Should We Trust Computers?
The question whether to trust computers might sound pointless since we really have no choice in the modern world. Think about it: if a computer of one of the nuclear superpowers goes berserk, then it can potentially end humankind as we know it. Since this is a chess website, the actual question I want to discuss is "should we trust computers in chess?"
Well, the only correct answer to this complicated question is "it depends." While for some people computers are synonymous with the ultimate truth, I have seen many kinds of computer errors. Let me start with one of the most bizarre cases.
About 30 years ago, when the dreaded "iron curtain" was finally lifted, many Soviet chess players got a chance to play tournaments abroad. So I was playing one of those big Swiss opens in Eastern Europe. A good friend of mine had a great tournament and before the last round the only thing he needed to achieve his final IM norm was another foreign opponent. In this case, even a loss wouldn't prevent him from getting the coveted title.
Unfortunately, according to the rules of Swiss pairing, he couldn't get such an opponent. I felt very sorry for him since it was really unfair that he couldn't get the title due to a stupid technicality. Surprisingly, he was in a good mood and practically celebrated his success. Answering my question, he puzzled me.
"You see, Greg" he said, "the word on the street is the computer in charge of the pairing really likes a five-star cognac!"
I don't know how he managed to make the computer drunk, but the next day he indeed was paired against the player he desired and consequently got his IM title. I had a bad tournament and don't remember much about my games there, but the main revelation for me, at 19 years old. was that computers like cognac!
Now whenever I see a weird tournament pairing or strange results in a local vote where people raise taxes for themselves, I immediately remember my chess friend's sly smile and hear his voice: "computers like a five-star cognac!".
But what about pure analysis of a chess position?
Can we trust chess engines? Again it depends on what kind of a position you analyze. Try to give your favorite chess engine a position like this and see the results.
You want a position from a real-life game? Here it is:
All the engines will insist that White should resign here; meanwhile it is a basic fortress. You can argue that while chess engines are notoriously bad in spotting fortresses, they are very good in everything else.
OK, here is a game hot off the presses:
Photo: Lennart Ootes, Grand Chess Tour.
Throughout most of the endgame, chess engines give Black a winning advantage to the tune of -3; meanwhile any chess player who's studied this kind of endgame knows that it is an easy draw. If you trust an engine in this kind of a position then you set up yourself for failure.
In the same endgame a chess engine severely criticizes Carlsen's move 60...gxh4 and instead suggests 60...hxg4. A nameless annotator analyzed the game for one of the major Russian websites and quoted the line as preferable way for Black to play for a win. In reality of course it is another basic theoretical draw:
As you can see, certain endgames are the weakest part of chess engines and you should definitely take their advice with a grain of salt!
Now let's examine the openings. We have to thank computers for many creative opening ideas, like this one:
After the game Aronian didn't hide the obvious fact that the whole 10.Rh4 idea was the result of his computer-assisted preparation. While we can find many examples like this, once in a while chess players pay a heavy price for their ultimate trust in the computer's power.
Who can forget the following famous game:
This is how Kramnik explained this catastrophe in his interview to the "New in Chess" magazine:
It was clearly a hole in my preparation, but it was a very strange hole. We analyzed this whole line (...) I was in the restroom and decided to play fast, not to give him time to think. To put further psychological pressure on him by responding immediately. I was checking variations and I already saw queen d3, but I thought it was just a perpetual. After all we had probably checked this with a computer so it shouldn’t be lost, because otherwise the computer would have shown that such a position is clearly lost. (...) This kind of thing had never happened to me before. You are White, you play all moves according to your preparation and you shake hands. This is something unbelievable (...) I don’t know if this ever happened in a world championship match before.
Have chess players learned from this debacle? Not really. Look at the next game where super-GM Radjabov's moves Nc6-b4-c6 clearly didn't make any sense and yet he played them only because he thought it was a computer's recommendation!
In this article you can find more details on this ridiculous Nc6-b4-c6 episode. But the key to understand what really happened is this:
"Radjabov said that he was following his notes, although he couldn't remember them exactly.
"Of course I looked at the natural move 13...Rc8, but I thought: we're living in the computer era," he said.
It is one thing when my students play a move that looks weird and they cannot even explain the point of the move, only saying that it was a computer recommendation. It is a totally new level of insanity when exactly the same thing is done by a top GM because "we're living in the computer era."
A popular stand-up comedian Elon Gold has a routine "ridiculous-diculous".
Rephrasing his punchline I can say that when Radjabov played Nc6-b4-c6 and drew a super-grandmaster it was ridiculous. But when GM Caruana played Nf6-g4-f6 and beat the former world number-two Aronian is diculous!
So, should we unquestionably trust our chess engines?
As you can see, some very strong chess players do. But I absolutely understand GM Svidler's joke after the Nc6-b4-c6 incident. When Radjabov mentioned that there was even a move Bb7-c8 somewhere in his notes, Svidler answered: "I would have quit chess. A mental institution awaited me immediately."
