Author: Robert Hyatt
Date: 12:40:28 05/25/99
Go up one level in this thread
On May 25, 1999 at 13:18:49, Paul Richards wrote: >On May 25, 1999 at 09:59:29, Robert Hyatt wrote: > >>It is simply time to realize that the GM players know so much more than the >>programs, that the only way the programs are getting by is through tactical >>skill. And at 40/2 the tactical skill of the program is not obviously better >>than the tactical skill of the GM. At game/30 it is, of course. But at 40/2 >>I don't believe so. > >I would still distinguish between tactical skill and strategic >understanding. Strategic understanding lets you create favorable >situations that are likely to offer tactical possibilities. This is >the way humans approach the game, and it allows them to create >situations where the payoff is too distant for the computer to see. >However I would not say that human tactical skill is equivalent to >the computer's, regardless of the time control. Even Kasparov uses >programs like Fritz to investigate positions, precisely because >computers are superior at tactics. GMs think in terms of plans, and >then calculate variations for their plan, and for what they believe >their opponent's plan is. They are just as incapable as the rest of >us chess players of analyzing all possible moves, and as a result >they will overlook things. GM Rohde had a strategically won position >but did not play optimally winning moves to finish the game either. >If he could calculate as precisely as Rebel he would have. Again, >GM calculation is devoted to carrying out specific plans, and this >strategic understanding allows them to efficiently use a tactical ability >that is very limited compared to the machine's. But since winning tactics >invariably spring from good strategic positions (centuries of human >experience have distilled the elements of what constitutes a good >position), barring a tactical blunder the human has the edge. The >computer is not outcalculated in any capacity, but does not have the >knowledge base built on centuries of practice that guides GM play, and as >a result is strategically squeezed into a position where winning tactics >are unlikely to exist in the first place. This highlights the difficulty >of finding algorithms that accurately describe strategic play. A program >that can do this well will win on ordinary hardware. Given the difficulty >of programming such knowledge though, brute tactical strength will make up >for the deficit. In any case when it comes to pure tactics the humans have >long since lost that battle. One simple example... Take the Shirov game last year, where he played Bh3. That was _all_ tactics, as he claimed that this was won at that point no matter what. And find _any_ program that can tactically see Bh3 winning. In some wild positions the computer is better, but in very forcing lines like that, the human is still formidable. It is solvable with the right kind of search (ie I wouldn't be surprised if DB finds Bh3 quickly since it uses that same sort of forcing extension logic) but the general program of today has little chance. Given hours. And yet Shirov found it in minutes. There are other such examples of very deep tactically forcing lines that the computers simply don't handle... but the human does. And there are other examples of positions where the human would be at the mercy of a computer when the board is wide open, pieces are developed and attacking everywhere...
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.