# Why can't computer-aided chess players wipe the floor with human players?

I don’t understand how a microprocessor, capable of performing millions of operations per second, can get defeated by a mere human at chess (think Kasparov vs. Deep Blue).

I know that Deep Blue won some of these games, but Kasparov also won as well.

How?

The quick answer is that while a processor can tackle millions of operations per second, in order to play perfect chess, you’d have to do kajillions of operations per second. Perhaps kajillions is an exaggeration but seriously, it’s a lot more than computers are able to do now. Probably you’re just not realizing how many possibilities there are when it comes to chess.

There are people here more knowledgeable than I in this realm, but as I understand it, chess is just the right level of complexity to make it interesting. A simpler game, like Reversi, is too easy for a computer to tackle, and the computer always wins. A more complex game, like Go, is too hard for a computer, and the human always wins.

Hey, I have yet to find a Reversi program that can beat me on a regular basis.

A chess player has, on average, around 30 possible ways to move at any one time, and an average game will have about 80 moves (I found those estimates through Google, not sure how good they are). If you want to look ahead ten moves and ten countermoves from a given position, that’s 30^20 possible moves to examine, or…

348 678 440 100 000 000 000 000 000 000 moves!

And that’s for looking ahead only 20 moves. Deep Blue might be capable of calculating hundreds or thousands of millions of moves per second, but that’s not nearly enough to consider all possible moves and find the ones that guarantee victory. At least not in the lifetime of a human opponent.

Well, a computer can wipe the board with me in Go, so humans don’t always win.

If you know how to play chess and do a little math, you can see that if you have about 30 possibilities for moving on any one move, and your opponent has a like number, that you get millions of possibilities only a few moves out. Grandmasters evaluate about 25 moves out, but only in generalities. Computers simply cannot do that yet, but can map out all possibilities fewer moves out.

Suffice to say that the current “Fritz” program, doing about 3 million moves a second evaluation, is a very good program, and if it had the processing power of Deep Blue, 200 million moves a second, that it could beat any human.

Something to consider, although I’m not sure it’s a factor here, a bazillion operations per second are only as good as the program they’re following.

I haven’t been able to find a good computer Go program, and I am just a mediocre player at best.

Anyway, what I really want to say is Go is different from Chess since you don’t think in concrete moves, you try to get a “feel” of the board (fuzzy generality) and then evaluate certain moves. Computers are exceptionally bad at doing that.

Actually there aren’t that many. A lot of those moves are bad so you automatically purge them from considerations. This is what the computer do too, with a pruning tree.

Now, of course, some moves that are initally bad can turn out to be brilliant moves later, so the program needs to look ahead a few (3-4 moves) to see if anything bad leads to something good.

This is, of course, done with 32-gambit processing.

[runs away, horribly ashamed.]

Maybe not, but something like that if you want the computer to play perfectly. Sure, in real life the software will use heuristics to reduce the search space. But as soon as the computer starts using heuristics because raw computing power isn’t enough, it can’t play perfectly any more and so it can be beaten. There’s always the possibility that five “stupid” moves actually lead to victory in the end.

Just thought I’d point out that a human brain is doing gazillions of operations per second. What exactly is going on is a bit of a mystery, but it’s obviously pretty sophisticated.

Also, keep in mind that a guy like Kasparov is a chess specialist. It’s pretty clear that the human brain is not engineered to play chess.

What if the contest were to successfuly get in the car; drive to the corner store; and buy a pack of gum? 99% of humans could clean the computer’s clock.

Computers are weak at strategy, which is as important a component of chess as tactics. They can’t “see over the horizon”. Whereas a human can call upon his instinctive understanding of positional nuances and formulate a plan, a computer can only crunch positional possibilities, and as Urban Ranger has stated, if it quits its analysis one move too soon, it can miss everything from a passed pawn to a forced mate.

So how does this explain the fact that I have yet to beat Gnu Chess on my Mac, even on the easiest setting? (Best I’ve managed is a draw.)

I’m obviously even worse than I thought…

You are probably tactically weak.

You are correct, it is not a perfect solution. It is adequate, however, for playing chess against humans, for humans aren’t perfect, either. No humans follow all possible moves, we prune our search tree as well. One thing we also do, however, is a lot of pattern matching (our brains are exceptional in matching patterns) by comparing board positions with ones we have seen before.

That is also something Deep Blue does IIRC, but computers aren’t good at doing this.

Has anybody ever actually built a better chess playing computer after Deep Blue? It seems that there was millions of \$\$\$ funneled into building huge, special purpose chess computers and, once it had beat the grand master once, interest just completely died down.

Would modern software running on Todays hardware completely wipe the floor of any chess player, human or computer 5 years ago?

Call me crazy but didn’t Deep Blue beat Kasparov in both the original and the rematch?

Also the latest incarnation of Chessmaster (9000) beat the US reigning champion and grandmaster.

I agree with Libertarian that humans win from their strength in strategy and the predictability of the analytical technique of the computer. When you get right down to it, The computer is still really just a tactician, albeit a phenomenal one. The computer simulates the oppostions counters like it’s playing itself - Kasparov may make a move that Deep Blue has already abandoned in pursuing if the long range strategic implications are strong.

A computer uses the ‘brute force’ approach to chess. It generates all possible legal continuations as far ahead as the processor can manage in the time limit for the game. Then it looks at this massive number of positions and basically counts the material (along with some crude stuff like how many squares each side controls). Finally it uses a simple minimax strategy to get the best possible result (it assumes its opponent plays the best moves).

A typical middlegame position has about 30 legal moves. So if you analyse 1 move ahead for each side, that is about 900 positions.
And so on - each look ahead by 1 move for each side increases the amount of calculation required about 1000-fold.
Only a computer has the patience and accuracy to play this way.

Humans use their experience and pattern recognition to form a judgement. A strong player will look at only 2-3 possible moves at each turn, and perhaps 4-10 moves ahead (that’s a total of both sides moves).

• you only assess a few positions, which is much quicker and less tiring,
• but you are relying on your judgement that the moves you ignore are really worthless.

The human approach works well in positions which are positional, where a calm build-up is required.
The computer is best in positions which are tactical, where a slip will lose material (or even get you mated).

The Kramnik match shows this dramatically. Kramnik cruised into the lead with quiet play, then lost a really exciting game. He was right that it was exciting, (and very sporting about his defeat), but it was a bad decision.
Incidentally there is ananlysis that suggests he resigned in a drawn position. See:

www.kasparovchess.com

for comments, then go to the match site (via a link) for the chess analysis.

Um, this is a reason why computers can beat 99% of all chessplayers!

‘30 moves’ is OK for the middlegame.
‘80 moves’ is absolute rubbish. A typical game is more like 30-40. In league chess, played over one evening, there simply isn’t time to play 80 moves. (And if 80 is the average, who’s playing all those 140 move games?!)

You can analyse 20+ moves ahead specifically in a King and pawn ending, where everything moves one square at a time. (I’ve done 28 ). But there is no chance of this in any other type of position. As above, I think 4-10 moves (and there will be mistakes!).

Maybe - it would be interesting.
The only way to beat such a beast is to use slow strategical build-ups.

I don’t know of any successful pruning algorithm.
Humans prune 90% of possible moves, computers 0%.

Deep Blue was a computer research project, not a chess project. IBM wanted to study how to multi-process. Chess was just the example they chose.
They also got a lot of publicity.
Standard chess computers are widely used for practice and analysis.

Human chess ability doesn’t seem to change very much. it is true that an occasional genius shows a new style is possible (Stenitz, Nimzovitch, Botvinnik), but the main advances have come from deeper analysis of human games by computers.
Really that’s the synthesis for best chess: human imagination with computer checking of variations.

You’re crazy!
Kasparov won 4-2 (in Philadelphia, 1996)

There have been several matches at kasparovchess between top computers and players like Smirin and Gulko.
I don’t know about the match you refer to, but an earlier Chessmaster version was rated about 20th:

http://www.elhchess.demon.co.uk/ehss.htm

Sorry, I don’t understand this.

I still don’t get it.

I understand that a computer is “only” capable of doing a couple of million operations per second… but compared to a HUMAN, who could perform 3, 4 operations per second at best, the computer should surely win every time?

glee… thank you… you’ve cleared it up for me.