Quote:
Originally Posted by ColliePoker
There are a lot of useful lessons to be drawn from the competitive chess community, I think.
First of all, I think stopping cheating in online poker will run into the same issue as stopping cheating at a high level in online chess. People have been saying that GTO Wizard could merge with Stars and other sites and then flag cheaters. Chess.com bans people all the time for playing so many high accuracy moves that the odds of it being human at their recorded strength are statistically too small.
At a higher level of cheating, Grandmaster chess players play as themselves for 90% of the game and then for 1-3 of the trickiest moves, they rely on a computer. This makes it virtually impossible to prove they cheated.
You can imagine all kinds of similar circumnavigations with Poker. I play 200nl and 500nl. Mostly I know what I'm doing. If every 1000 hands I used a solver, this would be impossible to prove as cheating.
Not to mention, let's say the solver says bet hand X for 25% pot 1/2 the time. Well first of all, we don't know if I'm randomizing exactly 50-50 or not, so it's tough to say if I'm following the solver output. On top of that, what if I bet 27%? You'd need a very smart AI to flag that.
Or, what if I'm a smart player and I look at a solver output and I think "Okay, looking at this output, I notice that it is extremely difficult to find natural bluffs for my opponent. Since I am looking at this output, it is very easy for me to get a concrete sense of how much value he has compared to bluffs. Ergo, I will pure fold this hand that actually I bet the solver is mixing. Or even pure fold this hand that is really a pure call."
Similarly, let's say the solver likes specifically the busted flush draw with 86 of hearts. All busted flush draws bluff at some frequency, but it pure bluffs 86 of hearts. Who cares! I look at the output, I see that busted flush draws bet at 70% frequency, so I randomize my specific busted flush draw and bet it.
People are obsessed with blockers but they are very little of the winrate. Mostly it is about not overbluffing or underbluffing backed by a reasonable amount of common sense. Back to the chess analogy, I don't have to make the best move every time. I am better than 99.999999999% of humans if I just never blunder.
And a blunder in this case would be something like bluffing with Ace high when there are way too many weaker hands to bluff with.
As cheating becomes more and more of a part of online poker, so will catching cheaters. Actually cheating in chess is somewhat irrational in the sense that if you are already eking out a living playing chess professionally, you are risking a great deal just to win 1 tournament by cheating. Just being suspected of cheating can be so damaging to your career. If it was ever proven, you'd be completely ****ed. If these sites invest in anti-cheating technology, there will still be cheaters. But hopefully few enough at the highest level to kill the games. And then at the lower levels, hopefully enough people that just totally suck at cheating.
The last five years, it seems like online poker sites have not truly prioritized the longterm sustainability of their market. Disaster will be if they simply can't bother to properly fund anti-cheating measures. One thing we have going for us is that recreationals and professionals probably hate cheating a similar amount. Online sites have failed to cater to pros which is short-sighted in the sense that people get into poker to become pros. If people stop aspiring to be pros, their market will die in the long term. In this case however, they might be worried about their market dying in the SHORT term. So, I'm like 80% confident most sites will try their best to keep up with anti cheating measures.
Hopefully that wasn't too ramble-y
The outcome of a chess game can often be determined by one or two critical moves. Poker is more like thousands of small edges adding up.
GTOW AI can already solve for whatever bet sizes were used in game.
Also, unlike chess, RTA-detection in poker relies on a large sample of hands. You have to measure aggregate stats. The cool thing about GTO Wizard is that they're building GTO reports, so you'll be able to measure things like total betting volume or how much EV you allocate to various lines.
Reports like this could potentially act as flags to identify superhuman play when analyzing a large sample of hands. It doesn't really matter if your particular dream machine uses 27%/71%/120% or 33%/66%, some of these stats are invariable between different solver algorithms and betting structures.
I also think it would be important to filter for the most important spots. For example, looking at just the stats of just the 30bb+ pots, or examining the stats of only 4BPs, etc. Perhaps filtering out hands that were played against obvious whales where we'd expect someone to deviate.
Last edited by tombos21; 08-09-2023 at 12:51 AM.