Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AI Beats Four Top Poker Players

bbc.co.uk

Post reply on HN
Print view View original

191–200 of 234 posts

Re: AI Beats Four Top Poker Players

#191

Post by mormegil »

andrewprock wrote: post #134
mormegil wrote: post #109

Earlier quoted context omitted.

That depends on what you mean by "correct". Sure, you could theoretically find Nash equilibrium of poker and by playing the equilibrium strategy, you can ensure you won't lose. But that does not mean this is the best strategy to use at a given table against the given opponents, who (being imperfect humans) almost certainly do not play the equilibrium strategy themselves. And, by playing a proper nonequilibrium strate…

The usual way these games are solved is to create an "abstract" game which is tractable, find the Nash equilibrium, and map state in the real game back to the "abstract" game. In the limit, the solutions for a well designed "abstract" game will converge to that of the real game.

You are explaining how current algorithms try to find the (approximate) Nash equilibrium (and those algorithms are far from perfect; as noted in the recent DeepStack paper, current abstraction-based programs are beatable by over 3000 mbb/g, which is four times as large as simply folding each game). But my point is that even the (exact) equilibrium strategy would not necessarily be the best strategy against given non-equilibrium-playing players.

Re: AI Beats Four Top Poker Players

#192

Post by adamlett »

wslh wrote: post #189

Could someone explain how Poker is more challenging that games like Go? Please eli5.

I'm not qualified to give a comprehensive answer, but I would have to imagine that it partly has to do with the fact that poker is a game of partial information. In Go or Chess the current state of the game is known. In poker, you don't know what cards your opponent hold, so you have to assign probabilities based on previous actions, and simulate many different scenarios.

Another part has to do with the fact that both you and your opponent have surprisingly many legal "moves" at each turn. because not only must you decide to fold, call or bet, but if you bet, you also have to decide how much.

Re: AI Beats Four Top Poker Players

#193

Post by StringEpsilon »

brilliantcode wrote: post #5

it won't be long before we hearing more headlines like: "AI beats top 10 hedge fund managers" to "AI run hedge fund blows up due to black swan event" regardless it's an incredible feat. It really casts questions into what our edge as humans are which is slowly disappearing and we didn't even need to put a brain in a jar and hook it up to a computer....it's deep learning reinforced algorithms that is appearing to outl…

I mean, by definition Black Swans can't be anticipated. So, no matter if it's an AI or human, black swans will throw you over.

But, an artificial intelligence probably has different black swans than humans, as their perception is inherently different. A tweet from Trump might not surprise us (anymore) but to an AI player, it might not be evident that a small amount of text from one Human can cause an uproar on the market. And it might not even have access to the relevant data (Twitter) at all.

Re: AI Beats Four Top Poker Players

#194

Post by BigJono »

_dps wrote: post #161
T-hawk wrote: post #154

Earlier quoted context omitted.

A 100% random RPS AI doesn't have a poor 1-on-1 win rate against any other RPS AI, but it absolutely can have a poor rate of winning tournaments , if "poor" is defined broadly enough. For a tournament that pays cash to the top 10%, most human players would consider anything in the bottom 90% to be poor, which would include a 50% win rate from a random AI. This happens because some entrants aren't 100% random, and the…

To add something, and to clarify (though I don't think you've made any mistake here, but many people misunderstand this subject): The 33%-each Nash Equilibrium is the mixed strategy Nash Equilibrium of the micro game (i.e. a single round of RPS, averaged over all possible randomizations). This is in no way the Nash strategy of the tournament game, which is "win the tournament given a pool of unknown participants and…

> You have to add additional assumptions (e.g. that everyone else is going to play the uniform random strategy) in order for uniform random to be the Nash strategy for the tournament game.

Isn't that literally what a Nash Equilibrium is though? It's my understanding that if there is players playing exploitably in the game then it cannot (by definition) be a Nash Equilibrium, so the Nash strategy may no longer be the optimal or maximally exploitative one.

Re: AI Beats Four Top Poker Players

#195

Post by yAnonymous »

leeleelee wrote: post #74
andrewprock wrote: post #56

Earlier quoted context omitted.

I'm surprised that Andrew Ng made this claim. The strategy that was built for Libratus' predecessor did not do sophisticated modeling of the opponents, or use new algorithmic principles. Poker is solved using a very large game tree, just as with the other games. The structure of the tree is modified to support the notion of hidden state, but beyond that it is essentially the same as the other games. The structure for…

This was my thought too. Poker is quite "solvable" meaning, whenever you're confronted with a decisions -- there is always a "correct" answer which does not have to depend on the other players' behavior or style. And you can find that answer by simulation, or game trees, and other methods. It's also important to keep in mind that the best AI can still lose, and the worst AI can still win (and everything in between).…

>there is always a "correct" answer

That's wrong. Even when you're holding a good hand, your opponent could hold a better one and reading them is a key element of poker. The opponent's hand is an important variable to decide whether you hold the winning hand or not.

If you look at the experiment in detail, you'll find that it was set up in the AI's favor.

>When a hand was all-in before the river no more cards were dealt and each player received his equity in chips.

While all that is less important when you can avoid all-in situations, the main statement -that the other player's behavior is irrelevant- is still wrong.

Re: AI Beats Four Top Poker Players

#196

Post by BigJono »

deepnotderp wrote: post #156
brilliantcode wrote: post #136

Earlier quoted context omitted.

well to be honest I never bothered googling that exact term but I was more interested in the details around his deployment of smaller and successful neural networks as part of a portfolio.

Basically, certain neural networks perform better at predicting different factors at different market conditions. You can string them together to get a decent trader, enough to make a profit, but not enough to justify hogging your GPUs.

As in the profit generated isn't enough to sustain the electricity cost of running the GPUs? Are you able to scale up with more initial investment? How much did you invest and what kind of return on investment were you getting?

Re: AI Beats Four Top Poker Players

#197

Post by kkoomi »

09bjb wrote: post #77
RyanRies wrote: post #30

Earlier quoted context omitted.

Yeah... I guess I was just expecting the AI to be more self-learning and self-tuning...

It built most of its decision model by playing itself over "trillions of hands", according to Prof. Sandholm. You could almost think of the nightly reviews as a similar process to a human sleeping: we benefit from memory consolidation, pruning, time to reflect, etc. I don't see giving the computer 12 hours a night to do something close to reflection and self-analysis as "cheating," or even very different from what it…

The four humans were even "colluding" too.

Re: AI Beats Four Top Poker Players

#198

Post by rograndom »

mod wrote: post #182
marklgr wrote: post #179

Earlier quoted context omitted.

Back when I played poker (up to ~2010), Negreanu was mostly a celebrity, as well as a pretty good blog writer. He was no match for the best pro players, but perhaps he has dramatically improved. Ivey was pretty good at managing his image, too.

Say that all you want, those guys play(ed? I'm out of the loop) in the big game, and...whoever you're thinking of, basically--didn't.

The "Big Game" which was run by Doyle Brunson, Chip Reese, etc had nothing to do with HUNL. It's mostly run hoping that some inexperienced player would drop in. It also almost always ran as a rotation of games to both bring in players who though they had an edge in "their" game and then exploit them in the other games.

The Big Game and online HUNL are both technically "poker" but they are truly completely different games.

Re: AI Beats Four Top Poker Players

#199

Post by yazr »

yAnonymous wrote: post #195
leeleelee wrote: post #74

Earlier quoted context omitted.

This was my thought too. Poker is quite "solvable" meaning, whenever you're confronted with a decisions -- there is always a "correct" answer which does not have to depend on the other players' behavior or style. And you can find that answer by simulation, or game trees, and other methods. It's also important to keep in mind that the best AI can still lose, and the worst AI can still win (and everything in between).…

>there is always a "correct" answer That's wrong. Even when you're holding a good hand, your opponent could hold a better one and reading them is a key element of poker. The opponent's hand is an important variable to decide whether you hold the winning hand or not. If you look at the experiment in detail, you'll find that it was set up in the AI's favor. >When a hand was all-in before the river no more cards were de…

>>If you look at the experiment in detail, you'll find that it was set up in the AI's favor.

Could you elaborate on this ?

Re: AI Beats Four Top Poker Players

#200

Post by deepnotderp »

BigJono wrote: post #196
deepnotderp wrote: post #156

Earlier quoted context omitted.

Basically, certain neural networks perform better at predicting different factors at different market conditions. You can string them together to get a decent trader, enough to make a profit, but not enough to justify hogging your GPUs.

As in the profit generated isn't enough to sustain the electricity cost of running the GPUs? Are you able to scale up with more initial investment? How much did you invest and what kind of return on investment were you getting?

Not enough to justify using up the compute capabilities of my GPUs since I need them for research.

I averaged a return of 7-13% a year. With less than 1000 in capital, it wasn't very profitable.

I don't see why it shouldn't scale.

Post reply on HN