The other day we reviewed Nate Silver’s new book which had a lot of poker content, and this reminded me of a post from 10 years ago on luck vs. skill in poker:
The thread of our recent discussion of quantifying luck vs. skill in sports turned to poker, motivating the present post.
1. Can good poker players really “read” my cards and figure out what’s in my hand? . . .
2. Some references on luck vs. skill in poker . . .
3. Rick Schoenberg’s definition . . .
Some interesting discussion there in comments too. Enjoy.
Reading the old post, your experience with the good poker player seeming to read my mind reminds me of my own test case. I was not a good poker player and my uncle was. He claimed he could tell what was in my hand by watching me as I picked up the dealt cards. Of course, I thought this was ridiculous.
So one evening we sat down at a table and he dealt me pairs of cards. Darned if he wasn’t right! Mostly he could tell pretty closely what I had been dealt. It was 60 years ago and I don’t remember the details any more but I still remember the feelings of disbelief and amazement.
One of the highlights of my (then) young life was a few years later when he and I were playing penny poker with friends and I bluffed him big three times in the evening. Somehow he must have taught me something!
“seeming to read my mind ” -> ” seeming to read your mind “
Poker players always complain about the brutal variance in poker, so you could be unbelievably good and still wind up losing money over the course of a year in a pro career.
I’ve often wondered if you could design a good “poker-like” card game with much lower variance. Then the best player over a relatively shorter period of time would tend to win, but hopefully the game would still test the same skills (playing near Nash, while knowing how to read your opponent psychologically and strategically and having a feel for when to switch to exploitation). Such a game would obviously be better for tournaments if you couldn’t just get lucky and win the whole thing.
I don’t know if such a thing is possible or not — maybe for some reason any interesting poker-type game would have to be high variance. Wonder if anyone has thought about this over the decades or tried to formalize it.
Another factor is that poker as a whole needs the variance to attract whales who will lose lots of money subsidizing everyone else, so you might not want to get rid of it.
Tic-Tac-Toe is all skill and low variance…. and nobody wants to play it. You see the problem.
If good players wanted lower variance, they’d play smaller stakes.
PS: I stand by all my comments at the linked thread.
Tic-Tac-Toe is complete information but more importantly, trivially solved.
People play deterministic, perfect information games like chess and enjoy them. But it’s widely agreed there’s something missing compared to stochastic, imperfect information games.
You could have a deterministic imperfect information game that doesn’t suck — people find Stratego interesting.
Poker has both randomness and imperfect information. There’s lots of e.g. European board games that also have this, but they are too complicated and you usually aren’t betting. Whereas poker is really clean and simple.
By “poker-like” I mean a game that you play with a 52-card deck, where the main goal is to have a winning hand between your hand and the community cards, and where the main decision is how much and when to bet. People really like that type of game, it seems to be something special.
But could you have one with much lower variance? It doesn’t seem impossible a priori given this whole range of other games which people also like.
it occurs to me that the “first best” impossible ideal poker game would be, if there are N players at a table, to create N copies of each player and have them independently and simultaneously all play every hand that is dealt.
but that’s more of an analytic philosophy thought experiment than a real game.
~30 years ago, the sociologist Eric Liefer wrote an amazing dissertation about skill in chess, titled Actors as Observers. (It’s the original inspiration for the concept of “robust action” which I believe has cachet in some political science circles).
One of the central arguments is that game theoretic decision trees are almost always useless for understanding skill expressed in the real world because skilled actors tend to select into deeply uncertain “game states”. This is why (to use a silly example) there are no professional tic-tac-toe leagues and why skilled chess players resign the instant a game’s outcome becomes calculable (e.g., mate in X-moves).
Leifer poses a cute thought experiment about aliens who know nothing about chess arriving on Earth and looking for the best players. An alien who thought chess was a competitive game about winning, would likely select a terribly ranked player while an alien who thought chess was a cooperative game about stalemating, would likely find a grandmaster. The explanation for this is that draw rates steadily increase as you advance in skill level. This holds both at any given time and across eras of chess: Fischer’s draw rate was 30% (2780 peak FIDE rating), Kasparov’s, 37% (2851 FIDE), and Carlsen’s is 43% (FIDE 2882) (Those numbers should be close but are not 100% vetted). The lowest draw rates are among the poorest players so, relying on the assumption that people tend to play similar-skill opponents, the highest win-rates will come from junior players on fluky win streaks.
All that is to say, Leifer argues that in many competitive circumstances, skill is best indicated not by winning but by the ability to sustain a competitive relationship. A checkmate tends to indicate a bad matchup of players. A stalemate tends to indicate a good matchup AND that the players had the skill required to sustain the game.
To finally connect to the blog post, I wonder if you could calculate a convincing index of poker skill just by averaging how many hands a player tends to play given the sizes of the pots and blinds (forgive any weird word usage, I am not a poker player).