Until recently, the chess machines had specifically programmed chess tactics into them. For example, they knew a queen was worth more than a rook, because that information had been programmed into the system.
Yet now, researchers are attempting to build chess computers simply by programming them with the most simple rules of chess. After this, they are supposed to figure out everything else on their own, which means they can come up with completely different techniques and plays that they can also show others.
It’s an ongoing debate as to whether chess should be considered a sport or not. What is certain, however, is that the nervous exhaustion experienced after a chess match is on a par with exhaustion felt at the end of a track race.
This is due to the fact that chess is ultimately a psychological game.
Since 2003, Kasparov has been studying chess matches played by famous grandmasters, including his own. He laid out his discoveries in his book My Great Predecessors and argued that even the best chess players make many tactical mistakes. Of course, it’s not because they don’t know any better. It’s due to the fact they’re anxious or psychologically worn down by their opponents.
The German chess player Emanuel Lasker, who was World Chess Champion for 27 years between 1894 and 1921, epitomized the psychological approach to chess.
The theory was that tactically, the right move does not actually have to make the most sense, but that it should make an adversary as nervous as possible. Once a match starts, this style of play requires close study of an opponent’s game. Weaknesses must be established, as must the actions most likely to destabilize him or her mentally.
No such rules apply when computers play chess.
A human will always have a psychological reaction to the stress of a match. But computers are emotionless, both in and out of chess games. For them, it’s purely a question of strategy.
By 1985, computers were already powerful enough to compute every possible combination of moves over the next three or four turns and pick the most appropriate one. But, if the player was able to strategize at least five moves ahead, it was quite possible for him to defeat a computer.
It’s a commonly held belief that success rests upon innate talent. But, as Malcolm Gladwell wrote in Outliers, this is debatable. What matters is many thousands of hours of practice.
For humans, Gladwell’s thesis holds some truth. But as far as artificial intelligence is concerned, there’s no uncertainty. Brute force is what counts.
Donald Michie, a British researcher in the field of artificial intelligence and machine-learning pioneer, was among the first to really take advantage of this when he began pairing computers with large amounts of raw data. He tested the concept in the game of tic-tac-toe in 1960.
Normally, you might give a computer a series of rules to apply in a game. But Michie gave the computer numerous examples of game moves and allowed it to work out basic principles from there.
We do see this kind of machine learning method with modern translation programs like Google Translate all the time. Probably they do not know anything about the languages at all. Instead, millions of example sentences with corresponding translations, produced by men, have just been served. Based on these, a correct translation of any given text can be compiled.
Such systems are not infallible, however. Computers that rely of huge amounts of data can also make massive errors.
In the 1980s, Michie tried to create a chess-playing machine. He and some other researchers stuffed the computer with raw data: millions of chess moves played during grandmaster games.
The computer became a great player, but one that would occasionally do baffling things, like suddenly sacrifice its queen for no apparent reason.
What had happened was that the machine had learned from the grandmasters that it could be a move to sacrifice the queen which meant victory. But obviously, the program had failed to realize that the gambit only operated when there were several other parameters in place. It was as if it had grasped anything, but nothing at the same time.
For many people, a game is just a game and nothing more. But there are also those who burst into tears or see red if they lose. Kasparov isn’t ashamed of this behavior. As far as he’s concerned, to be a good competitor, your dislike of losing has to be greater than your fear of competing. Otherwise, you’ll just quit.
But those games were against humans. Playing computers was another story entirely. Kasparov lost a game to a computer for the first time in May 1994, in Munich. Its name was Fritz 3.
At first Kasparov performed well and won an advantageous spot. Even then, he made only one unsound strategic move. The machine was back in the game, instantly. The error had been apparent. It was a blitz chess tournament, a style that sometimes takes mere seconds for players to think over every move. Although Kasparov ultimately won the whole tournament, it was the first time that a computer had managed to defeat a chess world champion.
Kasparov went on to face an even more powerful computer – IBM’s Deep Blue – under tournament conditions a few years later in 1996. This time it was a full match over 6 games. Kasparov won the first match, but at the rematch the following year, Deep Blue was the victor. It was a close match, but in the end, Deep Blue could calculate so many possible options for each move that Kasparov couldn’t keep up. It marked a major victory for artificial intelligence.
To Kasparov it had been a moment of recognition. He could now be defeated regularly by machines, and in the future they will certainly only get more strong. And with this Kasparov resigned himself to the losing experience.
As spectators, we generally see the glamorous sides of competitive sports. But behind the scenes, in the shadows, foul play is hardly unusual. Competitive chess is no different.
Those anecdotes can seem quite amusing from a distance. Take the bitter rivalry of the two dominant players Anatoly Karpov and Viktor Korchnoi in the 1970’s. At the 1978 Philippine World Championships, in an effort to hypnotize or confuse him, Karpov employed a psychologist named Dr. Zhukar to look at Korchnoi intently during the match.
Korchnoi refused to be outdone. During the same championship, he recruited some Indian sect members to meditate and stare at Karpov and his psychologist, in an attempt to intimidate them. What’s more, each of them constantly accused the other of cheating and would demand to have various objects the other possessed investigated. These included Korchnoi’s chair and glasses, and, famously, Karpov’s yogurt.
Computers haven’t removed the presence of foul play these days, it merely occurs in another type. For example, a certain amount of human interaction during matches is permitted on the computers. Technicians iron out bugs, reset machines when they crash and change the evaluative functions of the machines between games.
At the time of Kasparov’s famous rematch with Deep Blue in 1997, these routine modifications were already accepted. In fact, Deep Blue crashed twice during the six games and was restarted on both occasions. Because the restarts erased the computer’s memory tables, it would have led it to make different moves and decisions than it would have had it not crashed. Illicit triggering of this kind of event during matches could be a way for technicians to give computers an unfair advantage. As a consequence, technicians’ interventions are now more firmly regulated.
Chess is a complex and beautiful game but eventually it proved to be easy enough to master computers. As much was in fact when the author defeated Deep Blue, using only the computing resources available in the late 1990s. Computer science’s next challenge will be to get machines to learn more complex board games, with more squares and variables than chess. Anything like the Chinese game Go is going to do just fine.
Artificial intelligence is increasingly outgrowing human intelligence. For more than 20 years it has had the potential to beat world-class chess players at the game, but much more is to be anticipated. Right now, machines are primarily using brute computing force and their ability to process large quantities of data to do so. But a new revolution in artificial intelligence is in the offing. If computers can start to analyze the data, to formulate questions from it, and to develop solutions independently of human input, then we will have truly entered a new era.
Check out my related post: Are you thinking in bets?