Author: Terry McCracken
Date: 08:31:45 03/26/02
Go up one level in this thread
On March 26, 2002 at 10:42:19, Chris Carson wrote: >On March 26, 2002 at 09:30:46, Terry McCracken wrote: > >>On March 26, 2002 at 07:34:46, Chris Carson wrote: >> >>>On March 26, 2002 at 04:33:14, Daniel Clausen wrote: >>> >>>>On March 26, 2002 at 00:35:44, Slater Wold wrote: >>>> >>>>[snip] >>>> >>>>>Whether or not I believe computers play GM strength chess is no matter here. >>>>>What matters is that people who have given 30+ years to this field have more of >>>>>a right to an opinion than you *EVER* will. >>>> >>>>I don't think that he is more entitled of an opinion than any other here. The >>>>difference between what Bob says and some other says is that he does it in a >>>>scientific way. >>>> >>>>Jerry (among others) seem to be very emotional about this matter (which is not >>>>necessarily a bad thing) and are so fixed on their opinion, no matter what. They >>>>don't have the patience to collect enough data points before making a claim. >>>>Also they're not very critical when coming up with examples which 'prove' their >>>>'obvious claim'. Ie Mr Ballicora pointed out some things why some/most GMs are >>>>not _that_ interested in playing against computers and therefore don't take it >>>>as seriously as playing vs humans. These things have to be taken into >>>>consideration, as well as many other things. >>>> >>>>Sadly, many (if not most) people fail to think that way and prefer the >>>>unscientific way. Computers showed very impressive performances over the years >>>>without us inventing new ones w/o enough data to support them. >>>> >>>>Sargon >>> >>>Careful about Science vs Faith. There is plenty of data points to prove the >>>strength of computers vs humans. I work in the field of human behavior research >>>for a living, no ties to the commercial programs and have studied Human vs >>>Computers for a long time (many years). I do not care what the strength of a >>>program is (higher or lower), however, they have proven themselves to be >>>significant and at a 2700+ level on hardware that can be purchased for under >>>$2,000. >>> >>>Humans must perform significantly better than the top programs over a period of >>>200 games or more before you can say with any scientific certainty that they are >>>below 2700. The programs have proven themselves to be very strong over a large >>>number of games at 40/2 (even more at faster time controls). These games must >>>be in open viewing competition, tournament rules with an arbitrator. >>> >>>Ofcourse everyone has the same entitlement to an opinion (based on faith). I >>>hope the GM's do better, I have faith that humans can adapt quicker than a >>>static program on static hw, however, science and a lot of data show that the >>>top programs are 2700+ and are very dangerous opponents and they will only get >>>better with improved s/w and faster hardware. >> >>As a scientist I don't know how you can claim factually computers and software >>of today perform over 2700 at 40/2 T/C's. Unless we're talking about Deeper >>Blue, then I would tend to agree but not with absolute certainty. >> >>The "Verdict" is not "Out" yet and this attitude, can lead to premature >>conclusions. It's damaging. >> >>This is what happened with Deeper Blue in '97 and has fudged the science. >>Deeper Blues' win over Kasparov happened before it's time due to many reasons >>and "Pure Science" got second spot. >> >>The public and media are "Ignorant" even the "Business" people at IBM, except to >>"Turn a Buck", but the best players and even experts "Understood" why Kasparov >>really lost and it wasn't due to Deeper Blue bieng the better player. >> >>Believe it or not except for game 6 which hardly could be considered chess for >>obvious reasons, Kasparov outplayed Deeper Blue. >> >>I'm not going to re-hash the details. It's been discussed far too many times >>here already and people seem to remain unconvinced. Too bad. >> >>Until computers play thousands of games against GM's at 40/2 T/C's for "High >>Stakes" then much of the data is questionable. >> >>Much of the data you're refering to has holes all through it and is not >>convincing. >> >>Terry > >Thousands of games are not needed. Medium effect size with power .8 single >tailed only requires 52 subjects, even if the effect size was small (and it is >clear that it is not), then only a couple of hundred games would be needed. >This yeilds a confidence of 95%. > >There are plenty of games. There are plenty of games at ICC where GM's and IM's are still winning more than losing to comps. How many more at home? Who knows? That _must_ be factored in except what can't be known:) Terry
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.