Carlsen on Computer Chess

Sort:
Oldest
ElectricEel

So I read with interest Carlsen's visit to the MIPT a month ago, and while I've been meaning to call out a comment he made there, I was otherwise distracted (read: coding and debugging my admittedly very primitive chess engine) in the past month to do it. Anyway, you can read up the interview he gave there on Chessbase, I cannot post a link here, but search for 'Carlsen MIT Moscow' and it should be the first hit.

 

The offending comment was this: 'Maybe at some point the rules will have to change in chess because the computers know too much. I think the most obvious solution is the question of stalemate. I think if you change stalemate from a draw to a win for the attacking part [sic] the game will be completely different and harder, perhaps, to work out.'

 

Say what? To me this 'solution' is anything but 'obvious'. Am I missing something? There are a couple of reasons why I do not understand this statement:

 

1.    In Chinese Chess, where a stalemate is a loss for the stalemated player, computers defeated grandmasters back in 2006. Incidentally, in Chinese chess, perpetual checks/perpetual threats (incessantly attacking an opponent's piece, causing a position to be repeated) are also losses for the checking/threatening side. Chinese chess actually has a slightly greater game-tree complexity than chess with an average branching factor of 38 as opposed to 35 in chess. There are fewer than 30 grandmasters out of perhaps more than 100 million Chinese chess players.

 

2.    I should imagine that the proposed rule change would likely most affect the endgame – KP versus K only will always be a win, for example. Queen versus Pawn on seventh rank, KNN versus K also come to mind. Right, so the evaluations of these will have to be changed. However, most strong chess programs rely on pre-computed tablebases (or bitbases). After those are re-computed, the rule change will have zero impact on the strength of computers in these elementary endgames. Elsewhere, say in the middlegame, the main change that will have to be implemented would be something along these lines:

 

Instead of

 

if (position.isStalemate())

 

      score_eval = 0;

 

let’s have

 

if (position.isStalemate())

 

      score_eval = SCORE_CHECKMATE – depth;

 

Clearly, some adjustments will have to be made for the evaluation of endgames (and for the transitions into those), but will these significantly affect the strength of computers? Furthermore, what is stopping these new evaluation functions from becoming just as effective as the ones we currently have?

 

Far from reducing the strength of top engines, I believe the actual effect may be quite the opposite. Firstly, top engines will be able to score ‘extra’ wins whereas previously they would have been held to draws. Secondly, chess players used to the traditional rule-set (where stalemate is a draw) may have more trouble adjusting to the new rules than the computers.

 

Now I am no expert on computer chess, but is Carlsen’s remark just, well...wide of the mark? Then again, he is the world chess champion (and I’m just your regular woodpusher), so maybe he has some insight that I don’t? What does the Chess.com community think?

ChessMN16

From a chess programming perspective, your post definitely makes a lot of sense. If we do indeed want to "beat" computers, we'll have to expand the board size of chess, most likely (and add a couple of pieces, too). But then again, why do we care so much about "beating" computers? I mean, for all intents and purposes, chess computers have already put us in almost the exact same situation that would happen had they "solved" the game.

Mr. Stockfish, Mrs. Komodo, and baby Houdini can all crush Carlsen without any problems, so what difference does it make to our game if it's 2800+ vs. 3100+ or 2800+ vs. 10000+?

Computers are already too strong, but that doesn't mean human players can't enjoy the game! 

ElectricEel

Yes, I concur, we can still enjoy the game no matter how strong the computer is.

My post is rather wordy, but the question I am asking is not whether the game should be changed or even how it can be changed, but the meaning behind Carlsen's suggestion. Simply put, it does not make any sense to me whatsoever.

He may be the world chess champion, but does he understand anything about computer chess?

ChessMN16

Yes, with all due respect to Carlsen (and more), his statement doesn't make much sense. By the way, I didn't find your post wordy at all; perhaps not many people have the time to devote themselves to reading it so they may end up saying "tl;dr," but I think it's definitely a very interesting topic.

Anyway, I think that Carlsen does understand a thing or two about computer chess because he plays like one (probably because he grew up with these engines)!

Actually, I find myself "suffering" from the same style of play - I rely on chess engines to understand a lot of complex positions, and this means that sometimes I pick some obscure move because I "feel" that that's how an engine would have played.

Anecdote time: I was basically crushing a 1930 FIDE rated player (OTB) for almost the entire game (won game --> drawn --> lost in time pressure), and when, in a five move time span, I went from winning to losing the game, I got very angry at myself and started telling myself "that was a plus 6 pawn position!" Later at home, Mr. Stockfish gave the evaluation of the winning position at ~+6.8...It's true that I have become much more mechanical in my thinking, but it has actually helped my game improve quite a bit.

ElectricEel

That's a very interesting anecdote which I am sure resonates with almost all chess players. For example, when I was much younger, I was in a very nice position against a 2050+ player (up a pawn and positionally dominant) and then promptly lost to a one move checkmate under time pressure. Infuriating, but time pressure (or even just pressure itself) can get to anyone.

With regards to Carlsen and computers - most if not all professional chess players use computers for analysis nowadays, but how many of them actually know about the how engines are implemented? After all, you don't have to know how a car works to drive one. I'm not even talking about fancy stuff like Dynamic Tree Splitting or anything here, but a rudimentary awareness of the minimax algorithm for zero-sum games.

In fact, I was conversing with a very strong WGM the other day, and she had no idea how chess engines work. This may partially be attributable to her young age; years ago before I learned about basic game theory, I was confounded as to what rôle evaluation could possibly play in a chess engine! EmbarassedNevertheless, it does go on to show that there are strong players who are not so familiar with the inner workings of chess programs, even though they might be using them on a daily basis.

Carlsen also remarked (perhaps revealingly) that computers 'know too much'. Computers 'know' only as much chess what the programmers give them. Indeed, Larry Kaufman(at least I think it was him, may be wrongly attributed) stated that the top grandmasters are still better at evaluating positions than the top engines. Actually I don't fully agree with this statement either. While this certainly may be the case for static evaluation, it is kind of besides the point, since the forte of engines lies in dynamic evaluation, which takes priority over static evaluation under most circumstances. Incidentally, in dynamic evaluation the best engines are demonstrably stronger than any human.

Anyway, getting back to the point in hand, computers are much better than humans in applying models, since with them there is no random bias, enabling them to take a more objective view of any given position. This is of course not to say they do not suffer from systemic bias (which is the task of the programmer/tester to minimise). They only have as much knowledge as the development team chooses to put in. The rest relies upon the search, move-sorting and evaluation functions.

In fact (dare I suggest it), a huge number of people take a rather naïve view of chess engines. When I told a friend (a casual chess player) that I shall be creating a toy chess engine, his first response was 'that's a lot of moves to put in the computer', believing that chess engines operate on opening books alone, and store all possible positions.

Forums
Forum Legend
Following
New Comments
Locked Topic
Pinned Topic