
Last week we showed you a chess position presented by the famous mathematician Sir Roger Penrose – probably composed by his brother, GM Jonathan Penrose. It was devised "to defeat an artificially intelligent (AI) computer but be solvable for humans."
The position, which Penrose writes "scientists have constructed in a way to confound a chess computer," looks like a clear win for Black. However an average chess-playing human can see that a draw is possible: Black has no legal moves except with his bishops, so all White needs to do is simply move his king around and let Black make pointless bishop moves. There is absolutely nothing Black can do to force a win.
A chess program will not "see" the draw but display in its evaluation a very big advantage – 25 to 31 pawns – for Black. "The three bishops force the computer to perform a massive search of possible positions that will rapidly expand to something that exceeds all the computational power on planet earth," the Penrose Institute wrote, making a big deal out of this. But the position and the logic behind the experiment is not compelling: any chess program, playing the white side, will hold the draw flawlessly, even if it displays high negative values for White.
When writing the article I was reminded of a similar stuation that had occurred almost four decades ago:
David Levy - CHESS 4.8, Hamburg, 7.2.1979
The Scottish champion IM David Levy was playing against the most powerful computer in the world, and after 73 moves had reached the above position (the entire game is described below). CHESS 4.8 displayed an evaluation of plus nine pawns for White, but it defended the game perfectly, and Levy was unable to win.
The position is a theoretical draw, as chess engines today will instantly recognize – because the endgame queen vs c or f-pawn is apparently part of their chess knowledge. For CHESS 4.8 this was not the case, but the program was able to survive by doing a pure brute force search and simply not letting White capture the pawn. The defence, as given in the annotated game below, is to only allow the pawn capture on c2 when the black king is on a1, so that it would be a stalemate.
The situation is similar to the Penrose example. Here any human player who has been following centuries of chess research can immediately recognize that the above postion is a draw, but the computer did not and evaluated it as a dead loss. But: it played the position perfectly and defended it to the draw it did not "see".
I want to use the above position for a trip down memory lane. As a rookie science journalist I was working for the German TV station ZDF in 1979, having a lot of fun doing so. One day I read that computers were playing chess, and in fact one of them had beaten a grandmaster in a blitz game. I took the idea of doing a documentary on the subject to my boss, the eminent science journalist Hoimar von Ditfurt, who immediately commissioned a film.
At the time IM David Levy, who was Scottisch Champion in 1968, had just won a well publicized bet against a number of famous artificial intelligence researchers (John McCarthy and Donald Michie amongst them). They had predicted, in 1968, that in ten years a computer would beat the world chess champion. Levy bet them £1,250 that no computer would win a chess match against him in those ten years. In 1978 he had survived all attempts. As part of our documentary we decided to invite Levy to Hamburg to play an exhibition game against a computer, the strongest in the world at the time.
The was program CHESS 4.8, running on a CDC Cyber 176 located in Minneapolis, programmed by two pioneers of computer chess: Larry Atkins and David Slate, computer scientists at Northwestern University
Producing the ZDF computer chess documentary was quite exciting – I got to go to Northwestern (Chicago) to interview Slate and Atkins, and to Moscow to talk with Mikhail Botvinnik, who was also experimenting in chess programming.
In our studio in Hamburg we hooked up the Cyber 176 to a sensor board and an industrial robot arm. Using a satellite line the computer actually sensed the moves made by Levy and used the robot arm to execute its own moves. German GM Helmut Pfleger commented for the studio public and the TV audience.
The program was a great success (and dragged me into the chess world). Europe's largest news magazine, Der Spiegel, reported extensively on the game and our documentary, and in fact staged a second match between CHESS 4.8 and Viktor Korchnoi a short while later (Korchnoi won comfortably).
In our ZDF program we offered to send the game notation, annotated by David Levy, Helmut Pfleger and the computer itself, to viewers who wrote in. A week later I was called to the studio, where a somewhat distraught manager led me into a room which had a number of sacks of mail: a total of 95,000 viewers had requested our notation! I produced a three-page printout and we hired a bunch of students to send out copies in this record-breaking action.
I still have the printout, but it is all in German. So I translated it into English – back into English in the case of Levy's notes – and have prepared a replay board for our readers. It is meant as a document to show the state of computer chess in the very early days. Remember that the Cyber 176 was probably as powerful as the hardware in your TV set or microwave oven, and that many of the algorithms that have pushed chess engines today to 3300+ Elo strength had yet to be discovered. Still, it is an interesting game and of historical interest.
|