Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!nrl-cmf!ukma!gatech!rutgers!elbereth.rutgers.edu!harnad From: harnad@elbereth.rutgers.edu (Stevan Harnad) Newsgroups: comp.ai Subject: Re: THE MIND EXTENDS BEYOND THE SKIN Summary: On the Functionalisms: Wide vs. Narrow, Symbolic vs. Robotic Keywords: Mind, Brain, Robots, Searle Message-ID: Date: 15 Mar 89 15:06:53 GMT References: <305@edai.ed.ac.uk> <7337@boulder.Colorado.EDU> <7773@venera.isi.edu> Organization: Rutgers Univ., New Brunswick, N.J. Lines: 107 ON THE VARIETIES OF FUNCTIONALISM: WIDE AND NARROW, SYMBOLIC AND ROBOTIC smoliar@vaxa.isi.edu (Stephen Smoliar) of USC-Information Sciences Institute wrote: " Searle wants to argue about whether or not a collection of simple " agents (call them neurons or symbol processors or whatever) can have " properties AS A COLLECTION which none of the components have as " individual members of the collection. Anyone who has worked with " systems know that such properties do indeed exist; but both Searle and " Harnad would have us believe that "understanding" is too sacrosanct to " be such a property. It absolutely astonishes me how a simple point can persistently fail to register if all of one's resources are committed to its contrary. I will repeat, patiently, for the Nth time, that "call them neurons or symbol processors or whatever" simply is not good enough, because it is just the critical difference between the two that's at issue here! Supposing one were talking about the critical materials needed to get electrical conduction and someone said "call them metal or rubber or whatever"! Or the critical function needed to perform work, and someone said "call it energy or entropy or whatever." I could go on and on. There is nothing sacrosanct about "understanding" (though humility dictates that we acknowledge that the mind/body problem has been proving to be a rather tough nut to crack). Both Searle and I believe that it can be done by a physical SYSTEM (e.g., the brain). The point is that we have been giving reasons (Searle, one reason, I several) why a symbol-crunching system is the WRONG KIND OF SYSTEM to generate understanding. What is the response to these arguments? Endless repetition of the claim that the problem with Searle's Argument is that he underestimates SYSTEMS -- as if he would deny that even neural systems could understand. This is just wasted words. If you want to rebut Searle, stick to what SYMBOL SYSTEMS (implemented in symbol-crunchers) can and can't do, and why. Don't hand-wave about "systems" in general. And while you're at it, try to address directly the points I've been raising about the specific limits of symbolic vs. nonsymbolic "systems." " Bateson... [says instead] that "mind" cannot emerge merely from the " interactions of components within an individuals body but, rather, from " the interactions of that body and its components with other exterior " entities. I made no reply to the Bateson-related postings because they seemed too metaphorical and remote: Bateson is an anthropologist, not a cognitive modeler. But even philosophers (who are likewise not mind-modelers, but often quite good at point out the silliness of some of the latter's shenanigans), in struggling with the problem of meaning and understanding, have proposed two kinds of "functionalism" (the position that mental states are functional states, and that meaning is a functional relation between words and states of affairs in the world). One form of functionalism is "narrow" or "skin-and-in" functionalism, according to which meaning is something that goes on between the ears of the candidate, and the rest is just the causal history of the candidate in the world. Certain "twin-earth" koans by Hilary Putnam and others have led some philosophers to prefer a "wide" functionalism, according to which the critical functional relation involved in meaning something includes a wider "system" than the candidate himself: a wide functional "state" includes the candidate plus objects and states of affairs in the world. One of the features of "wide" functionalism is that meaning something or knowing something has little to do with knowing that you mean something or know something. This is simply not pertinent to the subjective state I have been emphasizing in these postings -- namely, the subjective EXPERIENCE of understanding, meaning, or knowing something -- which is surely as "narrow" a "skin-and-in" state as pain is. There are no "twin-pain" problems. Hence I reject wide functionalism as either flummery or changing the subject. The task of the mind modeler is to produce a candidate that will pass the Total Turing Test (TTT). The only functions such a mind modeler needs to worry about are the internal ones. The rest is just a matter of generating the right outputs to the inputs. (And once the mind modeler succeeds, we must accept that the TTT-capable candidate understands -- or at least that we have no better reason to believe it does than we do to believe that any other person but ourself does -- and that we can never expect to be the wiser.) There are at least two "narrow" functionalisms available: I've dubbed them "symbolic functionalism," according to which the critical internal function for passing the TTT and having a mind is just symbol-crunching, and "robotic functionalism," according to which the critical internal functions will be nonsymbolic function, in which "dedicated" symbolic function is grounded bottom-up, and not isolable as an independent functional module. " this reminds me of the old puzzle about whether a tree which falls " when no one hears it makes a sound... As usual, those who use this old koan about the mind/body problem to lampoon philosophers are really just displaying that they don't understand it. ("Angels on a pin" is at least a coherent retort, though it too usually signals that the complainant has been missing something...) REF: Harnad, S. (1989) Minds, Machines and Searle. Journal of Experimental and Theoretical Artificial Intelligence 1: 5 - 25. -- 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