Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / What did we learn from computer chess?
- - By Kappatoo (*****) [us] Date 2017-09-22 13:58
I was wondering what are the most surprising/interesting/important things we learned from computer chess.
- We certainly learned some things about chess - for instance, there were surprises discoveries about certain opening lines, about endgames (from tablebases), etc. What do you think are good examples here? And were there any more general insights - for instance, about the value of material, about piece values, about how close or far humans are from perfect play, etc.?
- Did computer chess teach us things about computer science? What was it?
- Did computer chess teach us things about the human mind? If so, what?
Parent - - By user923005 (****) [us] Date 2017-09-22 15:55

>>


I was wondering what are the most surprising/interesting/important things we learned from computer chess.
- We certainly learned some things about chess - for instance, there were surprises discoveries about certain opening lines, about endgames (from tablebases), etc. What do you think are good examples here? And were there any more general insights - for instance, about the value of material, about piece values, about how close or far humans are from perfect play, etc.?
- Did computer chess teach us things about computer science? What was it?
- Did computer chess teach us things about the human mind? If so, what?
<<
On the first point, I think we have a long way to go, but we have learned some very interesting things.
We have not learned, for instance, which is better, 1.e4 or 1.d4, a pretty simple and fundamental question.  I do not think the answer is certain, though I lean to 1.d4 (Berliner would smile, Fisher would wrinkle his forehead).  There have been interesting surprises.  Analysis of Philodor's games showed that (at least for the author of the quote "Pawns are the soul of chess") pawns were extremely valuable.  There was a tool set posted to the TalkChess programming forum that would derive the value of chessmen from a body of chess games.  When you aimed the tool at Philodor's games, you got a giant answer for pawns.  I think that computer chess has confirmed some long held beliefs (e.g. the approximate worth of a bishop pair in the endgame is about what the masters thought).  It has also shown the value of material imbalance and understanding material imbalance (one of the true innovations of Rybka was the material imbalance table).  Larry Kaufman wrote the most important paper on this topic.

About computer science:
The drive to play better chess drove many innovations in search technology.  This technology is applicable not only to chess but to the generic problem of search.  If I have an objective function that can evaluate a situation and grade it, and if I have some kind of gradient to choose a path, then the algorithms learned in chess can be used to find solutions quickly.  This algorithm could apply to an astronaut who is lost in space and has some various kinds of measurement to choose from as to how to regain his path.

About the human mind, I do not think that chess has taught us the answer to important questions like "How do we think" but it has exposed things like "What are we good at?" because humans are superb at pattern matching and also at long term planning.
Parent - - By Kappatoo (*****) [us] Date 2017-09-23 00:33
Thanks for the input!
Concerning advances in search: What are examples of techniques specifically developed for chess, which were then applied to other fields?

On humans being good at pattern recognition: I find it hard to measure this - good compared to what? I guess one could say we are good compared to computers. But given the recent successes with deep learning (e.g. AlphaGo), I wonder how long this will still be the case.
Parent - By user923005 (****) [us] Date 2017-09-23 02:47
Game programming search (things like A-*, C-*, Alpha-Beta, etc.) are simply graph search problems.  They happen to work well in chess, but ANYTHING that traverses a graph can use these ideas.  Graph search is used by lots and lots of industries, and it was usually game programming that drove some of the important algorithms.

The military uses this kind of thing a lot.

I cannot possibly predict all the ways that things like this can be used, but let me tell you about a surprising use of technology.

The University of California at Berkeley wrote an algorithm to simplify boolean logic circuits.
I used this algorithm to simplify SQL queries (which are, after all, complex and possibly recursive applications of boolean logic in the where clause).

Who would think that something designed to make less components on a printed circuit board would make SQL queries run dramatically faster?

So too, there can be all sorts of interesting usage of game theory applied to other fields.  I do not know what all of these things might be, so it is kind of open-ended.

I expect, at some time, to apply graph search to SQL problems also.  Some SQL queries have giant collections of nodes and edges, perfect for that sort of thing.
Parent - - By Labyrinth (*****) [us] Date 2017-09-22 23:03

>And were there any more general insights - for instance, about the value of material, about piece values, about how close or far humans are from perfect play, etc.?


I've asked this kind of question before and unfortunately the answer seems to be something along the lines of "not much". Here's a couple things that come to mind:

I think that computers have shown is that there is a lot more to defense than we ever realized. If you remember the tournament style match with the computers that included Karjakin and some others, they would sometimes get an opening advantage with their superior theory, but found themselves unable to convert or even have the tables turned after insane defense.

Personally, defense is where I feel the really big difference in playing strength between myself and the computer. Not saying I'm a particularly strong player or anything, but I'm not all that amazed by the way I lose in a normal game against a computer as the same thing would happen against a strong human. It's when I play winning positions, or just positions where I have a sizable advantage and would be sure to convert it against a human that I get amazed. The computer's ability to defend is simply astonishing. It can drag out what would be a done deal against a human for 50-60 moves and even set some traps that could turn the tables. One inaccuracy and that 50-60 moves could turn into a hundred.

The other thing I find impressive is that the increases in playing strength have not yet plateaued. I thought that for sure by now we would be looking at extremely high draw ratios between top engines. While the number of draws has increased over the years, it hasn't increased nearly as much as I thought it would. I was sure that it was getting to the point where so many positions were already played perfectly that it would take a horrendous amount of testing to get meaningful results. Not so! Every time I think it's about to plateau there's a new engine or a new version that somehow finds 50 Elo. Chess truly is vast!
Parent - By Kappatoo (*****) [us] Date 2017-09-23 01:52
My personal impression is that we at least gained the following surprising insights:
- Re-evaluations of endgames due to tablebases. For instance, KQ + h- and g-pawn vs. KQ is usually a draw. I think many people, including very strong players, found this very surprising.
- I think I remember posts by H.G. Muller and others about piece values, e.g. the queen being much stronger than usually assumed, and generally all pieces (in the middlegame) being stronger than assumed (compared to pawns, that is).
- My impression is that modern engines are very unmaterialistic and often sacrifice material.
- I think someone who had said, say, in 1990, that it is possible to have a chess-playing entity that scores around 99% against the world's best, would have been deemed insane.
- At least I wouldn't have expected that there are positions that are forced mates in 549 moves. I'm not sure how one should extrapolate from 7-man tablebases, but it seems like there might easily be winning positions that take thousands of moves to convert. (Crazy, no?)

What do you think?

On your point about engines being strong at defending: Personally, I am also quite impressed by their attacking abilities when I play them. :)
Seriously speaking, this might be related to my previous point about the existence of long mates. It might be that many winning positions take much longer to convert given perfect defense than we thought. (Of course, humans are also not really perfect at converting ...)
Parent - By Sesse (****) [se] Date 2017-09-25 10:48

> I think that computers have shown is that there is a lot more to defense than we ever realized.


And in a similar vein, that the draw margin in chess really is wide.
Parent - By leavenfish (***) [us] Date 2017-09-23 16:00
What did we gain?

People whose natural ego now lets them think they are good at chess. Without an engine churning away, 90% are duffers.

There, I said it.
Up Topic The Rybka Lounge / Computer Chess / What did we learn from computer chess?

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill