Note: I am asking this because I need to know whether to include your answers in my upcoming poker book The Theory of Poker Applied To No Limit.

When human chess, backgammon, and other games hustlers play somewhat weak opponents they sometimes play a theoretically incorrect move because it actually increases their chances. Chess computers don't do that and it wouldn't help them if they did since they always win anyway. But what if they were spotting a bishop or something like that?

More precisely put:

Given an opponent whose ranking is such that he would be about even money against a 2800 computer if spotted a piece. (What would that be? 1900 or so?),would I be right in assuming that a clever 2700 human, knowing he is playing that weaker opponent, would find slightly incorrect moves that would probably make him the favorite?

A 1600 human with enough time (i.e., not something super fast), and taking it seriously, will beat conventionally calibrated Stockfish every time, spotted a piece in the opening position. Full-strength Stockfish is a lot stronger than 2800 (in the sense of a lot stronger than a 2800 human, since computer ratings can't be directly compared). Although I can't see that difference mattering much in this particular case.

When the human is strong enough to understand the need to patiently liquidate to a decisive endgame, to carry out such a plan, and to not blunder while a piece down (and blundering is easier to avoid than with equal firepower), they are a comfortable favorite. My experience -- I have done this a few times out of curiosity and have seen others try -- is that an engine doesn't pose remotely enough problems.

I would imagine the 2700 human needs a qualitatively better opponent to dispatch them as consistently than would be needed against the engine.

Your wording was a bit hard to follow but I think that you are agreeing with me that a grandmaster that would be an underdog to the super computer would, in spite of his lesser strength, have a better chance against a weak opponent, when starting off down a bishop, than that same computer would against that same weak opponent. Am I right?

Correct.

I dno, Hikaru was getting smashed by Komodo (full strength) when spotted 2 pawns I believe. Not the same as a piece but I wonder if they ran that match too.

I agree that a 2800 GM should have a better chance vs. a weak opponent spotting a piece than a 2800 computer. But I'd imagine both are still heavy favorites vs. a 1600 player.

The difference between two pawns and a minor piece in this context is absolutely enormous, especially if one of the pawns isn't an f-pawn. I was actually considering whether to add that in the original post.

Some years back, after sailing through a piece up a couple times, I tried two pawns, underestimating how tough it'd be -- and it's basically a weird gambit that the comp inevitably recoups with flawless tactical play unless you're super sharp. Down a minor piece, the comp really doesn't generate much pressure.

Hmmm, well maybe not a dog against the machine then. But I'm still not convinced - I think the 1600 player would not have great results.

The 2700 human will play with both tactics and strategy in mind. The computer will play strategically, using tactics only when defending against them would give the computer a strategic advantage. That is the main difference.

I think it is a mischaracterization to say that when the 2700 human plays a move with tactics in mind, but which is evaluated by the computer to be a few centipawns worse than the "optimal" move, that the human made an intentionally slightly incorrect move. Because the human would not only make that move against the 1600 player, but also against a fellow 2700 player.

Another option would be a move where the 2700 human sees a refutation, but hopes that the 1600 player doesn't see it. In that case I don't think the "slightly" part of the statement is accurate anymore.

I'm a ~1600 player and you should bet on me against Stockfish in above scenario.

If you put effort into it (as if it were a serious OTB game), I think you'd do a lot better than you imagine.

I might give this a shot soon and report back. I've finished an OTB tournament today and according to the official rating calculator had a FIDE Elo performance of 1639 (which should bring my actual FIDE rating up to 1583), so I'm right in the range which you were saying should beat Stockfish 100% of the time with piece odds.

Stock fish is waaaaaaaaaaay above 2800, though. I thought Rei is saying a 2800 computer. So you gotta find whatever computer plays 2800 chess.

No, he's talking about "conventionally calibrated Stockfish":

Either way, if Rei's hypothesis is true, does it make a difference? I'm already highly skeptical I could beat Magnus Carlsen with piece odds. I don't think I'd have better chances against, say, Sam Shankland. But if he's correct that piece odds are good enough for a guy like me to beat a - for all intents and purposes - invincible chess-entity, then I don't think it matters much. Might as well play Alpha Zero powered by a Google cluster then.

Again, I don't think he's right, and we'll see in due time (on my computer, not some TCEC machine of course, but that should be good enough to whoop my 1600 ass).

I think it's not close that the human would have the edge at winning a piece down against a comparable opponent, assuming the computer is straight "out of the box". The ability to not just avoid moves but to steer the game in a particular direction and pawn structure as to make trades less likely requires more strategic decisions at the start than straight calculation.

So you think my chances against Carlsen (or Shankland for that matter) would be worse than my chances against Stockfish?

The superior player (down a piece) would play in a way to complicate the position in order to maximize the chances that the weaker player will make a critical mistake.

In it's most basic form, this would mean refraining from exchanging pieces in order to keep as many pieces on the board for as long as possible.

Sometimes that would mean playing theoretically inferior moves as early as move 1.

For example, if the 2700 player started the game without his queenside bishop, I think it's likely that player would avoid playing a standard 1. e4 or 1. d4 as their opening move. Those moves would lead to an open game in which pieces could be exchanged relatively early (to the benefit of the inferior player). Instead, the 2700 player would likely play something like the theoretically inferior 1. g3 and 2. Bg2, and not advance any pawns or pieces beyond the third rank in the opening.

I think this chess strategy of complicating the position on the board is somewhat analogous to an inferior NLHE player playing a high risk (heavy all-in) strategy against a superior player.

Here is a game. This is blitz (I don't have time for a longer game atm) but SF is still pretty strong. Missing queen's knight for black.

Shortly before the resign kicks in, Black is making nonsense moves because it evaluates itself down a ton I suppose.

Nothing special from me; haven't played chess in a while. I just played patiently and it acquiesced.

"Win 100%" is hyperbole. Obviously you can't predict people's temperaments and how they'd approach the situation strategically. But without contempt, the comp doesn't put up a lot of resistance.

[FEN "r1bqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1"]

1. e4 e6 2. d4 d5 3. exd5 exd5 4. Nf3 c6 5. Bd3 Bd6 6. O-O Ne7 7. Bg5 Qc7
8. h3 O-O 9. c3 f6 10. Be3 Be6 11. Re1 Qd7 12. Nbd2 Bf5 13. Bxf5 Nxf5 14.
Nf1 a6 15. Qc2 g6 16. Re2 Rfe8 17. Rae1 Re6 18. Bc1 Rae8 19. Qd1 Rxe2 20.
Rxe2 Rxe2 21. Qxe2 Qe7 22. Qxe7 Bxe7 23. Ne3 Ng7 24. g4 Kf7 25. Ng2 g5 26.
Ne3 h5 27. Nf5 hxg4 28. hxg4 Nxf5 29. gxf5 Ke8 30. Kg2 Kd7 31. Kh3 b5 32.
b4 Bd6 33. Kg4 Bc7 34. Ng1 Ke8 35. f4 gxf4 36. Bxf4 Bd8 37. Kh5 Kf7 38. Kh6
Be7 39. Nf3 Bd8 40. Nh4 a5 41. a3 Bb6 42. Nf3 Bd8 43. Nh2 Bb6 44. Ng4 Bd8
45. Kh7 Bb6 46. Nh6+ Ke8 47. Kg6 Kf8 48. Bd6+ Ke8 49. Kxf6 Kd7 50. Be7 Bc7
51. Ng8 c5 52. Bxc5 axb4 53. axb4 Bg3 {Black resigns} *

In the specific instance of starting the game a piece up, yes. However, I think there's a very small window where it actually makes a difference - for example, below 1500 I'm taking the stronger player missing a piece whether computer or human. Over 2000, I'm taking the weaker player with the extra piece whether or not the opponent is computer or human.

Somewhere in the middle of that is a small rating group that the human player is more likely to outplay the opponent than the computer player.

Damn, I'm in the "****** gets outplayed by human but can beat machine" window.

So you are saying what I expected and I can now include chess in a point I make in the book. But you confused yourself in your last sentence as the analogy would be the superior player playing "smallball".

Yeah, that's probably what he was trying to say. However, I would generalize this even further:

The superior player in NLHE will play exploitable, knowing the likelihood his inferior opponent playing exploitative is low enough to make playing exploitable profitable. Smallball is certainly one way of playing exploitable to extract maximum value from inferior players, but not the only one.

The analogy was a bit weird but correct. In both cases one of the players has a disadvantage (either being down a piece or being inferior at poker) and compensates for it by playing a high variance style. The reason for the confusion is that in chess we are looking from the point of view of the superior player and in poker from that of the inferior player.

GM Larry Kaufman on playing Lc0 with knight odds:

I think the analogy of exploitive play is the best.

Comparing the superior chess player and the superior poker player (when each are playing a significantly inferior opponent)

Playing the theoretically best computer chess move is like playing a GTO no-limit hold 'em strategy.

Whereas, playing a less than optimal chess move to complicate a chess position is like playing an exploitive no-limit hold 'em strategy.


The superior player could use either strategy and have the advantage. But, the superior chess player will win more games with the less optimal/complicating strategy and the superior poker player will win more money with the exploitive strategy.

The original question is very weird and I am not sure if I understood it correctly, but if a strong player is giving piece odds to a weak player in chess, he doesn't neccessarily have to play for tricks. The reason is that material advantage in chess usually doesn't lead to an immediate victory. You can play a lot of moves being a piece down, especially if the extra piece of your opponent is completely out of play. The best example are rook odds games by the old masters from 200 years ago. The amateur with the extra rook would get mated before he was able to make a single move with his extra rook.

Computers play without any strategy whatsoever. They just play from move to move and they always play the move with the highest evaluation. The evaluation is based on static criteria, so they are basically just counting advantages and discount disadvantages. A knight is +3.25 pawn units, a half-open-file is +0.3 pawn units, the pair of bishops is +0.5 pawn units, a backward pawn is -0.3 pawn units and so on.

Computers don't bluff either. Yes, there is bluffing in chess, especially on the highest level because super grandmasters have to memorize a couple of thousand positions before each game. You never know where you opponent is going, if you remember everything in that line and what dangerous novelty your opponent has in store for you.