Best human plan vs best computer plan

Question

Suppose that in a given strategic position, every grandmaster agrees that plan A is better than plan B, and for the first few minutes the strongest computer engines calculate that A is best and B is second best. But then at, say, depth 30 or so (at a depth beyond any human calculation), the engines find that plan B actually leads to better positions than plan A (even though there are no tactical refutations), with say, a +1.00 advantage over plan A. Would the grandmasters still insist that plan A is a better plan than B, or would the 4000 Elo rating of the chess engine make them rethink this? What would objectively be the best plan, A or B, if following the best line of play in plan B leads to a strong position dozens of moves down the road, while following the best line of plan in Plan A does not after dozens of moves (again, beyond what a human would be able to calculate)?

Answer 1

Any grandmaster worth their salt would then assume B is in fact better, and if they wanted to know why they could analyze it out with the engine. Being objective is an important quality for any strong player, and I'd find it hard to imagine a grandmaster would continue to truly believe their plan is objectively best, if an engine disagrees. The grandmaster might continue to believe that their plan would offer the best practical chances, or that it suits their own style best, but these are more subjective claims that they would recognize as such.

If the engine at depth 30 suggests plan B, then it's likely that it's better than plan A. Although note that at even higher depths, engines can still change their minds, so it's not always certain what plan it'll settle on way down the road. And then there's the issue that we don't know how close engines are to "objectively perfect" play. But still, current Stockfish at depth 30 is pretty damn good (like really good). Any opinion of a GM that agrees/disagrees with the engine's evaluation usually has no weight on what the objective evaluation is. Again, GMs can be good at deciding which plans give the best practical chances, but they don't hold a candle to engines in figuring out which moves are objectively best.

Answer 2

What GMs do - and I imagine most players - is force the computer to look at plan A and see what the result is.

For example, take this position with Black to move:

[FEN "rnbq1rk1/pp1nbppp/4p3/2ppP3/2PP3P/P2B1N2/1P1N1PP1/R1BQK2R w KQ h3 0 9"]

White obviously threatens Bxh7+ with the Greek Gift sacrifice. Black has two ways to stop it, ...g6 and ...h6. Which plan do you choose? Stockfish at d=23 says ...h6 is +0.8, while ...g6 is +0.9.

If a GM wants to get more sense of this position, then what he'd do next is input ...h6 and ...g6 and let Stockfish compute White's best move. This works because by inputting the move, Stockfish will focus on that move and that move only. If you do this, then Stockfish at d=24 says 1...g6 2. O-O or Nb3 is +1.0, while 1...h6 2. Bb1 is +0.7. Clearly Stockfish is becoming more convinced that 1...h6 is the best move. Is it right though? Well you can play through Stockfish's principal variation, check out more suggested moves (what exactly is White's plan after 1...g6 2. O-O? What about 1...g6 2. Nb3? Do the two moves transpose?) etc.

Then after you check out all the lines to whatever depth satisfies you, you decide if ...h6 is better than ...g6. Maybe it is, maybe it isn't.

Correspondence chess players do this to great depth all the time.