Skip to main content

Verified by Psychology Today

Why Can't Computers Play Chess?

We challenge AI developers to build machines that can physically play chess.

On the surface, the question “Why can’t computers play chess?” is ridiculous. Deep Blue beat Garry Kasparov back in 1997. Deep Blue, the IBM Computer, won 2 games, Kasparov, the reigning world champion, won 1 game, and three games ended in a draw. Current chess programs have gotten even stronger – at the time of this essay, 20 years after the Kasparov – Deep Blue match, computer chess programs such as Rybka, Shredder, Fritz, and Stockfish are much more powerful than Deep Blue. Artificial Intelligence systems have convincingly demonstrated that they can play chess and can outplay humans.

At least, that’s what I thought.

Then I read a description of another IBM program, Watson, and its ability to play Jeopardy and to beat human opponents. What happens is that the Jeopardy questions are flashed on a screen to the humans and to Watson’s chips. But Watson can react within six milliseconds. No human is that fast. So Watson gains a big advantage just by its circuitry, over and above its intelligence.

The same effect holds true in chess. Computers can spit out moves much more quickly than humans – it’s pretty demoralizing for a human to make a move and have the computer make a counter-move in less than a second.

But hold on. Part of playing chess involves decision making, and yet another part is the perceptual-motor reaction. If the essence of the game is the decision making, then we’d want to keep the perceptual-motor time equal. And it isn’t equal. Currently, computers have a pretty big advantage in response speed.

The human has to drag pieces to their selected destinations or work through a drop-down menu to select a specific piece and a specific square in order to make a move, or click on a specific piece and then click on the square that the piece is to be moved to. In contrast, the computer just announces the move it wants and that’s it. Milliseconds versus precious seconds. That doesn’t seem entirely fair.

We find that in many tasks, the developers of AI systems have carved out a niche that the computers can perform very well, and persuaded us that this niche constitutes the entire task of interest.

With chess, in the interest of accommodating to the computer’s strengths, abilities, and limitations, we have bent over backward and given the computers a pretty sizeable edge. We have inadvertently handicapped ourselves.

I think it’s time to take that handicap back.

Playing chess involves moving pieces on a board, then hitting the button on the chess clock to set the opponent’s clock ticking. We should stick to that concept. We should challenge AI developers to build machines that not only select moves but actually make those moves – reaching out to the board to grasp a piece, then lifting and depositing that piece on a different square, removing any adversary piece on that square, then reaching over to the chess clock and hitting the appropriate button on the top, taking care not to knock over any other pieces in the process. AI developers should have to cope with pieces that are not located exactly in the center of their squares, pieces that are visually occluded by other pieces, and chess clocks that may move a little during the course of a game.

This challenge is pretty straightforward, and with the wonderful development of robotics, the challenge is clearly within the current state-of-the-art. Consider the progress in robotic surgery as smarter circuits get inserted into smaller and more agile robotic arms. Ending the response time handicap we’ve given chess computers might level the playing field a bit. And it would add a little extra excitement to the contests – especially to blitz chess. Let the games begin.