Path: utzoo!attcan!uunet!lll-winken!ames!ucsd!rutgers!elbereth.rutgers.edu!harnad From: harnad@elbereth.rutgers.edu (Stevan Harnad) Newsgroups: comp.ai Subject: Re: Question on Chinese Room Argument Summary: More Cog Sci Fi: Simulated Worlds Message-ID: Date: 18 Mar 89 19:46:02 GMT References: <4298@pt.cs.cmu.edu> <416@censor.UUCP> Organization: Rutgers Univ., New Brunswick, N.J. Lines: 153 jeff@censor.UUCP (Jeff Hunter) of Bell Canada, Business Development, Toronto wrote: " 1) the TTT: (the same "irrelevant" example) Remove a person's brain and " replace it with a detailed, real-time atomic-level simulation. Add " transducers to convert real to simulated neural signals, and vice versa. " Now, this new person's simulated brain adequately reproduces "ALL " of the relevant causal and functional principles" of the real brain, " including the nerve signals. The behaviour of the human is unchanged. " The human continues to be able to pass the TTT, although he is now a " symbol cruncher with added peripherals. Please rebut, or stop talking about " the TTT. " I believe that the "symbol groundings" that you have "proven" to be " required for understanding systems are nothing more than the rules " governing how a transducer handles a signal. This is not to say that " your "mixed approach" as outlined recently will be fruitless. It " certainly uses more realistic hardware than my examples :-) However you " have failed to prove that it is NECESSARY for any conceivable tasks. You can't ask me to rebut anything I haven't claimed. I think I have been perfectly explicit about what I have and have not shown. It is now up to you to read and understand it. Here is part of a passage from my JETAI article that I have already posted to this discussion: (8) Robotics and Causality: Searle's argument hence fails logically for the robot version of the Turing Test, for in simulating it he would either have to USE its transducers and effectors (in which case he would not be simulating all of its functions) or he would have to BE its transducers and effectors, in which case he would indeed be duplicating their causal powers (of seeing and doing). (9) Symbolic Functionalism versus Robotic Functionalism: If symbol-manipulation ("symbolic functionalism") cannot in principle accomplish the functions of the transducer and effector surfaces, then there is no reason why every function in between has to be symbolic either. Nonsymbolic function may be essential to implementing minds and may be a crucial constituent of the functional substrate of mental states ("robotic functionalism"): In order to work as hypothesized (i.e., to be able to pass the Turing Test), the functionalist "brain-in-a-vat" may have to be more than just an isolated symbolic "understanding" module -- perhaps even hybrid analog/symbolic all the way through, as the real brain is, with the symbols "grounded" bottom-up in nonsymbolic representations. In other words, the only function that I have shown to be NECESSARILY immune to Searle's argument is transducer/effector function. But now consider the following: (i) If you examine the brain with a view to slicing off its "transducers" and "effectors," you come up against a problem, because even if you yank off the sensory surfaces, what is actually left over is repeated analog transforms of the sensory surfaces as you go deeper and deeper into the brain. Do you ever reach a point where sensory function leaves off ("cut here") and symbol crunching takes over? No. What you find is that it grades (in a way that is not understood) into sensory-motor function (modulated by arousal, attention and affective functions), and then into pure motor analogs, leading to the "effector" mechanism. So if you yank off the "transducer/effectors," you've got no brain left at all! (This is not to imply that the areas in question are just dumbly reproducing the sensory surfaces over and over, but that "transducer-effector" function seems to be intimately and intrinsically involved in everything the brain does -- and there's no evidence whatsoever that the functions with which it is so closely intertwined consist of anything like symbol-crunching either.) (ii) So the argument is really this: Searle has successfully shown that symbol-crunching ALONE is not the function that gives rise to a mind. It IS logically possible that if we hook up the same symbol-cruncher, which we just showed to be totally inert and mindless, to the set of sensors that opens and closes the doors at Woolworth's, suddenly the lights go on and there's an understanding mind in there! Searle's Argument cannot show it to be false that the system "symbol-cruncher + tranducers" understands, but I somehow doubt it's true anyway (almost as much, but not quite, as I doubt that "symbol-cruncher + chalk" can understand). But before you rush to say "just make the transducer/effector function more complicated and it'll work," I have to remind you of how it seems to have turned out in the case of the real brain: As the "transducer/effector" function approaches the requisite "complexity," it begins to grow to the size of almost of ALL of the brain's functions, and the corresponding room for symbol-crunching shrinks proportionately. [Let me add that in the JETAI article I also noted that it's an empirical question just how much of the internal functioning of the brain or any other TTT-passing candidate is or can be pure symbol-crunching (computation). There's no logical reason why some of it shouldn't be. For example, my own hybrid symbol grounding scheme, described in the Categorical Perception book, has a symbolic component too, but it is a specialized and dedicated one, with its elementary symbols grounded bottom-up in nonsymbolic representations. It is not an independent module; there's no place to "cut" so as leave transducers on one side and a pure symbol-cruncher on the other.] So the conclusion is this: Mental function cannot be just symbolic function. That's been shown. What function(s) it actually is or can be remain to be shown (by finding out what function(s) are sufficient to pass the TTT). "Symbol-crunching + transduction" (jointly) is still in the running, as a logical possibility, but it hasn't got much going for it empirically or conceptually. (iii) The formal power of computer simulation ("Turing Equivalence" -- not to be confused with the Turing Test) seems to have gone to some people's heads. You are free, if you like, to think of an airplane as just a set of transducers/effectors hooked up to a symbol-cruncher, but not many of the functions of FLYING will be generated by the symbolic component (mainly just the already computerized cockpit functions). When the plane is actually flying, almost all of the real work will be done by the nonsymbolic component ("transducers/effectors") rather than the symbolic one. I hope you can still see that -- and that it would be silly to speak of this as a plane at all if you removed its "transducers/effectors." If you can see it in that case, then try to see that the TTT-robot case is exactly the same. " if I say "Hawking has a deep understanding of theoretical physics that " has allowed him to make brilliant contributions to scientific " knowledge." the sentence does not depend on whether or not Hawking's " brain has been replaced with a simulation. Nor would it depend on it if you were talking about Ed Whitten. So what has Hawking's tragic handicap got to do with it? You're getting some spurious mileage by implying that Hawking, because of his infirmity, is closer to being a pure symbol cruncher. But that's simply false. He has a brain like anyone else (apart from the motor infirmity) and is able to draw on exactly the same nonsymbolic functions that the rest of us draw on. So why mention his case at all? " Consider this: on New Year's eve the entire solar system was " digitized and replaced by a symbolic simulation, a black hole, and a shell " of transducers to translate incoming/outgoing radiation. Your current " thoughts are simulations as are the ones at the time you wrote the above. " Rebut, or concede that introspection does not capture " Searle-understanding (and vice versa). As in the case of the plane and the brain, you are free to imagine the solar system (or the universe?) as just a bunch of transducers hooked to a symbol cruncher. So what? The vast majority of their critical functions will continue to be the nonsymbolic ones. Refs: Harnad, S. (1987) (Ed.) Categorical Perception: The Groundwork of Cognition. NY: Cambridge University Press. 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