Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!iuvax!rutgers!elbereth.rutgers.edu!harnad From: harnad@elbereth.rutgers.edu (Stevan Harnad) Newsgroups: comp.ai Subject: Re: Fallacy in Chinese Room experiment. Summary: The Symbol Grounding Problem and a Chinese/Chinese Dictionary-Go-Round Message-ID: Date: 10 Mar 89 00:27:16 GMT References: <1059@jhunix.HCF.JHU.EDU> Organization: Rutgers Univ., New Brunswick, N.J. Lines: 72 ins_cscs@jhunix (Surag Surendrakumar) of The Johns Hopkins University - HCF wrote: " According to Searle... Understanding has this additional biological " aspect... Saying that there is some biological demon within you which does " the understanding, and not having any idea what it is [is] not a " scientific argument. Searle said nothing about biological "demons." He just said that there must be properties that the brain has that symbol-crunchers lack. He was perfectly right to say that, and his Chinese Room Argument was quite valid, as far as it went. But you wanted a specific candidate for what the missing function might be? I've proposed one in my JETAI paper: nonsymbolic function (e.g., transduction, analog processing, A/D, D/A, motor effectors). And before you give the reflexive response that all you have to do is "hook up" those processes to your symbol-cruncher and you're back where you started: In my paper I give reasons why grounding a symbol system is not just a simple matter of hooking on peripherals to a symbol-cruncher. The nonsymbolc function may be INTRINSIC to TTT-passing power (and hence mental function) and may not be isolable as independent symbolic and nonsymbolic "modules." " Searle says that by learning just the rules for Chinese he will be able " to pass the Turing Test but... not really know Chinese... I think that " it is going to be impossible for Searle to learn all those rules and " still not understand Chinese... You are [taught] Chinese by learning " the syntax and symbols. This is very much the same way that Searle " learns Chinese... One version of what I called in my JETAI paper "the symbol grounding problem" is the "Chinese-Dictionary-Go-Round": Suppose we had to learn Chinese AS A FIRST LANGUAGE and our only source of information were a Chinese-Chinese dictionary. In looking up any (for-us-so-far-meaningless) symbol or symbol-string, all we could find would be still more meaningless symbol-strings ("definitions"). The trip through the dictionary would be UNGROUNDED: It could never come to a halt on something other than meaningless symbols. How do we break out of this meaningless syntactic circle to meaning, reference, understanding? It's obvious that some, at least, of the symbols (the elementary ones) must be grounded in something other than still more symbols. (This is true even for second-language learning, except there we have a grounded first language as a leaping-off point [to which the computer in the LTT is of course not entitled]. This is the only reason that cryptological feats like the deciphering of the Rosetta stone are possible at all.) In the real world there are real objects, which produce proximal (so far NONSYMBOLIC) projections on our sensors. We somehow learn (or in some cases have evolved) the ability to reliably pick out CATEGORIES of objects on the basis of our sensory input and to assign to them a unique, arbitrary symbol. My book describes how nonsymbolic representations may play an essential role in our ability to do that, thereby grounding our elementary symbols in the objects they refer to. No demons are involved; but there turns out to be no natural "joint" at which one can carve cognitive function that would leave nonsymbolic processes on one side and an "understanding" symbol cruncher on the other. Refs: Harnad, S. (1989) Minds, Machines and Searle. Journal of Experimental and Theoretical Artificial Intelligence 1: 5 - 25. Harnad, S. (1987) (ed.) Categorical Perception: The Groundwork of Cognition. Cambridge University Press. -- Stevan Harnad INTERNET: harnad@confidence.princeton.edu harnad@princeton.edu srh@flash.bellcore.com harnad@elbereth.rutgers.edu harnad@princeton.uucp BITNET: harnad@pucc.bitnet CSNET: harnad%princeton.edu@relay.cs.net (609)-921-7771