< Earlier Kibitzing · PAGE 12 OF 12 ·
|Feb-14-13|| ||WiseWizard: I applaud chrisowen for simply never breaking character, quite the achievement.|
|Feb-14-13|| ||Gypsy: <AylerKupp> Thank you friend.|
|Feb-14-13|| ||HeMateMe: I've never read a full Chris Owen post.|
|Feb-14-13|| ||TheFocus: Well, <chrisowen>'s posts are actually Swahili translated through Klingon and then rendered into Antartican.|
Get through those and you have it crystal clear.
Keep on truckin', <chrisowen>!! You da man!
|Feb-14-13|| ||HeMateMe: Maybe he is posting from the computers at MI-5, and this is the only "language" he's allowed to input on the keyboard?|
|Feb-14-13|| ||Eyal: <AylerKupp: OK, here is Critter 1.6a's top line at d=28 starting form the game's final position. |
1. [+0.42]: 45.h4 hxg4 46.Rxg4 Kc6 47.Rc4+ Kb7 (why this retreat from the center and k-side? Why not 47...Kd7?) 48.Rf4 (why not 48..Rxb4? A pawn is a pawn) [...]
But, of course, none of this is forced and many of the moves don't make much sense to me. You could think of Critter and this line as the chess engine equivalent of <chrisowen>, deep but incomprehensible.>
I've looked at the position after the "obvious" 48.Rxb4 a bit with my Houdini (which seems to share Critter's basic preferences) - it seems that with their tactically-oriented "thinking", the engines see in such a case enough counterplay for the black pieces that should lead to further drawish liquidation of pawns, without special problems for Black (especially since White's extra pawn is doubled).
Here are some sample lines after 48.Rxb4 Rxb4 49.Nxb4 Ba5: 50.Rd4 Re8 (aiming for Re3+) 51.Kf3 Rf8+ 52.Ke2 (or 52.Kg4 Rf1) 52...Rh8 53.Nc2 Bb6 54.Rxd6 Rxh4 55.Rxg6 Rh3 56.b4 Rb3; or 50.Nd5 Rh8 51.Nf4 (51.Nc3 Bd8 52.Rd4 Rh5 preparing g5) 51...Be1+ 52.Kf3 Rxh4 53.Rxd6 g5 54.Nd3 Ba5.
It's a bit funny, really, when you think about it - people who don't know much about engines tend to have the notion that they're very materialistic, but in fact today's top engines seem to be considerably less materialistic than most human players - from my experience, they usually have no qualms at all about giving up pawns for good piece activity.
|Feb-14-13|| ||Hesam7: After the game Anand said that his mistakes were 16...Rab8 & 24...Rb6. He also said that he wanted to avoid a repeat of Judit Polgar vs Anand, 2008.|
I was surprised that Anand did not play 14...Nh6 15 Qh6 Nd7 it seems to me that Black gets a slightly improved version of the game, for example: 16 Rfd1 Bf8 17 Qe3 Rac8 18 Nd5 Qc5
click for larger view
|Feb-14-13|| ||AylerKupp: <Eyal> You're quite right. It's a carryover from the "old" days when they were indeed materialistic and there was even one engine which played reasonably good chess using material as the only criteria in its evaluation function. But today I have seen instances of engines giving up a pawn for purely positional considerations, and give the resulting position a favorable evaluation even though they were still at a material disadvantage at the end of the line they displayed. Yes indeed, engines these days are very good indeed.|
But even in the "old days" (1971) engines were capable of sacrificing material for perceived positional advantages. I can't resist reposting my favorite game finish from "How Computers Play Chess" by David Levy and Monty Newborn, which is good for a laugh. White was COKO III and Black was GENIE, and they arrived at the following position after Black's 37th move, probably 37...g5.
click for larger view
The book says that "It would appear here that White held some advantage" which is certainly an understatement. But the embarrassment of riches became an embarrassment for COKO III since instead of ending the game by 38.Qb2# or 38.Bc4# it played 38.Kc1 since it evaluated an unpreventable mate in one the same as an actual mate. The game continued:
38...f5 39.Kc2 (mate in one is still not preventable, so this was just as good as giving mate) 39...f4 40.Kc1 g4 41.Kc2 f3 42.Kc1 fxg2 43.Kc2 gxh1=Q
click for larger view
Well, COKO III made it harder on itself but it can still gain the full point by 44.Qb2# or 44.Bc4#. Instead it played 44.Kc1 and the game ended 44...Qxf1+ 45.Kd2 Qxf2+ 46.Kc1 Qg1+ 47.Kc2 Qxh2+ 48.Kc1 Qh1+ 49.Kc2 Qb1+ 50.Kd2 g3 51.Qc4+ Qb3 52.Qxb3+ Kxb3 53.e4 Kxb4 54.e5 g2
click for larger view
And at this point COKO III's programmers resigned on behalf of their program. So I guess that engines have indeed gotten better.
|Feb-14-13|| ||Fish55: <AylerKupp> That's hilarious!|
|Feb-14-13|| ||Tomlinsky: <AK: So I guess that engines have indeed gotten better.>|
Running a Fortran program on an IBM 360 with 512KB RAM and a CPU delivering ~1.25MIPS one would hope so.
|Feb-14-13|| ||AylerKupp: <Tomlinsky> Sure, but it's not just the hardware that has gotten better. The software has also. And, to show how much better, here is a link to my favorite computer vs. computer game involving Rybka with 260 cores (!) against Jonny with 800 cores (!!!) as only <kingscrusher> can describe it: http://www.youtube.com/watch?v=CPFi....|
|Feb-15-13|| ||JPi: Funny game <AylerKupp> but as for pawn sacrifice to connect pieces of all part of the board to an attack, among many Pillsbury vs Lasker Nuremberg 1896(!)impressed me more.|
|Feb-15-13|| ||Eyal: Getting back to this game, it's quite funny at any rate that - as the round report on the official site notes - when Caruana accepted Anand's draw offer, Houdini (and apparently other engines as well) rate the line with 45.h4 as better for White than any other position that occurred in the game:|
|Feb-15-13|| ||OhioChessFan: 24..Rb6 was so ugly it hurt my eyes to look at it.|
|Feb-15-13|| ||Ulhumbrus: <chrisowen> Perhaps you can provide a first introduction to the grammar of your language. Can you tell me how you would say in your language, respectively, < I am, we are, you are, you are, he is, she is, it is, they are>|
|Feb-15-13|| ||chrisowen: <aylerkupp> <wisewizard> <thefocus> <eyal> <ulhumbrus> H5 for cyperus papyrus in scroll it evermore in heading her kindred h5 undulate again
empty black spite in a lesson sixty ninety plus moves to finish him off time advantage once promote nevermore in culpable why the big trekking b3 and b4 b5 in back in get off scott free in dutifuls crum bleed in give edge to white forge in g3 a never fear in here knight pave in d5 when position gauged it draw in organise key together blinking d7 no good you in church of reason d8 in g3 neck deed in will again nearly either dint bang one hammer king see victory it delivery in re d8 at a5 in bide it is your time, zealot in for pathway it is coin a term
in herrock a bill to rights so screaming rocket ala dink h5 ergo g3 lol free g4 you see the path he king of king it sand in wedges it h6 in sitter h5 free goodbind!|
|Feb-15-13|| ||TheFocus: <chrisowen> I can't argue with that at all.|
|Jan-17-14|| ||RandomVisitor: <AylerKupp>I pulled out my copy of Levy, Newborn, "How Computers Play Chess" and re-read the COKO III story.|
"Had there been only one forced mate in this position COKO III would have played it, but because there was more than one mate and because COKO III's scoring function did not distinguish between the value of a mate-in-one and and a mate-in-two, -three or -four, it had no incentive to choose the quickest mate."
So, this rare mate condition revealed what was essentially a programming bug.
|Aug-01-18|| ||keypusher: <Well, COKO III made it harder on itself but it can still gain the full point by 44.Qb2# or 44.Bc4#. Instead it played 44.Kc1 >|
AylerKupp -- that's a great story about COKO III. I'd read about it many years ago but forgotten all the details.
I still don't understand 44.Kc1 though. Couldn't the program "see" that there was no longer a mate in one (or two or three or four) if it allowed 44....Qxf1+?
|Aug-03-18|| ||AylerKupp: <keypusher> You're not the only one who doesn't understand 44.Kc1. Levy's book says after 44.Kc1: "Inexplicable! COKO III's programmers were so distraught now that they could not even speculate as to why COKO III had missed this final chance to mate." But I would think that COKO III's programmers were very familiar with the concept of an infinite loop.|
I'm sure that you're familiar with the famous Levy bet. Apparently, just before this COKO III vs. Genie game was played, one of the COKO III programmers, Ed Kozdrowicki, had a discussion with David Levy and he decided to join the bet for an additional US $ 1,000. After this COKO III vs. Genie game was over, Levy said that Kozdrowicki was heard muttering something about a "bad bet" as he left the playing hall.
|Aug-03-18|| ||NBZ: <AylerKupp>: That is an incredible story, thank you for sharing it. Can you explain a little more by what you mean by the infinite loop leading to 44. Kc1? My understanding is that Koko assigned a higher value to winning more slowly; that's why a mate-in-two was evaluated higher than a mate-in-one. Did Koko evaluate the position after 44. Kc1 as still winning, and thus preferable to the instant win via Bc4 or Qb2? Or is it that Koko just could not select between Qb2 and Bc4, and went for the third-best option? The latter might be a very human way to react when faced with impossible choices, but I am not sure why the computer would do that!|
|Aug-04-18|| ||AylerKupp: <<NBZ> Can you explain a little more by what you mean by the infinite loop leading to 44. Kc1?>|
Sorry about. After 40 years as a software developer and manager I'm so used to the concept of an infinite loop that I forget that others are not necessarily as "fortunate". There are 3 basic constructs in all coding; the straight sequence, the decision making, and the loop. Each loop has to have a way of terminating but, if the programmer makes an error (and all programmers do, some just more than others) the loop may not terminate and so the program would run forever. That's why it's called an infinite loop.
I'm not necessarily saying that this is what happened, but it sure looks like it. The book ("How Computers Play Chess" by David Levy and Monty Newborn") does not go into specifics, but it does say (not necessarily correctly, Levy was not a programmer) "It would appear that White held some advantage. He had at his disposal two different mates-in one (38.Bc4# and 38.Qb2#) as well as various mates-in-two, -three, -four or indeed almost any number of moves. But because of a defect in the program this plethora of mating continuations led to COKO III's tragedy. Had there been only one forced mate in this position COKO III would have played it, but because there was more than one mate and because COKO III's scoring function did not distinguish between the values of a mate-in-one and a mate-in-two, -three, or –four, it had no incentive to choose the quickest mate."
So I don't think that it was a case that COKO III assigned a higher value to winning more slowly, that wouldn't make any sense to me. Then again, you never know. I think it's just because apparently it couldn't select between the various mates since they were valued the same so it might have been just by chance that it selected a mate-in-two by 38.Kc1 over a mate-in-one by 38.Bc4# and 38.Qb2#. And, once it did that, there was no way to get out of it until it had to respond to the check after 44...Qxf1+.
Keep in mind that this game was played in 1971 when computer chess was still in its infancy (some will claim that it still is) so mistakes in coding were commonplace. I remember seeing one game reported in either Chess Life or Chess Review (they used to be separate magazines) when, with a pawn on c2 and a knight on c3 the computer played c2-c4. It didn't even know that this move was illegal under these circumstances!
|Aug-04-18|| ||john barleycorn: <AylerKupp: <<NBZ> Can you explain a little more by what you mean by the infinite loop leading to 44. Kc1?>|
... Each loop has to have a way of terminating but, if the programmer makes an error (and all programmers do, some just more than others) the loop may not terminate and so the program would run forever. That's why it's called an infinite loop. ...>
So, an "infinite loop" lead to 44.Kc1 and just kept going on leading there?
Is that an "infinite loop" or a "loop run through infinitely"? or are they the same thing?
|Aug-05-18|| ||AylerKupp: <john barleycorn> Is that an "infinite loop" or a "loop run through infinitely"? or are they the same thing?>|
"Infinite loop" is the usual term. I've never heard the term "loop run through infinity" but I would suspect they mean the same thing. For instance, I have a book titled "Infinite Loop – How Apple, the World's Most Insanely Great Computer Company, Went Insane". Which, given Apple's recent US $ 1T evaluation, means that a lot of insane people are laughing all the way to the bank.
|Aug-10-18|| ||NBZ: <AylerKupp>: Thank you for the explanation! Very helpful for a non-programmer.|
I don't know if you are familiar with Isaac Asimov's books? In his world, the robots have to make choices that satisfied the Three Laws of Robotics. Occasionally they just would not be able to choose between two options that they rate equally and then they would go into a "positronic freeze-out", which basically means their circuits would burn out. A skilled robotician who wanted to disable a robot could do so by generating a situation where the robot would be forced into a freeze-out.
It strikes me now that this is really similar to the "infinite loop". In fact I would not be surprised if Asimov was inspired by the infinite loop when he came up with this plot device.
< Earlier Kibitzing · PAGE 12 OF 12 ·