What's Wrong With Engine Analysis

It's something which you encounter a lot on the forums: people post games where they can't make heads or tails of the computer analysis. Alternately, there are those who post with great pride their latest immensely "accurate" epic (which turns out to not be much of an achievement at all--since the game was effectively over by move ten).

So why is it that the gadget often seems to be doing very little to boost our understanding, when really it should be bestowing its silicon-based wisdom upon everyone?

"Accuracy"

Okay, first off there's this whole business of nomenclature...

In chess, "accuracy" should really only apply to a position with a closed solution; e.g., something from Nalimov where all the lines have been analyzed down to mate. After all, if you don't know exactly where it is that you are going, how can you be certain that you are pursuing your goal with any great degree of "accuracy"?

The term is misused most flagrantly by the Silicon Swami both early on and later in the game (whenever things have become lopsided). An opening variation should never be dubbed "inaccurate." Dubious perhaps, or misguided--even (in the case of the Bongcloud) fairly looney. But not inaccurate.

Then too "accuracy" in chess has connotations of precision...of using one's endgame technique to grind things down to a win. In short, it implies that the position is still fairly close (otherwise you wouldn't really have the need to be particularly accurate). So anytime that somebody is--say--a queen down, accuracy becomes a fairly absurd term to use.

Indeed, whenever things get lopsided, the machine's pronouncements are often best ignored. I've seen king and pawn vs king endings where the pawn had a free road to queen...yet the Engenie castigated the winning player with an "inaccuracy" for pushing his pawn instead of making some king move! (which somehow or other would've shortened the solution by a move or two--I suppose). As usual in such cases of engine obtuseness, I could only snicker.

"Blunder"

Properly speaking, a blunder is not simply a mistake. It implies that something fairly elementary (and major) has been overlooked. Thus a dubious piece sac may well be called questionable, misguided, even unsound...but it is not a blunder.

I also have to let out a snicker every time the device charges somebody with a blunder in a dead lost position--advising them instead to (for example) give up their queen hopelessly. "Best is..." indeed! If any human were to make such a recommendation in person, such advice would earn nothing but a great big raspberry from me.

Still though, whenever someone does blunder away something--no matter what the material preponderance of one player or other--it should be indicated thusly. Not as an "inaccuracy" and not a "?!" move--or even simply ignored--but an actual, out-and-out honest-to-God hang. This is another of the complaints we see a lot on the forums--and justifiably so.

Where's your sense of scale?

It would be useful and instructive if Stockfish had built into it something like an "index of difficulty"...so you could find out whether the suggested moves were relatively simple (or immensely profound). But one major limitation to engines is that they seem to have very little sense of scale regarding these matters.

After all, from a computer's point of view, once a position is solved, that's all there is to it. All roads lead to Rome (so to speak)...and each alternative is fundamentally equivalent to another.

But we humans know how untrue that is. There are simple, easily solvable variations...and then there are ones which take a bit more (okay, sometimes a lot more) thought to uncover.

This unfortunate aspect of the silicon outlook can prove to be a decided shortcoming in defense (a trait at which the machines generally excel). I recall one time playing my microprocessor (back in the days when they actually still made those things!) a 30-minute game...and reaching an endgame where I had a queen versus rook and pawn--with only four minutes left on my clock.

Not much hope of me pulling that win off, right? But then the gadget seemed to flat-out hang its pawn; and a bit later on did the same with its rook. And thus I was able to mate it in time.

No doubt it had detected that I would be able to force the win of the pawn--then later on the rook--so it simply jettisoned them ("bowing to the inevitable"). But I hadn't actually seen the win of either; and if it had simply played a tougher line--as any human opponent would've done--I doubtless wouldn't have been able to notch the win.

You encounter the same effect every time you do the Puzzles here (primarily tasks with a set mate). There have been a number of occasions when the mate was fairly obvious, except for one tricky variation; but as soon as I punched in the first move, the engine would often choose one of the easy alternatives. A few times I admittedly hadn't actually been able to solve it right away...so I entered the "obvious" first move (hoping I would then be able to figure it out after that); only to discover that--once again--the device had chosen the easy way out. And so I never did learn what that mysterious line I missed was!

Oh yeah, and one more thing...

I also think the machine (and we) could benefit from a category called "missed opportunities." Time and again we see on the forums somebody posting a game where they made a quite reasonable-looking move, only to have it designated as a blunder. Why? It didn't hang anything, did it? Nope--but it turned out that there was a much better alternative to the move played. At any rate, it would be nice if that point was made somewhere in the analysis.

Human vs machine outlook

Okay, so I guess if you're sporting a 4000 rating, everything the gadget spouts makes perfect sense. But not for any of us flesh-and-blooders...and the whole point of engine analysis (it seems to me) should be that it's designed for those who are using it. Namely, us.

Sure, it would be nice to be able to find that mate in 15 every time (rather than having to settle for a meager +11). And computers (seeing most everything like they do) in fact have that luxury. But none of us humans do.

What we do have though is the ability to feel safe in the knowledge that--if we're a full queen up--we can fairly breeze along toward victory. Sure, we may not be able to announce a mate in 27--but the win will nonetheless be secured in the end. From the human point of view, a +11 is (or should be) the equivalent of announcing some far-off mate.

So if you see your opponent's queen hanging, by all means grab it (rather than spending undue amounts of time looking for some "brilliant" win). Because it's all too easy--if you do try to get fancy--to overlook some hidden resource (and be faced with a sudden reversal of fortunes). Then you'll be spending the whole rest of the game moaning to yourself: "I could've just won his queen!...instead I had to try and get cute..." We've doubtless all had those moments (and not just against silicon opposition).

At any rate, it would be nice if Mr Machine could be made more responsive to everybody's needs. So that we can all benefit from the sagacity of its pronouncements--rather than having to throw our hands up in the air one more time and letting out: "There it goes again!"