The Pitfalls of Computer Analysis

In the first post on Gligoric - Fischer, Bled 1961, I highlighted the sequence starting 17...c5, based on the different evaluations of the two GMs: Fischer: '!' & Kasparov: '?!'. Since Kasparov called 20...d3, 'The key moment of the game', I wasn't too far off the mark. The resulting position is shown in the following diagram. Note that Black has already sacrificed one Pawn and is now sacrificing a second Pawn to create complications.

Bled 1961
Fischer, Robert

Gligoric, Svetozar
After 20...d4-d3

Gligoric played 21.Qxd3, and Fischer commented, 'A double-edged game would result from 21.Bxd3 Bd4+ 22.Kh1 Nxg3+ 23.Nxg3 Qxd6 24.Qc2 Bh3'. Kasparov, undoubtedly using the best hardware and software of the time, zeroed in on that comment and criticized Fischer of 'ignoring' 24.Qd2. He gave '24...g5?! 25.Qg2 Bd7 26.f4 Rxf4 27.Rxf4 Qxf4 28.Rf1 Qe3 29.Bf5, with a powerful attack on the light squares'. Jumping ahead a few years, my software prefers Fischer's 24.Qc2, and improves on Kasparov's analysis with 24.Qd2 g5 25.Qg2 Kh8. This takes the force out of 26.f4, by removing the King from the g-file.

Now I have to return to comments from Some Truths Cannot Be Proven?, where I expressed my doubts about this type of computer analysis. What's more interesting -- the move as played in competition (by Gligoric), the analysis without computers (by Fischer), or the progressive deepening of the position with the help of computers (by Kasparov)? I suspect a professional player would favor the first, a trainer the second, and a correspondence player the third. The problem with that third option is that it will never be finished.