Path: utzoo!utgpu!watmath!clyde!bellcore!rutgers!aramis.rutgers.edu!klaatu.rutgers.edu!josh From: josh@klaatu.rutgers.edu (J Storrs Hall) Newsgroups: comp.ai Subject: Re: Sound and complete definitions of intelligence. Message-ID:Date: 8 Dec 88 23:59:41 GMT References: <2768@uhccux.uhcc.hawaii.edu> Organization: Rutgers Univ., New Brunswick, N.J. Lines: 50 I wrote: " Moravec estimates 10 teraops/10 terawords to be human-equivalent " computational power. ... Greg, lee@uhccux.uhcc.hawaii.edu replied: Surely such estimates are frivolous. They are not. Let me reccomend to you not only Hans' published work but Sejnowski in AAAI-88 and Merkle in AIAA Computers in Aerospace 87. It is obviously of critical importance to AI to have some understanding of the size of the problem it is trying to solve, relative to the tool they are trying to use. Surely the estimates are imprecise--I haven't seen even one that claimed to be better than order-of-magnitude-- but estimates I have read from widely varying sources fall into the 10e12 - 10e15 range with surprising consistency. You should not so blithely dismiss an area where serious, informed estimation dates back to von Neumann ("The Computer and the Brain"). We don't know what or how humans compute in at least one crucial area, language, except functionally by the gross results we can observe. So what? The parts we do know about, the retina for example, give us some guidelines for estimating an upper bound for whatever computation is being done. And we can theorize and conjecture. Our estimates may be wrong, but they are not frivolous. Could you estimate the computational resources consumed by an unknown program executing under an unknown operating system given some small samples of its input and output and fragmentary information about the device in use? Not feasible, without (re)constructing the program, at least, which we haven't yet managed to do for humans. Again, let me start vivisecting the computer with appropriate test instruments and I can begin to give you some believable upper and lower bounds. ...Mind you, this is not to say that there aren't significantly better ways the brain could be doing some of the things it does. Consider what a Cray could do to all those long division problems you slaved over in grade school. And the existance of "idiot savant" human calculators proves that there are significantly faster ways that even the brain can do some things like that. --JoSH