Sunday, June 23, 2019

Grandmasters don't play twelve-dimensional chess

A couple of caveats before I start:

First, in case it's not clear by now, let me say explicitly that I'm not a grandmaster, or any kind of master, or even close to being any kind of chess master.  I have, however, watched a few play, browsed through master-level games and read articles and interviews on the subject, not to mention losing quite a few 5-minute games to master-level players and occasionally asking them to explain where I went wrong.

Second, while grandmaster is a title with very specific qualifications, reserved for the best of the best, I'm using it here as a stand-in for any strong player with the same general mindset and skills that I describe here, regardless of how well they actually do in tournament play.

With that out of the way, imagine a critical point in a cheesy movie:

"Welcome to my trap.  I knew you'd come here, just as I know that you will now try to use your secret radio transmitter to summon help, which is why I have taken the precaution of jamming all radio frequencies.  Don't even think about trying to cut off power to the jamming equipment.  It's encased in an unobtanium vault with battery power for the next three months.  Checkmate."

"Ah, my worthy opponent, I did not anticipate that.  But I did not need to.  There is no secret radio transmitter, only this aluminum box of curiously strong mints that I knew your scanning equipment would register as one.  No, I will not be needing a radio today.  I need only wait here and do nothing, while the new head of security you hired comes to my rescue.  Next time be more careful who you use for background checks.  You never know who they might be working for.  Don't bother summoning your bodyguards.  They're already safely locked away where they will do no harm to anyone."

"But ... but ... my plan was perfect.  I accounted for every last detail!"

Characters like this may be casually referred to as chess masters, but as far as I can tell this is not really how actual grandmasters play.

While it's true that many games are won or lost by one player catching something the other missed, it's not necessarily from thinking one move farther ahead, or from calculating every last possibility, but from recognizing the importance of occupying a key square, or opening or closing some particular line of attack, or knowing which rook to move onto a particular file.  This, in turn, generally comes down to who has the better plan.

There will certainly be points in a game where a grandmaster will pause and think deeply about moves, countermoves, counter-countermoves and so forth, and they can do this better than most players, but that's probably not the real distinguishing factor in their play.  Likewise, while tactics are certainly important in top-level games, the better tactician may or may not end up the winner.

One good illustration of this is the simultaneous exhibition, where a professional typically takes on a few dozen amateurs, moving from one to the next for each move (with occasional pauses for a longer exchange) and often winning every game.  If human chess were a matter of sheer calculation then winning every game would mean one person keeping all the possible continuations of 50 different games in mind, or being able to re-calculate them in the few seconds they usually spend at each board.

But of course this is not what's going on.  The pro is relying on experience, often looking for quick tactical wins by exploiting typical amateur mistakes.  They're still planning, just not in detail, maybe something on the order of "That knight is lined up with the king.  I can pin it, then bring my own knight to this square, put my queen on this diagonal and either attack the king directly or just win that rook if they don't notice I have a fork there.  Am I leaving any pieces hanging?  No.  Is my king exposed? No?  Great. On to the next one ..."

A human grandmaster can do this because they see the board not as a collection of pieces on squares but in a more structured way, something like a collection of patterns and possibilities.  Even if the full details of how this is done are elusive, there are hints, like the well-known experiments in which a grandmaster can reconstruct a position after a quick glance while a beginning player or a non-player takes much longer and makes many more mistakes.

Someone without much experience might think "there was a rook on this square, a pawn on this square ... or was it a bishop?" and so forth.  A grandmaster might think something more like "both sides have castled kingside, white's queen bishop is fianchettoed ..." or "this started out as such-and-such opening" or maybe even "this is like that game I lost to so-and-so and that rook is where I should have put it"

When it comes to playing over the board, a strong player will know many more possibilities to check for than a weaker one, and many more possible hazards to avoid.  One site I ran across gives a checklist of more than two dozen things to check for on every move, just to play a strong amateur game.  I have no doubt that this makes for stronger play, but I personally don't have the patience for this, or the experience for it to have become second nature, so it should come as no surprise that I'm not a strong player.

The most important factor, though, is planning.  If you just play likely-looking moves against a grandmaster, but without a clear plan, you'll almost certainly find yourself in deep trouble before long because they did have a plan.  That innocuous bishop move, that weird-looking pawn push and the shift of a rook from a reasonable-looking square to a different reasonable-looking square were all setting up the attack that's now making your life miserable.  If only you could move your king out of the way -- and that's what that bishop move was really about.

As I mentioned in a previous post, AB engines really do play the kind of insanely calculating game that we associate with the cliche chessmaster character, while NN engines do the same sort of pattern recognition that allows a grandmaster to size up a situation without having to do a lot of calculation, but neither is doing the sort of large-scale planning that a human grandmaster is.  It's also worth reiterating that, while even early chess engines were able to out-calculate human players, it took decades before they could consistently win games against them.

In my view, it would be a mistake to think that the planning grandmasters do is some sort of humans-only ability that computers could never emulate.  There's no reason a computer couldn't represent and execute a plan on the order of "Put your bishop on this diagonal, push these pawns and go for a kingside attack involving the bishop, rook and queen," without exhaustively checking every move and countermove.  It just hasn't proven to be an approach that computers can use to win games.

1 comment:

  1. Chess is a war game, so it's worth quoting a warrior:

    "Plans are worthless, but planning is everything." --Eisenhower

    Also attributed to him: "No battle plan ever survived contact with the enemy."

    Moral: If you don't have a plan, the other guy's plan will work.