Home >Mint-lounge >Mint-on-sunday >The flawed IQ of artificial intelligence

It was 18 years ago that a computer finally managed to beat the best human player of chess. That was a program called Deep Blue, which defeated then world champion Garry Kasparov. Kasparov recently wrote an article in the New York Review of Books, speaking about his defeat.

It has been many years—but even so, he doesn’t sound particularly disappointed or bitter about his loss, as you might expect had he lost to another grandmaster. Instead, he seems resigned and philosophical. “Grandmasters had already begun to see the implications of the existence of machines that could play with godlike perfection," he wrote.

Once computers got into chess-playing, he implies, with their “enormous calculating power", it was only a matter of time before one managed to beat even someone like him. Inevitable.

Why is this? Sprinkled through nearly every report about computers playing games—and in Kasparov’s article too—is the phrase “artificial intelligence". The implication is that as computers became more and more intelligent, there would naturally come a time when one would overtake the intelligence of humans. And since you need to be intelligent to play chess at any reasonable level, a computer that’s more intelligent than you would likely beat you at the game.

So, if Deep Blue beat the best human player, it must be more intelligent than him; and therefore, more intelligent than most of the rest of humankind. Me, certainly.

There are questions to ask there, but let’s keep that for a little later. A few weeks ago, there was news of another triumph for game-playing computers. For the first time, one such defeated a human master at the old Chinese board game Go.

Why is this news, you might wonder, when the chess victory happened 18 years ago and demonstrated what computers can do? Well, Go is considered a far more complex game than chess. This means that programming a computer to play it well was that much harder and took that much longer. So, Go presents the greater challenge to programming maestros.

One way to understand that is to ask how many possible board configurations there are in these games, and try to make some sense of that number. For comparison, let’s start with an estimate of the number of atoms in our universe. How many, do you think?

The answer is approximately 10^80, or one followed by 80 zeros. If that seems unimaginably huge, that’s because it is. There is no easy way to comprehend the magnitude of a number that large. But here’s the thing: the number of possible chess board configurations is itself unimaginably greater than that number. There are about 10^120 of them—one followed by 120 zeros. (For various reasons, the number of possible games is far less, though still enormous: 10^40).

If that gives you some sense of the magnitude of the task of programming a computer to play chess, you will be blown away trying to understand the challenge of Go. There are claims that there are 10^761 different Go board configurations. One followed by 761 zeros.

Enough with the ridiculous numbers. But the point remains: even just going by the numbers, instructing computers how to play these games is a complex, difficult task. From any given position, the computer must evaluate possible future positions and choose the best among them to make its move. The farther it can peer into the future, the better it gets at the game. But the farther it can peer, the greater the count of possible positions there are to check.

Of course, humans also play chess like this, but only to a limited degree. The top players can visualize the game a few moves ahead at best, and not even every possible move.

For one thing, it is physically impossible to consider every possible move. (As someone once calculated, looking just eight moves ahead presents you with more possible games than the Milky Way has stars.) For another, there are plenty of moves which it would make no sense to waste time evaluating. Even amateur players know enough about the game to discard—to not even think about—some such moves.

So, in that sense, humans actually don’t play like computers. And that raises an intriguing question about these computers. If they don’t play the game like humans do, if they now beat humans solely because they can evaluate many more positions much faster than humans can, is the term “artificial intelligence" applicable at all? If a computer plays in this brute-force way, is it actually displaying any intelligence?

Here’s the dilemma. Artificial intelligence began as a quest to understand the human mind. If we now have produced world-beating Go and chess computers, has that taught us anything about how humans think?

You don’t have to answer that. Because here’s what Kasparov had to say: he was “dismayed by the fact that Deep Blue was hardly what their predecessors had imagined decades earlier when they dreamed of creating a machine to defeat the world chess champion. Instead of a computer that thought and played chess like a human, with human creativity and intuition, they got one that played like a machine, systematically evaluating 200 million possible moves on the chess board per second and winning with brute number-crunching force".

He went on: “Deep Blue was only intelligent the way your programmable alarm clock is intelligent."

And yet it won. What will a truly intelligent—whatever that means—piece of software do to the games we play?

Once a computer scientist, Dilip D’Souza now lives in Mumbai and writes for his dinners. His latest book is Final Test: Exit Sachin Tendulkar.

His Twitter handle is @DeathEndsFun

Comments are welcome at

Subscribe to Mint Newsletters
* Enter a valid email
* Thank you for subscribing to our newsletter.

Click here to read the Mint ePaperMint is now on Telegram. Join Mint channel in your Telegram and stay updated with the latest business news.

Edit Profile
My ReadsRedeem a Gift CardLogout