I recently visited the Computer History Museum in nearby San Jose, which has a new exhibit on the history of computer chess. Despite my total lack of interest in chess as a game, computer chess has a special significance in the field of computer science. Chess remains the most visible and public benchmark of the relentless increase in computer speed over the last 50 years.
This is a companion discussion topic for the original blog entry at: http://www.codinghorror.com/blog/2006/10/chess-computer-v-human.html
There was a study years ago, based on Bobby Fischer IIRC, that good chess players (and better) use neither Type A or Type B paradigms. They simply have memorized more successful games than their competition. The study found, in particular, that the good players don’t “think more moves ahead” than the opponent.
This might be thought of as Type B, but it really isn’t. Whether a prolog program could be loaded with the last 100 years of successful games and smoke the grandmasters, I leave as an exercise to the reader.
Cool post! Since the problem is parallelizable, it would be interesting to see a chess applicaation built on top of Google’s computer array.
but the best 200 chess players in the world are still holding their ground.
Not for long. I will soon unveil “Deep Haack!” and checkmate them all!
Joking aside, as buggyfunbunny points out, chess is a pattern recognition problem and we know that humans excel at recognizing patterns of other games they’ve watched, played, or read about.
However, considering that a computer could “remember” every significant game played in history, all it takes is a little bit better pattern matching, and I don’t see how humans can keep up.
At some point, it becomes a question of whether or not chess is “solvable”. It may well be that keeping up means simply not losing. Kind of like tic-tac-toe, a “solved” game. No matter how good you are at it, if you play someone else reasonably good, you’ll never beat them.
“As it turns out, computers have a hard time with the concept of ‘good’.”
Yo. Quotes page.
While I no longer play chess I do find that benchmark one of the more usable ones around.
Do kind of wonder what my old C-64, from when I did play computer chess, would of rated.
Also with the newer home chess programs do they respond with thier moves quickly or have developers taken the extra CPU speed and used it to make far more time intesive calculations?
For parallel use of home PCs etc it’s worth mentioning ChessBrain http://www.chessbrain.net/ , "a virtual chess supercomputer using the processing power of Internet connected machines.
On January 30th 2004 ChessBrain made history by becoming the first distributed network to play a game against a single human opponent."
I think your timeline of Deep Blue development is wrong for 1988. If I’m following the linked page correctly, the value should be 720k not 20k.
A very interesting article, though. Perhaps an idea for a future article is - I wouldn’t have gone to check the Deep Blue Timeline link if your table didn’t look so glaringly wrong. Sometimes it takes a big error to get us lazy guys to go check the facts!
Onset Computer Corp.
I have been intrigued by the advances of computer AI ever since I was a boy and witnessed the near destruction of planet earth by Dr. Falken’s “Joshua” in WarGames. I have watched a few televised computer vs. man events over the years (you want to talk about some riveting television…NOT!)
However, I have always thought and continue to believe that artificial intelligence it going in the wrong direction. The trend seems to be along the lines of greater speed and greater capacity. That is not the solution. For some time now computers have been able to process data (albeit by 1s and 0s) much faster than humans. The storage capacity comparison is open to argument but humans can’t match a computer’s ability to accurately hold tons of data and retrieve said data line for line.
The problem is the human brain does not work like a computer. The real answer is to replicate the exact way the human brain works with some modifications, etc. Until we pioneer new hardware and a new approach to AI I don’t see many advances in the field.
Maybe being in teams made the grandmasters less creative/daring - we’ve all seen how committees suck the cleverness out of projects.
Jeff, is the Blogger ban a new thing? You’re moving up my self-hosting timetable.
I think the most efficient chess AI programming award goes to Atari for making Video Chess for the Atari 2600 with complete, working AI in 4 Kilobytes.
Did you see the movie about that Deep Blue match?
It was amazing!
I think something really went wrong there. Someone from IBM should have stepped to the podium and put this in perpective, instead of beating on Kasparov.
As the quote article says “Kasparov did not evaluate 200 milion positions per seconds”, so that means that even if it looses one, he’s still better. Deep Blue had no intelligence, only persistance and practically infinit time.
Ken - Hardware is trying to move in this direction. It’s just very hard.
but the best 200 chess players in the world are still holding their ground
That is questionable. It is true that the world champion vs machine matches (some years back) resulted in draws (Kramnik and Kasparov), recently machines consistently mop the floor with top grandmasters. Remember Adams - Hydra last year? The result was 0.5 - 5.5 for Hydra… And Adams was around 7-8. best player in the world at that time. There was team matches also between a group of machines and top grandmasters. The result was the utter defeat. The best human was the recent world champion (Topalov) but he got still a score below 50%. And the opponents were not supercomputers, one of them was Fritz 9 running on a laptop…
The article by Jeff Sonas was written 3 years ago - an eon on computer chess terms. Have you noticed that very few GMs are prepared to play matches against computers nowadays?
The last significant match was played between GM Michael Adams (then rated number 7 in the world) and Hydra, in mid-2005. Hydra won 5.5 - 0.5.
Computers playing chess seems trivial when compared to Go.
I totally disagree that computers will never be able to consistently beat the best humans in the world. There are only a limited number of possibilities in chess. Its only a matter of time before chess is reduced to something as simple as tic-tac-toe for a computer.
Ah, lovely. Merge Zappa, Fruit, and Deep Blitz inside the Google memory space, have them decide on things by committee, and then watch them take over the government.
I, for one, welcome our chess-obsessed electronic overlords…
What everyone misses in this is that programs don’t really play any better than the best humans now - they just don’t make the mistakes that a human will always make because he or she is tired, distracted, or suffering a blind spot. Take back all obvious errors (caused by faulty calculation alone) and I can beat a very good program (and I’m not that good!) I lose against programs (and humans) because of blunders not because I can’t strategize. In the recent World Championship Topalov missed a two move sequence that would have forced mate! I found the correct moves while I was following the game and program found the moves later in less than a 10th of a second. But in that same game the same program rated his position as very bad before he reached the winning position and while his opponent did blunder to get in the mating position he would have still been at a disadvantage if he had found the move that didn’t lose immediately. In other words Topalov saw much more deeply into the position that the program could! But the program would have never missed the simple mate! Maybe brute force will eventually equal hundreds of years of experience - I think it will - but not yet!