Open Letter from Feng-hsiung Hsu, one of the main programmers of Deep Blue | Hacker News Reader

Open Letter from Feng-hsiung Hsu, one of the main programmers of Deep Blue

chesscenter.com

42 pointsamichail19 comments

Reply here: http://www.chesscenter.com/twic/owenfeng.html

Thank you for the reply - things make sense after reading it. I know a few people like Hsu - they usually are convinced that they are objective and are absolutely unable to grasp the fact that other people might have different values and goals. They are conviniently overlooking things they don't like. These people often do achieve much but are absolutely dreadful to deal with.

I don't know if I agree. Owen's letter seemed to pretend that Kasparov didn't lose the last match. Why would Hsu need to establish himself as a competitor when he won the last match? That doesn't make any sense to me. When he says things like, "All he had was a computer chip with no organization or sponsor behind him." it's ridiculous. That chip is the chip that contains the data and knowledge that beat Gary last time, and he can't get any sponsors because they won't commit to a match. Looking at both letters, it seems that all Hsu wanted to do was get another match and Kasparov didn't really want to do it. I certainly could be wrong, but it sounded like they were making excuses.

Fen-hsiung Hsu's letter (the original post) has a ring of truth to me: it sounds like a sincere story of someone who doesn't know how to continue with a project that's personally important without the corporate backing that he started with.

This reply is pure mockery -- it could have been written by anyone, there's no personal feel at all.

How much faith should we put in our truth-identifying heuristics?

I agree, Owen's arrogance is quite startling to me...

What I don't understand is how Hsu wanted to challenge Kasparov for "title match"... World Title matches are managed by chess federation (FIDE), so it's not up to Kasparov or Hsu to decide. All in all, I would take both letters with a big grain of salt.

Deep Blue '97, which beat Kasparov, was a: "RS/6000 SP Thin P2SC-based system with 30-nodes, with each node containing a 120 MHz P2SC microprocessor for a total of 30, enhanced with 480 special purpose VLSI chess chips." (from Wikipedia)

I'd be interested to know what exactly those VLSI chips contained ... My hunch is that a few dozen modern FPGAs should be able to match a few hundred ASICs from over a decade ago. Then it shouldn't cost more than ~$10K for enough hardware.

Hsu could easily bootstrap that sort of machine. Then, after winning a World Computer Chess Championship, he would be in a much better position to issue a challenge to Kasporov.

It depends. Having designed medium sized VLSI ASICs over a decade ago and worked with FPGAs recently, I'd have to say that it may still take several FPGAs to equal one ASIC.

I believe there is a consensus in the chess world that given any incentive, it would be easy for someone to put together a machine that can beat any human alive.

The algorithms at this point are well-known and tested; it's just a matter of having enough hardware to search deeply enough to match a world-champion player within the tournament time constraints.

edit: http://www.chessgames.com/perl/chess.pl?tid=54499

I believe this is correct. I worked at T J Watson during this famous match and if I recall the consensus in the halls was that these CMU grads had some interesting hardware hacks but the program was essentially the classic heuristic search with evaluation function. I think computers also have been used to work out some end game solutions

I do not think this is the case. If you follow chess and the matches where a top GM plays a computer you will find most of the time it is a draw. There was an article on chessbase many years ago that showed that chess program rating had stopped growing as fast and they were leveling out just as player ratings level out. http://www.chessbase.com/newsdetail.asp?newsid=1292

Perhaps we are coming closer to optimal play?

This is definitely the case. Check out Rybka - it has been rated 3200+. No human being comes close.

Time away breeds ignorance...thanks for the reference

After further reflection I'm not sure my belief is wrong.

The match between GM Roman Dzindzichashvili and Rybka has finished with a somewhat surprising (at least to me) score of 4-4.

http://www.chessville.com/Wall/DzindziRybka.htm

From the article:

"the odds were changed to - Roman gets White every game, the computer offers up a different pawn each game."

So the computer gave Dzindzi an enormous handicap.

I failed to make my point...sorry. A human rated 3200 would win more than 4 games even given the restrictions against a GM rated ~2550. Ratings are not as as transferrable as one would think, especially when a computer is involved.

I agree with most of this, but one should not overestimate the impact of general hardware progress since 1997. Each extra ply takes an order of magnitude more computation, and the state of the art in endgame tablebases has "only" progressed from 5 pieces (e.g. KBB vs. KN) to six pieces.

Pardon my ignorance, but why is it that Deep Blue can't be recreated on a reasonably fast modern PC in software? In fact, why is it that this sort of program would be considered difficult at all?

EDIT: answered my own question via wikipedia

"In a recent match, Deep Fritz vs. Vladimir Kramnik in November 2006, the program ran on a personal computer containing two Intel Core 2 Duo CPUs, capable of evaluating only 8 million positions per second, but searching to an average depth of 17 to 18 plies in the middlegame."

Here's a link to the games from the 1997 match. The game logs (showing Deep Blue's actual output) are pretty fascinating.

http://www.research.ibm.com/deepblue/watch/html/c.html

Reply on news.ycombinator.com