< Earlier Kibitzing · PAGE 2 OF 2 ·
|Jul-22-05|| ||Brown: <Computers have been around for a while, and played numerous matches against humans, losing many in the beginning, but they were always better at sportsmanship. Never heard of a comp making stupid excuses to cover for its losses, or going into raptures over its victories, two things that some humans tend to do an awful lot. I don't care if computers are better at chess, but they seem to be better also at some virtues that are supposed to be inherently human--and that's what's humiliating to us.>|
Computer's developed super-egos when I wasn't looking.
|Jul-22-05|| ||Caissanist: Call me naive, but I honestly think that Nemeth would score better in a match than Adams did and, depending on what the ground rules were, might well win.
The reason is that he is approaching this not as a chess match or even a reverse engineering job, but as a software quality assurance (QA) problem--i.e., he's looking for bugs. I've been in the softare biz for twenty years now, and I've seen two constants--at every company the QA people always know more than anyone else about the product and how it works, and they never get the time or resources to do their job properly. Except for the very simplest ones, there's not a software program out there that doesn't have significant bugs and design flaws that a savvy QA engineer (or hacker) could find. I would love to see Nemeth, or someone like him, make Hydra look silly. I'd like to see this not so much because of anything to do with chess, but because it would be a fine publicity stunt to highlight the sloppiness and bugginess of virtually all commercial software. Maybe this would be the nudge that would force the software industry to implement the kind of quality standards that are taken for granted with most other products. |
OK, yeah, that IS naive. I have more chance of convincing Christina Aguilera to marry me and have my babies than I (or Nemeth) have of convincing the software industry to build reliable products. Still, it would be fun to see a hyped-up media event that highlighted just how shoddy this stuff can be. Which, I believe, is where I came in :-).
|Jul-23-05|| ||korger: <Caissanist: there's not a software program out there that doesn't have significant bugs and design flaws that a savvy QA engineer (or hacker) could find.>|
That's true, but for a reasonably well-written program all of the design/programming faults manifest themselves through the improper use of the product. For instance, in the UNIX world the most common examples are the infamous buffer overflow exploits. (I deliberately avoid talking about Windows-related problems, as that is a huge design flaw on its own, rather than a software.) For someone who is communicating with a program through a chess interface, able to make only legal moves, the possibilities for exploits you are talking about are practically eliminated.
The only thing Nemeth--or any chessplayer for that matter--can do is to find shortcomings of the algorithm--and those aren't the same as errors in the code which hackers would exploit. Turning the shortcomings to your advantage may enable you to beat programs which are nominally 2-300 ELO points higher than you, but the enormous 700+ points gap between Hydra and Nemeth is just too big. No way.
<it would be a fine publicity stunt to highlight the sloppiness and bugginess of virtually all commercial software.>
Then your best hope is that during the next Hydra-Human match a hacker cracks the mainframe via Internet, and makes the computer play imbecile moves, or just resign in a won position. I agree, that would be a good demonstration about buggy software--but this has nothing to do with the playing strength of Hydra, so long as everyone plays according to the rules.
<I have more chance of convincing Christina Aguilera to marry me and have my babies>
There, you've taken the words out of my mouth. Not only you have better chances of a beautiful marriage, but that will give you undoubtedly more happiness than seeing the IT industry collapse and writhe in pain if its faults should be revealed. So even if the latter never happens--don't worry, this life offers you plenty of much nicer things! :)
|Jul-23-05|| ||moocow: <Turning the shortcomings to your advantage may enable you to beat programs which are nominally 2-300 ELO points higher than you, but the enormous 700+ points gap between Hydra and Nemeth is just too big. No way.
Fritz, Shredder etc are only 300 ELO better than Nemeth? I think not. Obviously Nemeth was playing around with these programs to come up with the techniques he's found and he would probably need a fair time playing around with Hydra to if he was to discover anything, but I don't think we can totally discount the possibility. Check this out for absurdity http://f27.parsimony.net/forum67838...
What I think is disappointing is that GMs who have obviously much more knowledge than Nemeth have not been able to come up with an equivalent of his chess lateral thinking, but on a GM level. Let's face it, one of his early ideas (the g5/g4 piece sac to open the h-file) is hardly sophisticated and the concept should be known to any club player with a decent knowledge of the exchange Lopez. Of course, I guess GMs have better things to than find amusing conceptual failures in chess programs ;-)
|Jul-23-05|| ||roni.chessman: IMO Hydra has no flaws. It depends greatly on "calculating power" from its hardware and not so much the software like fritz, shredder, or chessmaster. Therefore, because it is hardware-based, it has no bugs whatsoever. The only thing that will cause it to play like crap is when the hardware malfunctions. |
Nemeth will be ripped apart by Hydra.
|Jul-23-05|| ||moocow: Yeah, of course it has enormous calculating power from its hardware. The point of Nemeth's play, however, is to find perverse ways of setting problems for programs that are sufficiently long-term as to be outwith the level of ply that they are looking and they therefore cannot calculate their way through it (one rather more classical example of this is that programs need tablebases for certain endings to play correctly). Kasparov managed this in game 3 with X3DFritz using, compared to Nemeth, "normal" positional chess. How much more difficult this would be with Hydra is hard to say (obviously it would be more difficult). |
Nemeth (I think) has understood that chess programs are bound 100% by their programming and tries to construct situations where the algorithms that the program is bound by become a disadvantage. Obviously, if a position becomes a mainstream chess position, he will be annihilated. Perhaps an interesting contest would also be if a top GM took on Hydra with Nemeth as a second?
|Jul-24-05|| ||Caissanist: Roni: we are a very long way from having computer that can win a chess game through "brute force calculation". I posted the numbers in another cg forum a few months ago, but what I remember is it's something like 10^89 years to brute force calculate a move on the best existing hardware; by using parallel processing Hydra cuts it down to something like 10^87 years, which reduces but does not eliminate the dependence on software-based "smart" pruning algorithms. |
However, it is possible that this would be enough to make it impossible for a player of Nemeth's strength to exploit the flaws in the algorithms. That is, if the computer makes a mistake by only evaluating positions to a certain depth of moves, then it doesn't matter if that depth is beyond Nemeth's ability to calculate. I still think that Hydra-Nemeth would be more interesting thatn Hydra-Adams was, though.
|Jul-24-05|| ||Caissanist: Korger - I do not want to see the IT industry collapse and writhe in pain. I want to see it dramatically increase the reliability of its products, and get paid lots of money to do so. This is what happened with the whole industry when the Y2K bug popped up, and it's what happened to Microsoft as all the security flaws in Windows were revealed. I've seen this dynamic every day for years--the attention that the industry gives to quality is directly proportional to the outside pressure, and the outside pressure is directly proportional to the hype. |
There was a story earlier this month about a game that Shredder lost in a South American tournament (to IM Pablo LaFuente) as the result of an I/O error reading a hash table. The tournament organizers claimed that it was a "one in a million" error. Well, it shouldn't be--it should be one in a trillion at least, preferably one in a quadrillion. And I would love to see more incidents like this, not because I think it would "save chess" but because it would help publicize such geeky things as the unreliability of non-parity memory.
Anyway, this doesn't have a lot to do with chess, my apologies for the digressions. Regarding your chess related points, you may well be at least partly right, although I'd still like to see a match. Don't think it will ever happen.
Oh, and let me clear up one other point--I'm already married. Sorry Christina ;-).
|Jul-24-05|| ||Caissanist: btw, for those who are interested in the numbers quoted above, more details can be found at http://www.mssm.org/math/vol2/issue.... It's actually 10^92 years rather than 10^87, not that the human mind can make sense of that distinction. I see now that roni was part of the previous debate as well, so I guess that he was unconvinced by my iron logic. C'est la vie.|
|Jul-26-05|| ||Brown: Statistically, one cannot determine/prove whether Hydra has any "flaws" in its understanding of position, values, etc. Not nearly enough games.|
What is clear is that they seem very hard to find, at least with the current opposition.
|Jul-26-05|| ||Sneaky: <Texas No-Limit Hold-Em Poker> I think computers will ultimately master this game through carefully constructed behaviors programmed with the help of experts in the game.|
The element that they are weak on, of course, is bluffing in its many various forms. One of the most important principles of high-level poker is being able to evaluate bluffs by retroactively considering all of the player's behaviors in that single hand up to that point. A good bluff will present a plausible story that "adds up." An amateur will attempt an inplausible bluff, for example, they act elated when they see the third spade hit the table, and move all of their chips in the middle, as if to imply they have completed a flush. However, for this story to really add up, we must imagine what would have kept the player in the pot up to that point--if the bets in previous rounds were prohibitively expensive to allow him to pursue the odd chance of a flush, then it seems unlikely that he was really staying in the pot for that reason. Of course, you might be playing against such a fish that they were calling the previous bets "on a hunch"--fine, then the fish wins a hand, but such behavior will cost them their shirt in the long run.
Over time, this complex set of rules can be represented in code so that the best poker computers can go up against the best humans with at least equal odds if not superiority. But no matter what happens, I'll have a better chance against the poker computer than against Hydra.
|Jul-26-05|| ||Brown: <Sneaky> What tells can the computer reveal? Will it bluff on it's own, in a calculated way? And it will see "tells" from humans by a complex math equation? |
I think you are right that you have a much better chance in beating a poker program than a chess program, especially due to imperfect information and probabilities.
Whatever the case, the two programs (poker and chess) will look very different in code, I think.
|Jul-26-05|| ||OhioChessFan: <Caissanist>, My mind is able to know that the difference between 10^87 and 10^92 is about the same as my chances of getting a date with Christina after she dumps you.|
|Nov-01-06|| ||Granite: <Brown> Computer will use randomizers to vary their play much like human opponents do now. In "Harrington on Hold'em" by Dan Harrington he discusses the idea of simply using the second hand on your watch as a portable randomizer. Even if you opponent knows what you're doing they still will not be able to draw additional information from your play. Programming a good computer opponent will surely be difficult, but like all finite games an optimal solution does exist although it may never be discovered for both chess and Hold'em.|
As for tells, poker is not a card game but a bidding game and several types of betting structures are tactical like pins, forks, skewers and the like for chess. If a computer could understand and exploit those tactics it would be able to run some successful bluffs. In many respects, an opponents behaviour is revealed exclusively through their betting habits and if a computer is able to build a reliable set of "tell" bets on their opponent it's likely they'll be able to exploit that although it would be very hard to program.
Naturally the programs would be very different then those used in chess as chess is a perfect information game (you always know what your opponents moves are) while in poker you have imperfect information, and that has a significant impact on the game's theory.
|Nov-11-06|| ||2021: Perhaps Eduard Nemeth is related to:
|Dec-06-06|| ||Karpova: <Caissanist: Call me naive, but I honestly think that Nemeth would score better in a match than Adams did and, depending on what the ground rules were, might well win.>|
No, he wouldn't stand a chance. Nemeth has to play hundreds and thousands of games to determine a computer's weakness so playing a match versus Hydra without preparation... and what if he can prepare? well, Hydra's developpers would have enough time to check the computer also. So no advantage at all.
It's not like sitting at home in his chair, eating potato crisps and playing around with some old Fritz on an old PC.
<Caissanist: I'd like to see this not so much because of anything to do with chess, but because it would be a fine publicity stunt to highlight the sloppiness and bugginess of virtually all commercial software.>
Even if he did win against Hydra - Hydra is no commercial program and therefore irrelevant.
|Dec-06-06|| ||Caissanist: Well, for what it's worth I now agree with Karpova, which is to say that I disagree with what I wrote a year and a half ago. I guess that it's still possible, barely, that Nemeth could find a cheapo loophole to win a game from (say) Fritz, which of course is more than Kramnik was able to do. But it gets less possible every week. Chess engines are mature products now, which means that any significant bug has very likely been fixed. Nemeth himself doesn't seem to be able to bust any of the programs anymore, or if he is he's not publishing any of his wins.|
|Apr-20-07|| ||lukeprog: 170 games of Eduard Nemeth:
|Oct-15-07|| ||xeroxmachine: Hasnt the man plaid any homons? Does he always plai compotors?|
|Dec-23-07|| ||timhortons: i dont think these computer run on an 8 core machine...what nemeth posted here is his won game but maybe he played a thousand times against these computer and guess how many times he won....i beat roce at icc rated 2200 but gee i played with her a hundred times till i recognized a pattern where she would play like a broken record..i posted that game in my forum...the admin of icc warn me agaist lines repaetaedly used against there computer for scoring purposes....|
|Nov-22-08|| ||Richard Taylor: It's clear to me (from the above etc) that Nemethes is a genius who would indeed beat Hydra or any other computer - this is an amazing proof that Fritz is a moron!! The IT people have been beaten at last...!! |
It is clearly because they lack those deep human qualities of insight and subteley etc
|Jul-22-09|| ||birthtimes: In Capablanca's words, Nemeth, Bronstein, Pecci, and others who have had great success against chess computers, have all utilized his principle of position as being the most important factor in winning a chess game. These players don't care how much material they lose, if they gain a winning position in return. |
Chess computers will continue to lose to these types of players until they are programmed not to continue to go fishing after baited hooks, at the expense of position.
|Jul-22-09|| ||whiteshark: Blitz vs old Fritz 6 on a dusty P600. C'mon if you try long enough even you (a club player) will win a few games. I wonder what excuses you have for the hundreds of losses.|
|Jul-22-09|| ||birthtimes: Here's another Nemeth 10-minute game played against Deep Fritz in 2001 (not Fritz 6)|
1. e4 c5 2. Na3 Nc6 3. d4 cxd4 4. Bc4 Nf6 5. h3 Nxe4 6. Bxf7+ Kxf7 7. Qh5+ Ke6 8. Qg4+ Kd5 9. c4+ dxc3 10. Bf4 e5 11. O-O-O+ Nd4 12. Rxd4+ Kxd4 13. Nf3+ Kd5 14. Rd1+ Kc6 15. Nxe5+ Kb6 16. Nec4+ Kc5 17. Be3+ Kc6 18. Qxe4+ d5 19. Ne5+ Kc7 20. Nb5+ Kb8 21. Rxd5 Bd6 22. Nxd6 and White won.
I agree that even a club player could win some of these positions, and I believe that 90 or more out of 100 grandmasters would win this particular game after move 9 (especially if they were already familiar with the Nemeth gambit). Hopefully, today's Hydra, Rybka, etc., will not make the same mistakes that Deep Fritz did, during the first 5-9 moves. Or maybe they will...
Again, a game of chess is won only through POSITION, POSITION, POSITION...not material...
|Apr-19-10|| ||MarvinTsai: Some programmers tend to exaggerate the power of their software. I saw an article describing how a program defeat a super GM. The GM's play sounds normal and the program is just a little bit stronger to take down him tactically. The article emphasized on that it's the first time human super GM is defeat by a computer. But days later I found out the fact else where that it's just a simul...|
< Earlier Kibitzing · PAGE 2 OF 2 ·