So when I peruse the many strategy books I have, I have found many positions where there is a clear best plan, that seems clearly much better than all other plans. The variations given by the experts, and the game continuations seem to confirm that the stated best plan is truly the best. But now with 4000+ rated engines, the best plan that, say Stockfish, finds is at odds with those experts' opinions with some of these positions. Before simply concluding that these engines, despite their ratings, simply don't understand these positions, I looked carefully at their analysis, and indeed there are moves by the opponent, not necessarily tactical and some pretty deep down the tree, that the experts did not mention, and the best lines of play found by the engines seem to suggest that the "best plan" does not turn out to be as good as the experts say after the apparent best lines of play found by the engines. But without calculating any variations, the quoted best plan does seem to make the most sense, meeting the strategic needs of the position. But the engines analysis change that outlook. So what does this mean for the meaningfulness of these quoted best plans?

1 Answer 1


This is a tricky thing to answer. What is boils down to is the following:

On a game theoretical level there is no such thing as a plan. There are no such concepts like "strategical advantages", doubled pawns being bad, the bishop pair being good and so forth. There is just calculation. A position is a win, a draw or a loss based on the concrete variations that happen with best play.

However, for a human that is not very useful. No human, in fact not even any engine can get anywhere close to solving chess and/or calculating trillions of variations. This is where especially traditional engine development and human play diverge:

For humans it turns out that beside developing an intuition for positions, some general guidelines can be helpful. The bishop pair being better than two knights may not be true in every position, however, it turns out to be true in many if not most positions. And so, due to not being able to calculate trillions of variations, for humans those general guidelines are a more practical approach that often works well. That being said, especially very strong players will also at times break those guidelines when they feel like in the position at hand the guideline does not work well.

On the other hand you have engines. While they can not calculate trillions of variations either, they can by now calculate at least millions per second and even many decades ago already calculated thousands of variations per second, more than any human could do. With this calculation power it can become feasible to attempt to go through many concrete variations. Engines will still need heuristics similar to those used by humans, however, with the much more powerful search, concrete calculation will always overturn the general guideline if it has to. With new neural network engines their mode of operation has changed a bit, however, they just the same will always have calculation overturn general guidelines if need be.

When looking at general plans in openings the same is necessarily the case: the human approach is to find something (a plan), that works most of the time and is practical to use. Meanwhile the engine will use brute force calculation to figure out that actually, if the opponent plays in just the right way, the plan will not work and thus a different sequence of moves is preferable. However, while this is correct from an objective point of view, it may not be practical for a human. So the good old simple plan in an opening might still be a humans best bet, even if the engine rightly points out that things are more complicated than that.

That being said, there are also some variations where entirely new plans have been found by engines. (to give one example, since engines became really strong some 2 decades ago, we see a lot more h pawn advances in top level chess)

So I suppose my bottom line is: For a human such plans are still useful in practical play, even though they will not always work.

  • @ koedem Let's say a middle game position is a tablebase draw except for one move that wins, but that move violates many strategic principles, while the move that every grandmaster says is clearly the best because it meets the strategic needs of the position and gives a strong advantage is actually a tablebase draw with very accurate play by the opponent (but not calculatable by any human). The human should scrap the tablebase winning move and still go with the recommended one?
    – prestokeys
    Commented Oct 31, 2023 at 21:30
  • 1
    It depends on the position of course, but yes I would say so. In some highly regarded books like the Seven Deadly Sins, final section of chapter 3 the author mentions the following: when analyzing with great players he'd often ask them whether some position was winning. They would reply "There are good winning chances" rather than an objective evaluation. In practical games, the concrete evaluation may matter less than how likely it is to win in practice against another human.
    – koedem
    Commented Nov 1, 2023 at 8:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.