Author: Robert Hyatt
Date: 10:26:34 04/30/01
Go up one level in this thread
On April 30, 2001 at 11:26:48, Albert Silver wrote: > >Here's a little story: > >Somewhere around 1988/1989 in Paris, the large department store chain FNAC >organized some kind of chess challenge in order to draw publicity for their >section selling chess playing machines (all stand-alones). The challenge went >like this: 8 players would qualify for a knockout event of 20 minute games, the >top 3 of who would win some prizes. To earn a spot, you had to beat the >Constellation Forte B (rated around 2000 Elo - in France - in 40/2h) in a 5 >minute blitz game! No doubt to guarantee players of at least master strength. >The trick was that you could play against the machine all you wanted, but had to >call over the attendant of the department to watch your qualifying attempt. You >could only try once, and if more than 8 succeeded, the shortest wins would be >awarded the spots. You must understand I was rated 1580 Elo at the time. I spent >the whole afternoon there, noticing how it treated a few openings, and finally >was confident of my ability to score a win at will. I qualified by mating the >machine on the 20th move in its ill-fated Najdorf (0-0-0, g4 and h4 and kill the >comp!). Naturally, I also lost to my 2280 Fide rated opponent in the first >round. > >I understand that this was a long time ago, and that programs and hardware have >gone a LONG way, however the fact of it remains that the program's opponent >isn't rated a measly 1580, he is rated a good 1200 points more and is the >current World Champion. He won't be preparing for a single afternoon, but rather >3 full months. No disrespect meant to Amir or Frans, but you can imagine what my >prognostic is. The only positive thing is that it may serve to recover mankind's >pride after the DB-Kasparov fiasco; though I think the general media (and >people) will simply conclude that PCs have a long way to go before reaching DB's >heels. > > Albert I have a similar one. Bert Gower and I (along with several others) started a chess club tournament many years ago. We had an odd number of players, so we elected to include Bert's Super Constellation to round out the event and avoid any byes. I had played this thing many times and could simply crush it at will, because it was very aggressive but it didn't know anything about endgames, nor about kingside attacks, and so forth. In fact, I had noticed one particularly bad point where it would sacrifice a piece for what it thought was a good attack, but it could never win against me with that line. I decided to try it in my game vs the SuperCon, and by move 20 I was a piece up, and by move 35 the game was over. Bert was paired against it a couple of rounds later and he followed my game exactly and won easily. The best player in our club was rated just over 2200 and he played it and lost, not knowing about my "busted line". When we showed him, he was pissed as I won the tournament and Bert finished second. So long as a machine will fall for stuff like that, humanity doesn't just have hope. It has good chances. And before you say this was a book learning issue, it wasn't. It was just a detectable evaluation problem that you could exploit from many similar positions once you knew the 'pattern'. Today's programs try to combat this with book learning, but it isn't enough. I watch Crafty/Scrappy play GM players all the time and repeat the same openings because they find a very narrow and forced line that book-learning can't get around. You can do a search for JRLOK on ICC and see that he has success against crafty, tiger, deep fritz, you-name-it. No, he doesn't often win most games. But he wins more than you would suspect when you consider that at blitz the computers have a serious advantage. And these are blitz games.
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.