[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Fwd: Re: Strong players only?]

> Dear Bruno Bouzy,
> 
> If my reply to your comment went out only to you, I'm having trouble
> sending it to the whole computer-go group ... could you post it to the
> whole group?
> 
> Merci / Bob
--- Begin Message ---  

Bruno Bouzy wrote:

Darren Cook wrote:
>How many people agree or disagree that a poor go player cannot be a
>good go programmer, and why do you think that?

Better to be as strong as possible in computer.
A poor programmer cannot be a good go programmer.
My two cents.
:-)
Bruno

Computer chess is "settled" now ... that is, the success computer Go programmers are now striving for has been pretty much objectively achieved with Deep Thought's designs leading to Deep Blue's implementations.

But "How Computers Play Chess" by David Levy and Monty Newborn (1991, Computer Science Press/W.H. Freeman) was written during Deep Thought's career and contains a lengthy and detailed historical survey of computer chess, and the authors conclude that throughout the history of computer chess, fairly weak, often even naively lousy players achieved the most striking results and advances in chess programming.

I suspect many of you already have this book. (Not having it, it seems to me, is like wanting to ride a zebra without ever having ridden a horse.) But here is an amusing anecdote which ends with a conclusion about Darren Cook's question:

*   *   *

"The weakest player ever to develop a strong chess program was Claude Jarry. Jarry developed L'EXCENTRIQUE while an undergraduate student at McGill in the late 1970s. L'EXCENTRIQUE's shining moment came at the 1980 World Championship in Linz, Austria where it defeated the reigning World Champion CHESS 4.9 in the first round. Jarry, a fantastic programmer, knew so little about chess that the only way he could tell whether L'EXCENTRIQUE was winning or losing was by reading the score provided by the program after each move. His success dramatizes another lesson of computer chess: A programmer can develop a program much better than himself. [emphasis mine]

"Advice from a strong chess player can hinder a project unless the player has a keen comprehension of computers. Without some understanding of the correlation between chess concepts and the amount of programming work to implement them, a strong player can waste a programmer's time. In the past, many strong players were unable to understand the complexities of the search process  sufficiently well to suggest realistic ways to improve tactical play."

*   *   *

Levy and Newborn offer several other technical analyses of why a programming (and in the case of Deep Thought's Hsu, a circuit-design) whiz is the key member of the team, the player-master whiz a less significant member. But here are two analyses of mine:

1. The Machine, whether by brute force or logical algorithm, is the entity that's supposed to come up with the magical, winning answers, not the Human. That's what computers are for; that's how they do certain classes of intellectual jobs better (faster) than humans. The ideal game-playing machine should win based chiefly or entirely on just The Rules of The Game as First Principles. (Not at first, of course; but by learning from thousands of games worth of its mistakes.)

2. Why would a person want to invest so much of his or her life to create a powerful chess or Go-playing machine?

Here I think the lousy chess/Go player has the great psychological edge. Decades of being humiliated by 13-year-olds ... that terrible moment when you finally realize (like Solieri in "Amadeus") that the best you'll ever be is a mediocre player ... such people who also have programming skills just naturally develop an obsession to build a chess or Go Frankenstein monster for revenge. And revenge can be every bit as sweet as you fantasize it to be; look at what an unsportsmanlike whining baby Kasparov was when Deep Blue beat him.

I'll never even get the chance to be beaten by the word's greatest Go master ... but I could write a program ...

Bob Merkin

http://www.javanet.com/~bobmer/ --- End Message ---