Rant on The effect of Deep blue on chess

[Aaron]

Geschrieben von:/Posted by: Aaron at 10 March 2000 11:12:51:
I'm always steered clear of such threads in RgCC where people talk about how deep blue's victory over Kasparov has lead the general public to think that Chess has being solved..
Untill today, where my Lecturer made a off hand remark about how Humans with "gut feeling" is superior to statistical packages and data mining tools that merely crunch numbers.
He mentioned that this was changing with the coming of netural nets that can "learn to learn", and expert systems . Valid enough i guess except i think "Expert Systems" alone usually come with a knowledge base installed by a knowledge engineer ..
However I was shocked when he mentioned that deep blue was a exception that actually knew how to learn!!!
As far as I know this is hardly true..Deep blue has nothing beyond what free winboard programs has except superior speed and maybe/a big maybe a well tuned evalution function..!!
as for learning, I suppose the open book learning to avoid bad moves is the only form of learning most programs have..
I suppose this perception of Deep Blue as a super breakthrough computer is quite wide spread among the well read and educated profession public but with little interest in chess/computers..
A quick poll of some friends..

"It doesn't learn but just follows strategy/games preprogrammed into it" - This was the closest comment probably..I think he is refering to the fact that DB was supposed to have a database of all GM games..But he thought that DB somehow "learnt" from the games, in the sense of a Human learning from playing over GM games
It doesn't does it?

[Mogens Larsen]

Geschrieben von:/Posted by: Mogens Larsen at 10 March 2000 13:00:39:
Als Antwort auf:/As an answer to: Rant on The effect of Deep blue on chess geschrieben von:/posted by: Aaron at 10 March 2000 11:12:51:

"It doesn't learn but just follows strategy/games preprogrammed into it" - This was the closest comment probably..I think he is refering to the fact that DB was supposed to have a database of all GM games..But he thought that DB somehow "learnt" from the games, in the sense of a Human learning from playing over GM games
It doesn't does it?

No, Deep Blue didn't learn anything. It is possible for a program to compare the present situation on the board with a massive game collection. It requires a very fast searcher to make a reasonable number of comparisons in the time allowed and with the necessary quality. Deep Blue was a very fast searcher. This sort of pattern recognition is mostly known from neural systems, but you would want the system to make its own conclusions based on results. A lot of results. This isn't necessary if you have a database present. That is my opinion, but I didn't build the thing ).
Best wishes...
Mogens

[Thorsten Greiner]

Geschrieben von:/Posted by: Thorsten Greiner at 10 March 2000 16:11:41:
Als Antwort auf:/As an answer to: Rant on The effect of Deep blue on chess geschrieben von:/posted by: Aaron at 10 March 2000 11:12:51:

But he thought that DB somehow "learnt" from the games, in the sense of a Human learning from playing over GM games
It doesn't does it?

At least Deep Thought - DB's predecessor - did. They used a large collection of grandmaster games to tune the evaluation function parameter so that its move decisions match the moves of the grandmaster. You just might call this *learning*...
I think there is an article about this in 'Computer Chess and Cognition' or Levy's book. I don't have them, maybe someone can look it up?
-Thorsten

[Pete Galati]

Geschrieben von:/Posted by: Pete Galati at 10 March 2000 23:56:21:
Als Antwort auf:/As an answer to: Rant on The effect of Deep blue on chess geschrieben von:/posted by: Aaron at 10 March 2000 11:12:51:
[....]

"It doesn't learn but just follows strategy/games preprogrammed into it" - This was the closest comment probably..I think he is refering to the fact that DB was supposed to have a database of all GM games..But he thought that DB somehow "learnt" from the games, in the sense of a Human learning from playing over GM games
It doesn't does it?

Learn is too human a word to _really_ apply to computer programs. Crafty "learns" to an extent, it keeps a record of positions and gives them scores etc. in effect somewhat learning. It also learns in it's opening book by flagging certain openings as to their effectiveness, you could maybe compare it to Crafty putting a bunch of little post-it notes in it's opening book to remind it of what worked and what didn't work.
DB probably had a huge database of GM games used in it's opening book, but once outside of the opening book, it would have been calculating moves on it's own until the endgame.
So "learn" doesn't _really_ apply all that well. I think that it was a term that he used to over-simplify the situation so that he could talk about Deep Blue which is a bit of a Pop-culture item and people love bringing it up, it's an old overused buzz-word.
If you related what he talked about well then he probably had a very shallow understanding of what was happening. My take on it is the the mid-game was all Deep Blue's calculations, any positional understanding would have been the work of the programmers and consultants.
All of that is my opinion, that doesn't make it true or not.
Pete

[Aaron]

Geschrieben von:/Posted by: Aaron at 11 March 2000 15:29:52:
Als Antwort auf:/As an answer to: Re: Rant on The effect of Deep blue on chess geschrieben von:/posted by: Thorsten Greiner at 10 March 2000 16:11:41:

But he thought that DB somehow "learnt" from the games, in the sense of a >>Human learning from playing over GM games
It doesn't does it?

At least Deep Thought - DB's predecessor - did. They used a large collection >of grandmaster games to tune the evaluation function parameter so that its >move decisions match the moves of the grandmaster. You just might call this >*learning*...
I think there is an article about this in 'Computer Chess and Cognition' or >Levy's book. I don't have them, maybe someone can look it up?

That's interesting..I would call that "learning"...Wonder how they did it?