I'll be getting back to you in due course about my poser involving an email chess game and the preferred move of the computer. Contributions to the comments box on that post are still welcome: in the mean time, it may help illustrate my theme if we look at a completed game, Rasstrigin-Horton, also played under the aegis of the IECG.
What makes this game interesting (apart from any intrinsic merit, and it was an exciting game) from the immediate point of view is that until move thirty, I was playing without a computer. The game began just after I moved to Spain, at about the time I learned that I was too late to register for any local chess in 2006 and would therefore have to kick my heels until the following January. So I registered for some email games - and this one lasted almost nine months. However, it was not until five of those months had elapsed that I was able to get Rybka installed, and my play before then tends to show it: missing the knight-and-queen manoeuvre that allows the capture on b7, then fumbling away a pawn shortly afterwards. Meanwhile, after the b7 capture my opponent's move coincides with the computer's choice on almost every occasion.
This was obvious to me even before I was able to get in Rybka and check it for myself, since White's whole approach, after going a pawn up, is very computer-like: sit on the material, start pushing pawns forward, it doesn't matter if your pieces are a little out of play. Which gave me some hope - as long as White continued to play like that, perhaps I would get time to move my pieces over to the kingside while White was fiddling elsewhere.
So I gave up a second pawn on a6 to get the knight out of the way, threw my remaining pawns at his king and prayed that just over the horizon, something would come up: and also that he would continue to follow the computer, hanging on to all his material advantage and not find some way of giving some of it back to exchange off material, then race the pawns home. (I've not looked at it recently, but it's hard to believe that White couldn't have cashed his winnings in at some time by capturing on c7, with three passed pawns easily compensating for a piece.)
So this is the position after Black's 31st move, to put it in the traditional manner - or the position in which Black got hold of a computer program, to put it in a manner both more modern and more pertinent. As it happens, this is also the point at which the evaluations started changing, and once a move or two were entered, White's advantage, having previously been over 1.00 (indeed, well over) suddenly dwindled to nothing.
An exposed king, chased by the heavy pieces, Black's bishop far better placed to join in than the knight - that's fine compensation for two pawns or even three.
After this point, while I was checking everything with the computer, I did not always play the preferred move, at least once choosing instead to throw more material on the fire in order to open lines (e.g. 38...e4 instead of a recapture on g2). I didn't win in the end - my king was probably too open itself to let that happen - but I got to force the draw, and to choose the moment when that happened. (What draw, you ask? Well, I'll let you or your computer find it. As it goes, the computer prefers not to force the draw and opts, instead of 43...Rg5, for 43...Qb2+ 44.Ra5 Qc3+. But I judged, rightly or wrongly, that the resulting ending, which the computer prefers for Black, is probably unwinnable - and potentially more risky for Black than White.)
What's the moral? The moral isn't that you can give the computer user a couple of pawns start and still outplay them. Not really. For a start, I could never have played the last dozen moves without a great deal of computer help - give it a go yourself and see how you get on. (34...Rf7 in particular would have been very hard to see, and the consequences of any given line were lost to me without computer help). Moreover, I would have been better placed in the first instance if I'd had a computer and my tactical shortcomings had not forced me to gamble that my opponent would look at the computer evaluations rather than the board. (Could he really not have taken on c7 at some point before I got organised on the kingside?)
Nevertheless, the game does show what happens if you do rely on the computer. An enormous points advantage can disappear, in a wild position, just like that. We know, playing OTB, that we should be cautious when winning, consolidate our advantages, be prepared to give back some material and avoid complications if we can. The odd thing is, we do that because we're worried about our human weaknesses. The possibility of blundering, of getting nervous and so on.
Yet playing with a computer program, it's the weaknesses of the machine which ironically make the same demands on us. Left to itself, it won't consolidate - won't cut down its arithmetical advantage in order to cut down on the risk of encountering the unexpected. Computers are supposed to be very good at winning won games. Perhaps it isn't necessarily so. We need to help them.
It's also supposed to be true, these days, that computers don't just grab material and don't underestimate the possibility of compensation. I am not sure, having played this game, that I believe this is true. I think it's quite possible that if a computer evaluation shows compensation that is substantial but inadequate - for instance, with a material advantage of 2.00 it shows an evaluation of something like 0.95 - then in fact the compensation may be rather greater than the computer thinks. Go for the line - which you never would if an equal-material position was assessed at 0.95 to the other side.
Which suggests, in its turn, that a possible strategy for playing a computer may entail a preparedness to sacrifice material where the computer's assessment fits that pattern. Where you would feel good about it over the board but are wary of the computer thinking it doesn't quite work. Be confident, because it may well be that your gut feeling is a better guide than the calculations of the computer.
It also goes, almost without saying, that the same is true when you are playing correspondence chess against people who will almost always follow their computer's advice.