Author: Wayne Lowrance
Date: 22:08:14 11/13/03
Go up one level in this thread
On November 13, 2003 at 21:30:31, Jaime Benito de Valle Ruiz wrote: >Ok, no one doubts that humans are much better (at the moment) at evaluating >positions. >For those who have studied maths -or have a natural intuition towards it-, chaos >is the (so-called) study of the logic underlying the supposedly unpredictable >outcome of a particular event, which in fact, it is not as random as it >appeared. >Chess offers virtually infinitely possible combinations, making it an attractive >and unattainable game to analyze (ultimately). However, all “logic” used to >evaluate this game has little meaning if a tactical combination takes a pawn >from you in 11 plies. Rules are far too complex to be reduced to a single >algorithm in this case. Inversely, all combinations are pointless if a long-term >well studied strategy defeats brute force. We don’t need to recall the so-called >horizon-effect! >The design of a computer-chess program has been pragmatically reduced to a >statistically successful design that blindly uses previous knowledge in >situations where brute-force doesn’t work, and makes a smart use of selective >brute-force where necessary. The result is a search algorithm capable of >punishing the slightest tactical mistake (enough to ensure a victory), and a >very systematic optimally-well-tuned algorithm capable of maintaining a game >against any opponent, no matter what the position is. Let’s be honest: Chess is >not a fully random map, where anything can happen, no matter how good or bad the >position looks like: There are trends; I believe that formal chaos analysis >should be taking chess as a prime study object. >Statistically, there seem to be too many tactical positions easy to miss by a >human, so the game remains even all the time. We get tired, we neglect >possibilities that we regard as futile, and we are far from being comprehensive. >A machine does not. Considering an almost-infinite analysis tree, most human >players fall most of the time into one of these tactical traps that will >inevitably lead to a loss for them. >Assuming a person does no tactical mistakes, this person still needs to outwit >the machine strategically in an extreme long-term plan. Here is where people >find their limitations: The human brain induces rules and develops new abilities >in a remarkable way, but induction is not a logical/full-proof mechanism! A GM >surely reaches a point where he can tell whether a game between two >useless-players is a victory for one side, or the other. This GM is using >previous knowledge in an inductive/holistic way, no doubt, and his – is >prediction will be most likely nearly perfectly accurate most of the time! The >problem is that these GM are basing their assumptions upon some past assumed >knowledge, and this knowledge has never been systematically put under test. >Machines have found unexpected outcomes from old positions (specially with the >use of tablebases), and very often, they offer unexpectedly resistance against >positions that were supposed to be lost. >Machines are showing us logically sequenced paths that we never considered in >certain positions where any small variation can decide a game (tactical ones). >They have opened new ways of thinking about chess: a dull and scary one, >actually. People now know that it doesn’t matter how “good” they play now, >because chances are that even they won’t be able to keep a perfectly-non blunder >tactical game all the time, and the machine will take profit of it at some >point, unbalancing the game and massacring you! Keep yourself away from this >danger… and you’ll have to find a way of outwitting a long-term plan, previously >considered by the machine several plies ago, that will surely lead you to a >drawish position (if you’re lucky). >Let’s be fair: If you want to systematically beat a good machine, you have to be >not only tactically perfect, but also able to strategically outperform your >engine to a point where its evaluation algorithm (and its extensions) does not >detect any treat! We’re asking too much. Surely some GM can do this…. sometimes. >For the rest of us, it’s like a battle against a calculator to see who can to a >root 27 of 0.223254242. Forget it! >Chaos guarantees some of these unexpected “lines of play” where the score can be >successfully inverted because of a 20/30 plies horizon effect, but statistically >discourages players who attempt to beat computers. >Let’s face it: We’ll never get a root 33.4 of 0.2324242 faster than a >calculator, as well as we’ll never run faster than a 1200 cc motorbike. >This doesn’t mean that we cannot give a calculator a problem that it cannot >solve, or that we cannot climb a mountain faster than a motorbike (not design >for climbing). I’m not interested in being able to hit a screw harder than my >hammer! Chess computer machines have become very efficient in finding >complicated routes that lead to a material victory, and strong enough in >challenging GM in non-tactical positions. The only fun we can get is betting >about their strength or by finding positions that we “assume to be easy” but >machine cannot see at first sight”. >I’m not good at chess myself, but I don’t plan to waste my life trying to outwit >a “chess-calculator” at positions where my chances are close to nothing. >Kasparov is a person; an amazing player, no doubt. But he cannot keep a 100% >reliability in tactical moves against a computer (or other player), and this >makes him vulnerable to the machine-trap. Any little-tiny mistake, and he’ out!! >IMHO he kept a decent/balanced game against Fritz, until he neglected this move… >Even one of my old-engines (with no Hash tables or Futility pruning) happily >takes that pawn and gives a score of nearly +2.00!!! That’s a human blunder. >We’re likely to make them… sooner or later! >These computer-human games are just getting boring to me: Unless the human plays >a confidently-non blunder-game from the start to the end, there is no fun to >watch!! If the person is so confident/so precise, still has to find a long-long >term strategically well design plan to beat the machine. Statistically speaking, >chances are quite low. >Why do we keep worrying about human-machine war at chess? I no longer care about >Motorbike-human speed championship!! >Are we going to forget about adding or multiplying just because calculators can >do it faster and more accurately? NO. Are we going to loath chess just because >they always win? No way! It still amuses us! Are we going to stop running just >because our old car is faster than we can be? What about writing faster than a >printer? Life is not just a competition! >As computer-chess fans, we find this hobby intellectually and challenging >amusing, because the ultimate algorithm for perfect chess is still far to be >found. There is still room for opinion and human intuition… as well as >“programmer-skills” rewards for some of us! Chess is still a formal space where >the number of solutions can be described as a finite (huge number) sequence, and >formal proofs can be possible. I think it’s time to change our philosophy about >chess: We cannot win in certain positions unless we’re infallible (and no human >is). But we can still improve the game by suggesting new innovations and make it >more interesting! >Whenever I see a GM-machine game, I look attentively at the “expected” score (in >my case usually Shredder), and I wonder about their game… if the score suddenly >goes up to 1, more than 1, or even more for the machine, I know that the “human” >has just blundered. No way back! The machine will win here! I’ll be very >surprised if this evaluation is flawed!!! Whenever this score remains negative >against the machine, I still believe there is a (small) chance for the human >side; but it is still an extremely improbable fight against a “calculator” that >will make no tactical mistake. In mathematical terms, the chance to outwit the >machine in a deeply-comprehensive analysis space such as chess, becomes smaller >and smaller as plies move on! >I’ve seen Kasparov-Fritz games. I’m not good at chess, but I have several >chess-engines helping me to understand. Most moves were nearly forced. The >situation was “arguably” better for either side, no matter which one. >Let’s enjoy our debates about computer chess programs! >There’s too much dogma here! >Who cares about human-computer supremacy? (My pocket calculator can beat you all >for sure at simple calculations) >Let’s enjoy the chaotic unexpected result of computer vs computer results. >Regards, > > Jaime Okey dokey, i'm with you. Nice work Wayne
This page took 0 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.