20

I can choose to hire a titled player to play with me; at the same time, playing with the computer is usually free and more flexible.

As far as improving one's chess skill is concerned, is there more value in playing with a titled player than playing with a chess bot with similar playing strength?

Zuriel
  • 9,063
  • 6
  • 40
  • 98
  • 8
    Usually playing against a chess engine is not recommended. You should play with the real people in your own level and try to find your mistakes. – Minot Dec 12 '21 at 06:38
  • 23
    If you're willing to hire a titled player, hire an actual coach instead. – Comic Sans Seraphim Dec 12 '21 at 11:15
  • 2
    At first I read this as "tilted" player, and was very confused – justhalf Dec 13 '21 at 05:46
  • Note that some engines can give a (rather meagre) feedback on the nature of your errors (or even warn about hanging pieces), but unless those are quite obvious tactical patzers, a trainer is still better (and be it in the form of a chess book). – Hauke Reddmann Dec 13 '21 at 08:32
  • 1
    I think this question is very opinion based. – anion Dec 13 '21 at 10:06
  • A human player can tell you why one move is better than the other. Engines (like the one on lichess.org) can just say that a given move is better than others but you need to do the appropriate thinking on your own. – Thorbjørn Ravn Andersen Dec 13 '21 at 23:24

3 Answers3

23

Yes, as long as you go over the game with them afterwards.

A titled player can explain why another move or idea is better, and not just point out what you did wrong.

The engine can't explain in human terms the reason for its evaluation. This is true especially for non-tactical lines.  

Federico Poloni
  • 515
  • 5
  • 15
Herb
  • 4,197
  • 1
  • 20
  • 42
  • "The engine can't explain in human terms the reason for it's evaluation" -> decodechess.com enters the chat – nathanchere Dec 13 '21 at 13:39
  • The difference being, decodechess doesn't answer the questions you might want to ask about the position. It's better at it than an engine, probably, in most circumstances, but nowhere near as good as a human. – Arlen Dec 13 '21 at 23:15
12

The common advice that you shouldn't play against a computer has a few facets.

It is true that chess AIs play differently from humans, not because they cannot play like humans but because most of their designers are concerned with winning rather than losing like humans. So if you play against an AI, you would usually see a different distribution of responses to your play than against humans. In particular, the AI is unlikely to make the same variety of mistakes that humans make, so you lose out a bit in learning how to punish such mistakes. Also, the AI is unlikely to take safe solid winning routes, since if they can see an ad hoc 10-move-deep tactic that gets a better result they will take it over a slow-and-steady endgame-win, so you would also lose out in such understanding.

That said, it is a general truth that one of the fastest ways to learn to play a strategy game is to keep playing against a player that is slightly better than yourself (this has even been mentioned before on Chess SE). If you do precisely this, it is possible to learn quite fast whether or not your training opponent is human. After all, even the Lichess AI is designed to make some amount of mistakes, so if you are able to punish these then you would also be able to punish human mistakes.

The biggest problem with playing against an engine is that you must be proactive in identifying and learning from your mistakes, playing out alternative lines against the computer to figure out where you went wrong and what you could have done instead. If you cannot understand why a move is a mistake because the tactic involved seems too deep, it may have been an issue in the past if you do not have a hired trainer, but not in modern times since you can ask on Chess SE...

Ian Bush
  • 4,387
  • 1
  • 23
  • 31
user21820
  • 2,809
  • 1
  • 12
  • 30
  • Chess as played by recreational players is a very different game from chess as played by experts. One will learn best by playing the strongest possible opponent who is still playing the same game, but most computer programs aren't equipped to play the same game as recreational-level human players. – supercat Dec 13 '21 at 23:20
  • @supercat: I don't understand your comment. Chess AIs are playing the same game, whether or not their play feels different from humans. – user21820 Dec 13 '21 at 23:27
  • Imagine a chess variant where every turn one would announce a move, roll a die, and if the die came up six one would have to pick a different move. Success in such a variant wouldn't be affected as much by one's ability to formulate coherent plans as by one's ability to exploit situations where an opponent played a sub-optimal move. In chess as played between recreational players, someone who is able to exploit an opponent's mistakes will often win against an opponent who cannot, even if the former player makes more mistakes. An ability to formulate plans twelve moves deep would be... – supercat Dec 14 '21 at 00:05
  • ...a huge asset in expert-level chess, but would be irrelevant if one would be unlikely to get anywhere near the completion of the plan before an opponent makes a mistake, and the reward for exploiting that mistake would exceed the reward for successfully completing one's original plan. – supercat Dec 14 '21 at 00:06
  • @supercat: I understand what you are saying, but how is that relevant? The higher the chess AI level, the less frequently it (usually deliberately) makes a mistake, and also the lower the centipawn loss in each such mistake. As I said in my post, you need to set the chess AI level a bit higher than your own, so if you become good at exploiting its mistakes rather than formulating long-term plans, then you would have to increase its level to the point where those mistakes become irrelevant to your training. – user21820 Dec 14 '21 at 14:22
  • If a pair of experts and a pair of novices were to arrive at a particular board position, it's entirely possible that the best move for an expert to play might be a very bad move for a novice to play (e.g. because it drops a queen without any compensation in the next 11 moves, but allows a mate in 12). If expert-level chess, novice-level chess, and recreational-level chess were all "the same game", a good move in one should be a good move in all. A person with a given level of understanding would benefit most from practicing against an opponent whose moves would generally appear to be... – supercat Dec 14 '21 at 16:57
  • @supercat: What you just said is already covered by the last sentence of my first substantive paragraph. – user21820 Dec 14 '21 at 17:04
  • ... the best possible, given that level of understanding. A beginner whose opponent initiates a queen sacs that feed into 12-deep mating combination is unlikely to learn any positive lessons from the experience, but might instead perceive that queen sacrifices are likely to be magical miraculous plays because all of the queen sacs by the opponent seem that way, even though queen sacrifices played by novices are almost always losing plays even when they happen to be sound. – supercat Dec 14 '21 at 17:05
  • @supercat: As for queen-sacrifices, I want to remark that I do not consider a win-in-12 move to be bad if made by an opponent of a beginner. The very existence of such a winning path implies that the beginner's position in the first place is dangerous. And no, it cannot give the impression that queen-sacrifices are miraculous because if you play enough times against the chess AI the vast majority of AI wins will not be via 12-move deep tactics! – user21820 Dec 14 '21 at 17:08
  • I guess my intended point was that if one were to rate players separately on breadth of knowledge and cleanliness of play, an ideal practice opponent for most players should generally combine a breadth of knowledge which is roughly comparable to their own, with either extremely clean play. A sloppy player will gain more benefit from a practice opponent that consistently punishes sloppiness than one who would often let it slide, even if the former opponent would, at least initially, beat the player 95% of the time. The only time sloppiness would be a virtue in a practice opponent would be... – supercat Dec 14 '21 at 20:35
  • @supercat: I don't know why your last comment trails off. I'll just say that there are plenty of details that do not make it into a tiny SE answer-box. If you wish, we can continue this discussion in chat. – user21820 Dec 14 '21 at 21:26
4

Several factors are at play here:

  1. Humans tend to be more interesting to play against. Therefore, you will play more and learn faster! If you regularly play a bot, you will typically always win or lose at a given difficulty level. The best chess players in the world cannot beat a well-optimized AI, even if it's powered on your phone. This tends to be... boring for most players. You might be different, but probably not.

  2. The mistakes an AI makes are not like human mistakes. AI mistakes look nothing like a human mistake, since we start by looking at moves we think are good, a bot has no such compulsion. It might, for an (absurd) example, check from left to the right for moves, and always make mistakes on the left side of the board because we have to tell it to make mistakes on purpose. On the other hand, human mistakes are stylistically consistent compared to AI ones - we overvalue kingside attacks, or misunderstand the value of connected passed pawns. The "mistakes" made by an AI can be manufactured by several methods:

    • The bot might generally play "badly" by truncating the search tree after a certain time period. The AI makes whatever move it thinks is good as soon as a timer occurs.
    • By deliberately selecting moves that are bad n% of the time. The bot looks through the list of moves and looks for ones that are within a tolerance of "best". Perhaps +/- 0.5 points (half a pawn), and selecting one of those. These look... strange and not appropriate for the position.
    • Major point here: There is one bot, Maia, which is specifically designed to make human-like errors. It was trained, basically, on tactical errors made by real players at a particular Elo rating. From my experience and the evidence attached, it does seem more human-like, and this has given me a better appreciation of what mistakes seem "normal". However, it struggles very much in drawn positions (see point 3).

  3. One can often beat bots like by playing slowly, reaching a closed position and then just waiting for it to do something foolish. Doing this, you will not learn to make a good plan, "force mistakes", or drag yourself through a complex position.

  4. As already mentioned, human players can say why you lost. An AI simply stomps you into the ground and tells you "Don't be so bad at chess" while pointing to specific moves. A human can watch 5 games and say: "You are not attacking the true base of this pawn chain! Reread your Nimzowitsch!"

Glorfindel
  • 24,825
  • 6
  • 68
  • 114
RegressForward
  • 141
  • 3