As human beings we’ve tried to define our relationship with machines since they first appeared. When the development and spread of computers became unstoppable, we subconsciously began to fear that perhaps they would replace us in the future, taking control of our lives. It should come as no surprise, then, that we began to talk about a chess match between a human being and a computer as soon as we had the chance.
Games have always been used by human societies as an instrument for physical, emotional and mental growth, but none of them have been able to do it like chess. This game is a battle between two minds, which requires the desire to compete and to win using our brains: to think and train ourselves to think in different ways in order to adapt to the situations that arise during a game.
In 1985, at the age of 22, Garry Kasparov became the world chess champion by beating Anatoly Karpov (my idol), starting a long reign that lasted more than a decade. Shortly before winning the title, Kasparov had played in simultaneous matches against 32 of the world’s strongest machines capable of playing chess.
Humanity’s representative won all the games, quite easily. It was no surprise: the machines were still too weak. Twelve years later, Kasparov found himself fighting for his and our survival against a single computer in a match that Newsweek called “The Ultimate Brain Challenge,” so as not to put any pressure on the Russian champion.
From Turing to Deep Blue
The Deep Blue computer had been developed by IBM research department in the 1990s with help from chess grandmaster Joel Benjamin, who built the database of openings to feed to the machine.
But what was innovative about this project? Absolutely nothing! 70 years before there had been a great debate among the founders of AI on the subject of chess. Turing had developed a program in 1952 that could play chess or, rather, only played one game, and did it manually, instruction after instruction since there was not yet a computer capable of running it. In 2012, that machine was built as part of efforts to remember Turing and it played against Kasparov and lost after a few moves.
The founding fathers of AI believed that the way machines should play chess was through an understanding of the position and not through brute force, but any attempt to achieve this result failed miserably, and eventually, everybody agreed that brute force was the only possible path. In its second version, Deep Blue was able to evaluate 200 million positions per second, but from an artificial intelligence perspective, it was like an elaborate electronic alarm clock or any other electronic device: any chess program that can run on a mid-range smartphone today would be able to do better.
The match against Kasparov
There were two matches between the champion and Deep Blue. The first match began on February 10, 1996 and ended after seven days. Kasparov eventually won 4–2, but everyone remembers the outcome of the first game: Deep Blue won in 37 moves: there was no going back.
Kasparov admitted to making several mistakes in his preparation for the match and today believes it was totally inadequate. The main one was that he did not treat Deep Blue like any other human player: when he clashed with Karpov, for example, he knew all about his opponent and in particular knew all about his recent matches. Instead, in this case, there was no game played and documented by the machine against other humans. It was a black box completely unknown to human beings. IBM folded the rules in the contract in its favor: Kasparov’s imprudence in this regard was lethal to him in 1997.
For example, the human operator in front of Kasparov was connected through a terminal to Deep Blue, which was in a room that was inaccessible to the media and without cameras. Today we know that Deep Blue crashed several times, thus losing each and every one of its activity logs during those matches: it was necessary to restart it and provide the previous position. If our opponent had a heart attack during a real game, he would be taken to the hospital and we would win the game by forfeit!
Between the first and second matches, what had been a simple research project for IBM became a matter of life and death for the multinational that had now found itself at the center of attention around the world. The investments were huge and the second Deep Blue that clashed with Kasparov was much more powerful than the first, and once again it was an entirely unknown opponent. Psychologically, the Russian champion began to collapse under the pressure, seeing conspiracies everywhere: the result was 3-5 to 2-5 for the machine.
The real outcome
As we have said before, the outcome is not the most important aspect. It was inevitable that the day would come when the brute computing power of a machine would get the better of a human chess champion. Even changing the boundary conditions, the current world champion (Carlsen) would have no chance against a computer: the required level of accuracy would be too high for a human being.
Deep Blue, despite its victory, had provided no understanding of the mystery of human intelligence: he had “only” managed to calculate 200 million moves per second. For this reason, and mainly thanks to the spread of the Internet, human beings have continued to play chess even more than before, despite the most catastrophic predictions.
There is a very beautiful phrase, from Kasparov himself, that we must overcome our fears if we want to get the most out of our technology and, at the same time, bring out the best of our humanity. It should come as no surprise to us, therefore, that the Russian master licked the wounds of that clash and continued to work in the area of artificial intelligence as applied to chess.
His idea was simply brilliant: instead of insisting on clashing with a computer, why don’t we play together, combining our forces, against another man-computer pair? Putting together human intuition and machine computing power? You might wonder: what’s the point? We have all the brute computing force we need. Think instead, just for a moment, about how many of those 200 million moves per second were completely useless, pure combinatorial garbage. If we could use human strategy to guide the tactics of the machine, and our experience to shape the computer memory, we would certainly achieve better results.
The Kasparov’s Law
In 1998, the idea came to life in a match called Advanced Chess, where the participants were Kasparov and Fritz-5 against Topalov and ChessBase 7.0.
The results of that match were not exciting but showed a new way forward that later became incredibly widespread as the Internet. In 2005 a so-called free-style tournament was organized according to the same rules and a discovery was made there which is the real reason why I told you this story. Several grandmasters and the best chess machines participated in the tournament, but the winners were neither grandmasters nor a supercomputer. The winners were a pair of amateur chess players operating on a network of 3 ordinary PCs. Their ability to guide their machines won against the chess knowledge and computing power of the opponents. This result inspired what is now known as Kasparov’s Law
A “weak” man, plus a machine, plus a better process, are superior to a much more powerful machine left alone and, above all, to a strong man, with a machine and an inferior process.
But what do we mean here by process? We can think of it as a protocol stating how a human operator can interact with machines and support them. As Kasparov tells us: “We need better interfaces that help us better train our machines towards more useful intelligence”.
Garry Kasparov told his story in a book called “Deep Thinking: where artificial intelligence ends and human creativity begins”, I highly recommend it:
On YouTube, you will find a lot of his speeches, for example here:
but the first part of the 1996 match is not to be missed:
Suppose we are faced with a text written in a language unknown to us and let’s pass it on to an online translation tool. We all know that the result will not be perfect but we will use our human experience to extract meaning from that phrase. This is a model that is spreading more and more also because the machine, thanks to us, can learn from its mistakes. Always repeat the mantra that: “machines perform calculations, but we have understanding; the machines follow instructions but we have a purpose; machines are objective but we have the passion”. And I’m also saying this because we are fed up with fears of dystopian futures.
Keep following us!