Computers: Grandmaster plots the next move: Tim Jackson looks at the human skill and expertise behind Kasparov's defeat by the computer chess program Genius 2

Click to follow
The Independent Online
AT FIRST SIGHT, it is tempting to see last week's defeat of the world chess champion by a computer program as evidence that the human spirit is being subjugated to the raw processing power of silicon. Behind every impersonal computer program, however, there is a group of human programmers. The program that humiliated Garry Kasparov so comprehensively was the work of one man: Richard Lang.

Mr Lang, arguably the world's leading expert on computer chess, is a refreshing throwback to the old days when computer nerds designed products in their garages that changed the world. A physics graduate of Imperial College, London, he first became interested in computer chess when he was working as a researcher at British Gas and started writing programs to play the game at home in his spare time.

His lucky break came when one of his programs won a European competition for microcomputer chess and a small software house offered him a full-time job. Within a few years he was confident enough of his expertise to go freelance - and for the past decade he has written programs from his house in Dorset and marketed them with the help of distributors. The victory over Kasparov of Genius 2, his latest program, is likely to provide a huge boost to sales.

Genius 2 is a tiny program, occupying only one-tenth of the space available on a single floppy disk in an age when software is growing so complex that even a word-processor or spreadsheet can need five or more full disks. For all its technical wizardry, however, Genius 2 is not a mass-market product. Priced at about pounds 90, more than the average chess program, it is sold through specialist computer chess shops rather than on the high street. This is probably because Mr Lang has devoted most of his efforts to allowing the computer to play chess as well as possible. No space is left for gimmicks like displaying the board in three dimensions, or including a facility that allows the program to announce its move through a loudspeaker. All the same, Genius 2 sells about 8,000 copies a year worldwide.

Mr Lang is reluctant to give away the secrets of how it works and competitors who want to find out will have a hard time in doing so. To allow the computer to play at top speed, he wrote the program in assembly language as opposed to the higher level languages which are easier to understand but less efficient. Genius 2 is also copy-protected. 'It would probably take a good hacker a year or more just to get a look at the code itself,' Mr Lang observes with satisfaction.

In principle, however, Genius 2 works like any other chess program. It considers all the possible moves from the position it is faced with - usually there are about 36 choices - and then all the possible moves and countermoves thereafter. The millions of different possible positions resulting from this search are then given a score and the program chooses the move most likely to lead to a high-scoring outcome. One of the secrets of a strong chess program is the ability to choose quickly the lines of play that can be discarded as of no interest and concentrate on deeper analysis of the rest.

Genius 2's edge over other programs lies mainly in its high speed - which comes from a highly efficient way of 'pruning' the tree of possible moves and counter-moves. Mr Lang has a bias against ruling out avenues of investigation early on, in case an apparently bad move could produce a big advantage in several moves' time. On average, the program looks five moves ahead; for the most promising possibilities, it may consider as many as 32 moves and counter-moves. The difficulty is that there are so many choices at each point that inspecting each branch just one more move ahead can halve the program's playing speed.

The secret criteria that the program applies in rating one position over another vary in complexity from a simple count of the number of pieces on each side, with the different pieces being given weightings that vary at different stages of the game, to an assessment of the strength of the pawn structure and the broader competitive position.

Since tournament chess specifies time limits for a given number of moves, Genius 2, therefore, saves up calculating time at the start of the game by playing its first few moves from a library of standard opening tactics. If its opponent plays a particularly conventional game, all the first 20 moves could come from the library in a matter of microseconds. Players who want to give the program a rougher ride will choose unusual openings, forcing it to use more thinking time in the game's earliest stages.

The contrast between the program's performance at different speeds gives a clue to the huge difference between the way that humans and computers play chess. Because human grandmasters probably instinctively look deeply at the consequences of a small number of attractive moves, only later, considering the results of less plausible tactics, they tend to be caught out by computers more easily when the time available is short. Kasparov was beaten at blitz-chess - a five-minute version of the game - earlier this year by Fritz 3, another program. The human advantage carries more weight as the games become longer.

Since familiarity with the program makes it easier to beat, Mr Lang might have been expected to limit access to the latest version of it to the small number of international masters and grandmasters who have helped him develop it. In fact, he does no such thing: Genius 2 is a commercial product, he says, and the version of it that appears in the shops is the best chess program that Mr Lang has written to date. The amendments made before its victory against Kasparov will be incorporated into Genius 3, which Mr Lang will deliver to his distributors next week. Within a month, therefore, owners of standard personal computers will be free to play with the software that beat the world champion.

What next? Mr Lang has an interest in the Oriental game of Go, which is no less formal than chess, but has to date proved so hard to program that the best software is still beaten by mediocre players. But first he wants to write a program that can beat the human world champion at standard tournament chess. Before the Kasparov game, he expected this task to take another 10 or 12 years. Now he is more optimistic and thinks it can be carried out in only five or six.

There is only one cloud on the horizon. In the past, increases in computer processing speed used to translate clearly into better chess performance. That may no longer be certain in future; and Mr Lang may find that moving even to a microprocessor four or ten times as fast as the 100 megahertz Intel Pentium chip that helped beat Kasparov produces only a modest improvement in playing skill. If that is the case, the deadline will have to be put back - and Mr Lang and other chess programmers may have to teach computers to mimic more closely the way humans play chess.

IndyBest product reviews are unbiased, independent advice you can trust. On some occasions, we earn revenue if you click the links and buy the products, but we never allow this to bias our coverage. The reviews are compiled through a mix of expert opinion and real-world testing

Comments