or because we refused to believe that the bot could be exploited in that fashion. The strategy would look "bad" to us even though it might work better than the types of strategies we would generally advocate.

I'm not saying that's what's happening itt obv.

yes, i would for sure expect that the optimal strategy vs the bot would be something that looked funny to a lot of ppl. Also, if the bot was tweaked off of GTO on purpose it would make sense for the owners to tweak in a way that exploited how most humans played.

this is true c-bets in position and raising from its sb 65-75 percent of the time is going to beat most players. If it bets from first position most of the time it has a hand worth showdown value.

Could we re-raise half of the sb raise range to counter the computer value raises? Computer raises 70% I re 22, A2, K7o, K4s, Q9o, Q8s, JTo, J9s giving us above 50% equity on our re-raises. When the aggressive BETBOT is playing I have thought about folding slightly more pre and making the money from it overplaying its hands against my stronger range. Computer does not adjust. Changes the gears by having multiple personalities. Once again I'm new to heads up.

The simpest example would be limping on the button aginst the machine.

I was wondering if you would chime in here since we know you've spent a decent amount of time playing these things. Since my posts were kind of theoretical, I'd be interested to know if you're saying your experience suggests a limping strategy is in fact an effective approach versus the IGT machine.

Also, would you care to offer an estimate as to the frequency with which it raises the bb in response to an open limp? We have a few reports (one from a HS pro) estimating it at about 75% but a published 2p2 author saying he doesn't believe it.

I haven't read the whole thread, but I have a few comments. You said you don't want to be raising into it when it has shown strength, but it shows strength every hand. I have played many hours on it. I did well at the beginning playing 2/4 on it. I had a day as big as 1600 playing a couple hours. Now, the machine will hardly ever let me get ahead. I took some pictures of some hands that the only conclusion I could come to was that it takes into account my whole cards.
EX: I chk raised w/ 45 on an A33 board where I had flsh draw. The computer raises me on the turn. The river is check-check and computer has J9offsuit. I found that I had the worst results when I tried to bluff it.
One thing I also tried was short-buying, so that it could not bluff me out of pots. It is unreal how many times his holdings win when I am going w/ nothing worse than Q high preflop. I feel it just decides when to get lucky and does.

I would try then to play more passive with marginal value to let it barrell off his junk with tiny equity. And also to delay pulling the trigger with draws on drawy heavy flops until turn on river (ofc u need to take into consideration turn/rivers give u legit hands u can actually rep something on).

Maybe try to bluff with some less obvious hands yourself and then keep the pressure when ur fake outs get there and give up when the board bricks.

Seems to me the bot has some explotive strategy built into it, but these hands arent enough to say that for sure ofc.

Not bloating the pot too much early should also be good for u especially OOP with less than strong hands if he plays really aggro and isnt dumb about it