Hacker News Reader

Comment by blueside | Hacker News Reader

I disagree and it's not ignorance on my part. Due to the nature of the poker and heavily weighted component of luck, a terrible AI can theoretically win.

I am sure the AI they built is a mighty and spectacular achievement, but poker is about the only game I know where the worst player can easily beat the best player. Prove me wrong and I'll happily accept your insult of "ignorance"

edit I should have been more clear, as I am extremely impressed with the AI's results of hands over time. I am referring to the context of a standard tournament where the loser is eliminated after losing their chips.

View on HN

On a hand by hand basis, sure. In freezeout tournaments, yes. In heads-up over 120k hands? No. This was not a lucky win. And I say this speaking as a one-time full time player in a number of formats. Variance does not explain this loss.

Consider a game played in a series of rounds in which N% of rounds are decided by luck (random player wins) and (100-N)% are decided by skill (better player wins). As the number of rounds approaches infinity, the win percentage of the better player will approach (100-N/2)%

That is, if the game is 100% luck, the better player will have an expected win percentage of (100-100/2) == 50%

If the game is 0% luck, the better player will have an expected win percentage of (100-0/2) == 100%

If the game is 95% luck, the better player will have an expected win percentage of (100-95/2) == 52.5%

Conclusion: If there is any non-zero amount of skill in the game, the better player will win in the long run.

"the worst" player might win a few hands, or even beat "the best" player due to luck.

They will not be able to consistently beat a better player, especially not professional players.

That's what this AI is doing. It didn't beat them once or twice. It beat them consistently over the course of 120,000 hands in a 20 day event.

Is it theoretically possible that "the worst" player could do that based entirely on luck? Sure, probably in the same realm as monkeys, typewriters, and Shakespeare.

In these contests they typically play each hand twice with the sides reversed to remove the luck factor.

So the deck for hand 1000 might be the same as hand 113853 but they swap who is the button.

> " I am referring to the context of a standard tournament where the loser is eliminated after losing their chips."

Why would you refer to that? That's not at all what happened here.

In the game that was played (long-term HUNL cash) the worst player cannot beat the best player at all. Not easily, not at all.

Not to mention they didn't run the cards out.

You're commenting without reading the article, seemingly.

Yes it's possible, but it's also possible that a terrible chess AI could get lucky and win by the opponent getting a heart attack in the middle of a game.

Reply on news.ycombinator.com