[article] How does Artificial Intelligence learn to bluff at poker?
I have been a recreative player since 2008. I've spent so much much time on this forum. I was a big fan of High Stakes P
How does Artificial Intelligence learn to bluff at poker?As an answer to the main question, I would say that good poker AI doesn't necessarily need to learn in real time how to bluff. You (the programmer) can teach it how to bluff and do it in a tricky way, so that it's not easy to discover a solution. If you program AI which is learning during the game, you can as well program
It seems you have not yet understood on what principles are based Poker Artificial Intelligent agents.
It seems you have not yet understood on what principles are based Poker Artificial Intelligent agents.
Ok maybe, but AI can be "fixed" (fully developed at the start) right ? Or does it necessarily need to constantly learn ?
I always laugh at AI stuff like ChatGPT which connects to the net to search for knowledge. Or a robot with an internet connection. In my opinion, AI and robots should be self sufficient in this matter. And if it must learn, let it learn from the people it talks to, not from the internet.
It's theoretically possible that poker AI also could be internet connected and learn from there, like browsing this forum's content for example.
Ok maybe, but AI can be "fixed" (fully developed at the start) right ? Or does it necessarily need to constantly learn ?I always laugh at AI stuff like ChatGPT which connects to the net to search for knowledge. Or a robot with an internet connection. In my opinion, AI and robots should be self sufficient in this matter. And if it must learn, let it learn from the people it talk
AI in poker don't look for knowledge anywhere as you said. It's an algorithm that is able to find Nash Equilibrium. And nobody can beat Nash Equilibrium. That's it. Nothing else.
Unbeatable poker AI ? I wanna play it ! 
And question also - what would be the exact amount of hands a human would have to play against this AI in HU match to see who's better ? Is it possible to define this minimum amount precisely ?
Current gen AI requires just ONE showdown hand before it can tell you what socks you're wearing.
Gen 2.0 (due Q3 '25) will go further and tell you where your sister got married. Gen 3.0 (Q1 '26) will go back in time and marry your sister. Good news for your sister, bad news for her fiancé, Brian.
Any poker AI is beatable - it's just the matter of the match format (total hands quantity):
If it's:
1 hand - I can win
5 hands - I can win
10 hands - I can win
50 hands - I can win
200 hands - I can win
So let's say that you claim that the minimum number of hands would be 1500 hands. I can then win in a 1499 hands match.
But it's impossible to set this minimum number precisely. There's no number that the term "long run" starts from.
And question also - what would be the exact amount of hands a human would have to play against this AI in HU match to see who's better ? Is it possible to define this minimum amount precisely ?
Imagine flipping a coin over an dover. After 100 flips you record that it lands heads 60% of the time, and tails 40% of the time. Is the coin fair and you're just running bad, or is it a loaded coin?
Well you can never know for sure, but you can get increasingly confident as you flip it more times.
In statistics you would state this as a range of probabilities with some confidence level, e.g. "we are 95% sure that heads is a between 55%-65% favorite".
You can apply that same process to poker. For example, you can measure your win rate against the AI and your standard deviation in bb/100, then calculate the probability that you have the edge using a tool like
Unbeatable poker AI ? I wanna play it !
Any poker AI is beatable
I know you're trolling, but you can't beat a GTO bot. Go practice agaist the GTO Wizard trainer. Hell, the drills in the study plan are free,
. You can even see exactly what the solver is doing, but you can't beat it. The best you can do is luckbox against it for a while, but you'll never have the edge.
1) I know you're trolling, but you can't beat a GTO bot.
2) Go practice agaist the GTO Wizard trainer.
3) The best you can do is luckbox against it for a while, but you'll never have the edge.
1. I'm not trolling, I just love to play against AI. Any poker bot is beatable. Like I say, it's a matter of a match format or tournament format. Simulating long run isn't the only solution. After all, long run is good to decide who wins more money, and you're not playing against a bot for real money. Besides, what's the point of measuring long run results if the bot doesn't get tired at all ? It's not fair.
2. I'm using Windows XP (I'm not trolling), so my old laptop won't handle this stuff you linked. But still, thanks. Maybe soon I'll be able to use it.
3. What about MTT against GTO bots ? I'm sure 1000 players or less is beatable.
1. This is silly and misses the point. You can't outplay a GTO strategy. I encourage you to study any modern poker theory book, video, or course before claiming otherwise in a poker theory forum.
2. These tools are websites. You can access it anything that has an internet browser. You can even use them on a cell phone.
3. GTO in MTTs, even with large fields, has
Ok, but long run (as a judgement) can be played only once per person. For example, I play 10 000 hands against a GTO bot and I lose. I won't play again, because this would be too much time consuming and the chance is to small. Nothing extraordinary. But what you're trying to tell me is that if 3 000 000 players worldwide play the same challenge (10K hands against GTO bot) they will all lose. All 3 million players losing ? Something is not right here.
We're talking about bot being "beatable" or "unbeatable" as an opponent. It may be a matter of terminology also. Let's say a very good player plays 250 hands high stakes HU challenge against Phil Ivey. And Ivey loses. So he can say then: "I beat Phil Ivey in a heads up match. Phil Ivey is beatable." Simple as that.
IDEA: Let's combine a GTO bot with a talking AI system and let it speak before the match. Let it say how many hands does it want to play against you to see who's better. I wonder what it will say then. If it's so smart, it should know that number before the match.
ITryDeuces, you're confusing luck with edge.
Having an edge doesn’t guarantee you’ll win every 10000‑hand match. Grasping this concept of long-term EV is crucial for any poker player.
So what are the numbers? Let’s look:
- Pluribus (a GTO bot) crushed 15 top pros in 6‑max over 10,000 hands at roughly 4.8 bb/100.
- LLinusLLove lost to Pluribus HU at about 0.5 bb/100. [2]
- A sample strong mid‑stakes cash game regs lost about 3 bb/100 according to my GTO Wizard reports below:
Now plug those win rates into a variance calculator and you get something like this for a 10k hand match:
For example, even at –5 bb/100 there’s still about a 30.9% chance that the human beats the bot.
If you want to learn more about edge and variance calculations, you can read my
about it here.
Bet I could hack a bot...
PASSWORD: bot
USER ADMIN: bot
Guys I'm in !!!
*Increases edge 25%
Ok, but long run (as a judgement) can be played only once per person. For example, I play 10 000 hands against a GTO bot and I lose. I won't play again, because this would be too much time consuming and the chance is to small. Nothing extraordinary. But what you're trying to tell me is that if 3 000 000 players worldwide play the same challenge (10K hands against GTO bot) they
Do you think you would beat a bot at paper-rock-cissor game?
