Originally posted by Ragwort
Top engine assisted players on ICCF for example are now saying they do not send moves without analysing to a depth of at least 40 ply - which requires more than a laptop by the way. Many say they no longer use human OTB databases on the grounds that the moves already represent the best engine assisted analysis or are simply weaker and thus too risky. Quite f ...[text shortened]... 40 ply? To what extent is a rule of thumb a reliable ball of string to take into the labyrinth?
So chess becomes a game of analysing an opponent's past games to work out which engine they are using and looking for potential weaknesses in the algorithm...
From the point of view of which move is best it does depend on one's opponent. Consider a chess oracle - 32 piece tablebases. From any given position it tells you which moves win, which draw and which lose. In any given position one can only make one's position worse in that if the position is drawn then no winning moves are available, but, in general losing moves will be. So, in a sense, any move which does not change the result is as good as any other, and all moves that change the result are bad, how bad they are depends on whether they blow a won position into a draw or a defeat or allow a drawn position to drift into a losing one.
The problem is that against human opposition some moves are more difficult to maintain the result against. This is extremely hard to quantify. Counting lines of play leading to wins, draws and losses won't really help as just because along a sequence of ten moves only one of the available moves each turn is correct (i.e. maintains the result) doesn't mean that finding that one move in twelve or so is particularly difficult (sometimes "only moves" are easy to find). Humans find defending difficult, they tend to miss entire lines of play, but machine calculation is more or less comprehensive so they tend to defend very accurately without becoming flustered. This means against a machine what is more likely to work is a positional approach preferably involving a gambit for some long term benefit the positional evaluation algorithm hopefully underestimates. Whereas against a human a gambit for a quick direct attack on the king is more likely to be effective.