Author: Steve Lim
Date: 18:39:29 10/18/02
Comments in [] are editorial comments. CrazyBird: shall we start now? Darooha: YEs, let's do it ..... introduciing CB hsu, system designer extrotranier Question from Diepeveen: did DB in 1997 do 'no progress' pruning in the software depth too, so with hashtables turned on? CrazyBird: no progress is in hardware, which means it happens everywhere. [“No progress” can be considered a generalized repetition detection mechanism. See “Deep Blue”, in Artificial Intelligence, January 2002.] Question for CrazyBird: There have been some discussion going on, who is stronger, Deep Fritz or Deep Blue, what's your opinion on this? CrazyBird: my opinion should be obvious:). CrazyBird: they are not in the same class. [General Introduction] CrazyBird: I started working on computer chess in 1985. CrazyBird: in roughly 6 months, i created a single chip move generator. CrazyBird: this chip was the basis for chiptest, dt i & dt ii for the next 10 years. CrazyBird: deep blue are based on new chips. CrazyBird: the 1996 & 1997 matches used different chips. [That is, a new 1996 chip was used in the first Deep Blue, and an even newer 1997 chip powered the second Deep Blue] Question from theinz: is there any chance that Deep Blue will return? CrazyBird: deep blue, as in IBM deep blue, is being donated to Smithsonian this month. [IBM is donating one frame of the two-frame Deep Blue to Smithsonian Museum, while keeping the other frame for archival purpose. Some of the chess cards may be donated to Computer History Museum later.] CrazyBird: theoretically, i can create a new version, but only if there is a strong incentive to do so. CrazyBird: i don't think there will be one though. CrazyBird: i am working on shogi on the side these days. Question from Ram: can you ask him to explain the technical differences that make deep blue better or was it just speed CrazyBird: the 1996 & 1997 version of Deep Blue are different mainly in the amount of chess knowledge. CrazyBird: [between the two matches] we went to Benjamin's excellent chess school.:) CrazyBird: the hardware evaluation was completely revamped. [as a result] [Descriptions of Deep Blue’s evaluation hardware can be found in IEEE Micro, March/April 1999 issue. Speed is certainly a big factor as well, and so is the use of extensive selective search extensions] Next question: Seems like people wants to know what the exact meaning of ""12(6)" in the Deep Blue log files, can you explain this? CrazyBird: 12(6) means 12 plies of brute force (not counting the search extensions & quiescence). CrazyBird: 6 means the maximum hardware search depth allowed. CrazyBird: this means that the PV could be up to 6 plies deeper [than what is displayed by software] before quiescence. [The 6 plies is the maximum. Deep Blue had a dynamic horizon between software and hardware search. This horizon could be as close as 3 plies from the true quiescence horizon or as far as 6 plies from it in this case. Tactical positions tend to have the software horizon pushed closer to the true horizon. During the two matches, the maximum PV length in middle game (“quiet” or tactical) shown by the software was of the order of 40 plies, which means the true PV length may be up to 46 plies + hardware search extensions + quiescence search (which may include up to 8 checks).] Question from BenjaminBlue: with the success of deep blue, why wasnt it released as comercial software to run on a pc and do you think fritz would have played better powered by an ibm RS6000 CrazyBird: deep blue is hardware, not software alone. CrazyBird: the software cannot exist without the hardware. it is tailored for the hardware. [The evaluation computation done on each Deep Blue chip (Deep Blue used 480 of these chips) would require something like 100,000 to a million instructions for each chess position evaluated. If we assume that chess knowledge is important for top level play, then Fritz, assuming you can add the same chess knowledge to it, would require a multi-teraops general purpose supercomputer to match Deep Blue.] Next question from gremista: does mr hsu think it is possible to compare deep blue's strenght to deep fritz's CrazyBird: it is. Deep Blue is way stronger:). Next question from Schurick: What does he think about today's chess programs and computers. And what is to his mind the way on which they will be improved. thanks CrazyBird: they are good enough for us mortals:). it is not quite clear how to improve them. speed is still important in computer vs. computer. [This means, among pure software programs, a “smarter” but much slower program would have a hard time to come on top of the heap.] CrazyBird: but we have some evidence that knowledge can make a huge difference. CrazyBird: it is just not easy for software program to add chess knowledge. Next question from fishbait: I heard that you guys ran Deep Blue on the opening books and found some Theoretical Novelties. Have those / will those be published, and if not, why not? CrazyBird: i guess by now, they are published indirectly. The grandmasters on the project get to use them:). CrazyBird: i don't work for ibm any more. [and would not have access to them] [Also, short of publishing Deep Blue’s entire opening book, there might not be a way to publish them.] Question from CallTheFBI: is with today's technology couldn't IBM create another Deep Blue that was more powerful at a smaller cost? CrazyBird: sure. deep blue was done in 0.6 micron cmos. you can get 0.09 micron today... [Assuming that you can ignore the fact that IBM no longer has the full Deep Blue team.] CrazyBird: it is even possible for a pocket pc to be as powerful as deep blue. [0.09 micron gives you something like 40-fold increase in area density and 6-fold increase in speed, which means through technology alone, a single chip can be more powerful than the entire Deep Blue. Put that on a CompactFlash card and plug it into a PocketPC…] Next question from reinforce: how do you view the developement of chess knowledge and search techniques after 1997 and the future of these aspects...I am interesting if you think the same strategy will be CrazyBird: i don't think there is any major development within the last 5 years, other than intel's processor speed increase & the use of multiprocessors in commercial programs (crafty's contribution) [There are a few notable developments though. Ferret seems to have shown that null move pruning and singular extensions may be compatible, to some extent. The development of tablebases is another one. Although, Deep Blue did have 6-piece ending databases obtained from Lewis Stiller on special commission.] Next question from EeEk & fishbait: Are Deep Blue's training games published somewhere? How many games did Deep Blue play vs GMs that are available to look at? EeEk also heard that Deep Blue DID play a game vs Deep Fritz, do you know anything about this? CrazyBird: the training games are ibm properties, but a few games are published, I believe. Chris Chabrid [Chabris] may have them published somewhere. they are against deep blue jr. though. no, there was never any game between deep blue and fritz, despite claims to the contrary. we considered fritz too weak to be interesting. [in the sense of “positionally” weak] CrazyBird: should be chabris. CrazyBird: [the assertion that Deep Blue played Fritz in 1995 is a] false advertisement. deep blue does not exist until 1996. CrazyBird: the new chip was not completed until january 1996... CrazyBird: they [commercial vendors] relabeled the machine [Deep Thought II] they played to take advantage of the publicity. Question from arconia: do you agree that by developing this standard of chess play that the analytical side of chess will now be more prevalent than the strategical side and that this will have an effect on otb chess and the popularity of it? CrazyBird: i am afraid that it is beyond my expertise to answer that question:). Question from mpjaya: - How many programmers worked on the project Deep Blue in total? CrazyBird: it was a three-person project, Murray, Joe and I. Next question from theinz: Is it possible that the games were published in the American Chess Journal? CrazyBird: you mean the training games? i think it was in some sort of game magazine. CrazyBird: they are not really training games proper, but the test games against grandmasters for benchmarking purpose. [There was also one game between two versions of Deep Blue that was broadcast in an episode of ABC Nightline. Kasparov apparently missed it completely. There was a short story related to this in the book “Behind Deep Blue”.] Question from EeEK: You said earlier that you believe Deep Blue is stronger than Deep Fritz, what do you have to say to Kramnik's claim that Deep Fritz is stronger? CrazyBird: money talks. CrazyBird: actually that is not completely fair. sometimes people get blindsided. [Further clarification. People tend to trust their computers too much in tactical problems, and this could give them a false sense of machine’s capability. Some of the GMs the Deep Blue team worked with in pre-Deep Blue days had this problem.] Question from TomBrooklyn: question is Was Deep Blue a unique type of computer or was it comparable to any standard IBM computer at the time with special software? If it was comparable, which regular IBM product was Deep Blue similar to? CrazyBird: it is not really a standard product, although the base machine was (RS/6000 SP). CrazyBird: in terms of real computation speed, deep blue was comparable to something like a 10 teraops machine. [as a result of the 480 special purpose chess chips] CrazyBird: ibm does not count that way when they advertise though. [When IBM makes comparison between a new machine and Deep Blue, they are really comparing it to Deep Blue’s RS/6000 SP host.] Next question from theinz: Can you give us the last names for "Murray and Joe"? And do you think you could create a chess program that could crush Deep Fritz? CrazyBird: murray campbell and joe hoane. yes to the second question, in simul even. [but not with software alone…] Question: As a young chessstudent I read that Bent Larsen claimed to have calculated that the possible possitions in chess is about 10 in 120... When your team was visiting denmark in 1993 ? I read that the Deep Thought project had calculated it to be about 10 in 120 too... Is this asessment still right ? CrazyBird: i don't think we did any calculation of that kind. murray quoted something from literature, i believe. Next question from pille: Which strength would DeepBlue have actually after updating with new chips - which ELO value? Thanks CrazyBird: we don't know. best guess would be kasaprov +- 100:). [The question is somewhat ambiguous. If we are talking about state-of-the-art chips today, then no one knows.] Question from schroeder: YOu were talking about incentive to bring deep blue back ... how MUCH on an incentive are you thinking about ? CrazyBird: maybe more like emotional incentive. when i quit ibm, i wanted to do it, but kasparov was not cooperating:). not as interested these days. CrazyBird: it also depends on how easy it is. if i have the shogi machine completed, it would not be so hard. Question from SJLIM: Is it possible to get any proof of games between DB/DB2/DBjr and the commercial programs? Just to finally dispell any doubt. Are game logs available from you? CrazyBird: i no longer have access to the games or the machine, for that matter. it is up to ibm. it is not that important to them, anyway. [It is possible that Murray is not against releasing them, but it is unclear whether he can just release them or not.] Question from EeEk and others: Can you tell us something about your book? Why do you want people to buy it? CrazyBird: i started working on it [Behind Deep Blue: Building the Computer That Defeated the World Chess Champion] in 1998. one of the reasons for doing it is to let people know about the real stories. the poeple invloved and so on. I also wanted to let people realize that it was not just a machine, there are real people involved... Question from reinforce: the question is: in its match against kramnik, deep-fritzt has proven to have ample tachtical capability but less strategical one. This would as I understand it also be that case with deep-blue. Do you see any ways for chessprograms to overcome this drawback? CrazyBird: no easy way around. you just have to do the dirty work of adding the knowledge. [In hardware, this is a matter of design time and the designer’s will. Not so clear in software, you would have to deal with the tradeoff of search speed and chess knowledge—the more chess knowledge, the slower your pure software program is.] Next question from guzzzler: What's your comments on Kasparovs claim of faul play? And: Were the conditions fair, for example in the Deep Fritz match, they are not allowed to change anything but the opening book. CrazyBird: i have ample comments on kasparov's claim of foul play in the book. CrazyBird: the conditions are fair, as long as you agree that both sides can have secret weapons as in human matches. [It also makes it a far more interesting match if both teams (yes, we are really talking about teams on both sides) can learn from each other.] Next question from Forsaken as a follow up: After Kramnik wins this match and if they approached you about having a Deep Blue 3 play Kramnik would that be incentive enough? (to return with Deep Blue) CrazyBird: that might depend on my employer [HP], as well as the match condition. the fritz match conditions are only acceptable when you are not interested in winning the match, i.e., you know that you are [not] competitive. [Of course, Deep Fritz, on the Compaq machine, seems to be doing a good job of staying competitive as of game 6.] CrazyBird: that is, if [you really think] you are competitive, you would not accept the conditions. sorry for the slip up. Question from TomBrooklyn: Since the Deep Blue project was completed, have you thought of any ways to further improve the program? If so, how much stronger do you think you could make it? How many man-hours of work would you estimate would be required to do so? Would you need the help of any GMs? CrazyBird: you always need the help of GMs. there are many things we could have done, but elected to skip due to time constraints. how strong could we make it? i don't think there is a real upper limit. [Before the 1997 match, Deep Blue was improving at a very rapid rate. If the rate of improvement could be maintained (a big if) for say, another 6 months, it might be able to beat Kasparov by more than one game.] Question from many: Back to your book. If I wanted to buy the book, where do I go? CrazyBird: currently, amazon seems to give a decent discount (30% off). half.com might be lower though, but i have not been checking up on that. Question from SJLIM: Hsu, I'm wondering if you would be open to having another discussion to answer more technical questions from chess programmers. I ask this because I believe both this session and your book are geared more to the general chess public as opposed to computer chess groupies. =) CrazyBird: hm, i guess that is possible. check with darooha:). Question from MrSerious: Are the other two programmers, Murrey and Joe involved in productions of chess computer play? CrazyBird: i don't quite follow that question. we all were involved in the programming, although i am more hardware oriented. CrazyBird: oh, you mean whether they are still active? CrazyBird: joe has left for a DSP startup, and murray is doing data mining these days, i believe. CrazyBird: and ibm is giving deep blue to smithsonian... CrazyBird: don't know whether it would be an active display or not. Question from Celes: I understood that 2 chips were mixed, old ones and new ones with more knowledge. Why didn't you use just the new ones? CrazyBird: sorry for the mixup. the 1997 deep blue only used 1997 chips. CrazyBird: all 480 of them. Question from many: What is the chance of Deep Blue returning after it was donated to the Smithonians? Are you remorsefull that Deep Blue is going to the Smithsonian? CrazyBird: it is an appropriate place for deep blue. deep blue, as it was, probably won't ever return--both Joe and I are no longer with IBM... CrazyBird: i did get the right for the chip from ibm, so a descendent is possible, just don't bet on it. Question from wohl: Are there restrictions on what the smithsonian may do with it? Can they set it up for visitors to play? CrazyBird: that i don't know. it might depend on the software status. they [there] should be enough hardware left over to create some working system. Question from theinz: Does Mr.Hsu think the the human mind would ever be as strong as the strongest man made computer? CrazyBird: human mind is far more adaptive. "strong" is an unclear terms to use in this context. CrazyBird: and computers as of now are only idiot savant. Question from darkone: For CB: Is your book written more for technical people or the layman? Also, what is the title of the book? CrazyBird: it is for layman. "Behind Deep Blue: Building the Computer That Defeated the World Chess Champion". the book is really supposed to be out in november. so probably all the book reviews won't come out until nov. amazon does have one reader review already though. CrazyBird: it contains a few reference[s] at the end of the book for the more technically inclined. Question from Frantic: According to what was published DB was evaluating 200 million positions per second (vs 2.5 to 5 million for the 8-way Simmons server running Deep Fritz). How fast would be Beep Blue today if the project continued? CrazyBird: if we redo the chip in say, 0.13 micron, and with a improved architecture, it should be possible to do one billion nodes/sec on a single chip. CrazyBird: so a trillion nodes/sec machine is actually possible today. CrazyBird: i was planning to shock kasparov should he [have] agree[d] to a new match:). Question from Naisortep: Do you think todays top programs, such as the latest version of fritz, would be stronger than Deeper Blue if running on Deeper Blue's hardware ? CrazyBird: no, they would not know how to make use of the hardware:). Question from EeEK: Can you also explain how to translate nodes/sec into positions a sec? what exactly does it mean? CrazyBird: it is the same. nodes/sec == positions/sec. just jargon. CrazyBird: some people were asking my current status. CrazyBird: i left ibm in 1999. joined compaq last year, and got merged into hp this year. deep fritz is using a compaq machine, so i am not going to be harsh on them:). Another question that seems to be popular: Do you think computers are able to SOLVE chess ? Could that ever happen? CrazyBird: solving chess? interesting question. CrazyBird: it is not impossible, although unlikely. it is estimated that the total number of different chess positions is something like 10^40. CrazyBird: square root of that is 10^20. this could be a lower bound for the size of the proof tree. CrazyBird: and 10^20 is not an impossible number. CrazyBird: 1 followed by 20 zeros. or 100 billion billion. CrazyBird: that is an optimstic estimate though. Question from ardee: Does "12(6)" mean 12 total ply or 12+6=18 total ply? This has the been source of huge arguments for years! CrazyBird: 12 total in terms of brute force. 6 is just the max partition in hardware. CrazyBird: to all the book readers, if you do like the book, please tell your friends would [who] might be interested. thanks. CrazyBird: replace would with who:). Question from parabola444: You mentioned Deep Blue searched about 12 plies brute force + extensions, which is similar to what pc programs these days get on a fast pc - since Deep Blue hardware was much faster, how come it didn't search significantly deeper ? CrazyBird: we were using fairly extensive search extensions [which exchange brute force search depth for selective search depth], and the decision not to use null move pruning was an [a] deliberate one. [That is, Deep Blue is actually significantly deeper in critical lines—its PV is frequently around 40+ plies, and the non-PVs sometimes get even deeper.] CrazyBird: there were several reasons. CrazyBird: first, we were always at the top of the heap, and the occasional error introduced by null move could cause us to lose games to lesser programs. [politically embarrassing…] CrazyBird: second, we observed that we were zugzwanging null move using opponents, which made us suspicious of it. [possibly a result of the deeper selective search on Deep Thought & Deep Thought II, the players of most of these games] CrazyBird: third, it is not clear how to incorporate singular extensions with null move pruning. they did not seem to be that compatible. though Ferret seems to suggest that it is possible. anyway, given that singular extensions are considered far more important in creating deep lines, we keep what we know. CrazyBird: fourth, and not least, speed was more than adequate, and we did not need to resort to null move. Question from oddg: Back to the WC 1995, There was an entry with the name Deep Blue (Fritz won against DB), did it not have any relations to your Deep Blue? (EeEk: any idea how Deep Blue's name got in there, is this completely false?) CrazyBird: deep blue did not play in 1995, since it did not exist yet. it was just deliberate relabeling on the part of commercial vendors, for obvious reasons. CrazyBird: the program played was deep thought ii. which was vastly inferior to deep blue in chess knowledge as well search speed (1000 to 1 ratio in effective speed, 100 to 1 in raw speed) CrazyBird: and we were as unlucky as kramnik is today:). Question from TomBrooklyn: Follow up to 12(6) ply: What does max partition in hardware mean? Not answered [See earlier notes] Question from Diepeveen: tell crazybird sorry for the off topic question, but where can i order your book. i need ISDN number i fear if i want to order it here. creditcards not common in europe. CrazyBird: i will answer diep's question first. you can find the isbn number in the news item for this lecture. CrazyBird: or you can find it on the amazon web site. Question from winechess: How far do you see computer chess and A I coming in the next 10 yrs and how strong do you think they will be as far a elo? CrazyBird: now back to the normal question. it is hard to say. speed will increase for sure, intel is going to make sure of that. CrazyBird: but the knowledge part may be more difficult. some of the things we did in deep blue are very expensive to do in software, and i am not sure the commercial programmer would be willing to go for them. CrazyBird: that [the elo question] might be an impossible question to answer. it depends on the will of the people involved. There have also been a couple of questions regarding the rematch. Why didn't IBM want to offer Kasparov a rematch? CrazyBird: inside ibm, what i heard was that lou, ibm ceo, was all for giving kasparov a rematch, purely for sportsmanship reason. CrazyBird: but with the incessant attacks on ibm, even the best intentioned people had their limits... CrazyBird: it [the attack] played a very big part of it. ibm actually had an advertisement/endorsement deal with kasparov--of course, it did not go anywhere. Next question from many: Do you follow the chess world or play yourself? or are you only into computer chess? CrazyBird: the play part. no, i don't play chess--i only played the world chess champion on tv:). CrazyBird: well, i do play a few games. I am no better than 1800. i stopped following chess or computer chess for a while. but people constantly remind me of these things:). Question from SJLIM: Just for the record. Could you explain to the chess public the history about a certain Chess card that you wanted to build once upon a time? CrazyBird: you mean the kasparov butt kicking device? CrazyBird: just joking. since kasparov was not interested in playing, i gave up. CrazyBird: you could build something similar with FPGA's today though. CrazyBird: that is how i am going to build a shogi machine in the first step. Another personal question, from DiMarco: Mr. Hsu I read in 1991 you had a Ph.D. where did you do your B.S. or where did you grow up? How did you become interested in computer chess? CrazyBird: In the book, there is an appendix that gives details about my life before the project. CrazyBird: i was born in taiwan and attended national taiwan university there. CrazyBird: i was trained as an EE, hence the hardware interest. CrazyBird: my interest in computer chess was probably triggered by media reports on computer chess. CrazyBird: finding the book "chess skill in man and machine" while in college was probably the final straw. Question from bootneck: why are you called crazybird? CrazyBird: Crazybird was my nickname at high school. [Crazy also has the same sound as Feng in Chinese.] Question from Aldos: Is there such a think as computer chess style, or all is about machine strength and optimized algorithms? CrazyBird: Are there different styles for different computers? was that the question? CrazyBird: hm, styles can come from different things. search algorithm can make a big difference. CrazyBird: chess knowledge can do the same thing as well. before the 1997 match, joel benjamin made an interesting comment. CrazyBird: he said, "you know, sometimes deep blue plays chess." [In this case, the new style came from all the new knowledge added between two matches.] CrazyBird: a few[more] questions. my wife would kill me if i stay here much longer:). nice talking to you all though. CrazyBird: sorry, a few more questions. Question from Frantic: There have been media reports that Deep Fritz is running on a Simmons 8-way server . You mentioned Compaq - did you mean in the match vs Kramnik? Do you know any details about that box? CrazyBird: the simmons box? there is a photo that showed compaq logo on chessbase web site... Question from Karol: After what happened in game 2 vs. Kasparov, why did IBM never deliver the printouts of the game as promised? CrazyBird: they were delivered. [Kasparov received all the relevant log he asked for right after the match. Also, prior to the completion of the match, printed versions of the entire logs were given to the match arbitor as well as the independent observer, Ken Thompson, who also got to see the game log live while the games were ongoing.] CrazyBird: kasparov has the relevant log, and of course, everything is on the web today. [The IBM match web site, www.chess.ibm.com. The entire logs have been there for several years now.] CrazyBird: the entire log, that is. Question from TomBrooklyn: The (revised) Question Is: Have you read the book "A NEW ERA, How Garry Kasparov Changed The World Of Chess" which is about Kasparov-Deep Blue 1996 and 1997 (and has a couple of pictures of you in it.) What is your impression of that book? CrazyBird: i did not read that, and don't intend to. Ok, one last question: Question from EeEk: Do you think computers will dominate the chess world, and if yes, when? Do you think chess has a future then? CrazyBird: chess will always have a future as long as people find it fun to play. will the computer dominate? nay, we will just ban them:). CrazyBird: thanks you all for coming. once again, if you like the book, please do spread the words around, thanks.
This page took 0.01 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.