Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!rutgers!elbereth.rutgers.edu!harnad From: harnad@elbereth.rutgers.edu (Stevan Harnad) Newsgroups: comp.ai Subject: Re: Question on Chinese Room Argument Summary: On the Proper Treatment of the "Systems Reply" Message-ID: Date: 5 Mar 89 18:17:36 GMT References: <4298@pt.cs.cmu.edu> <7408@polya.Stanford.EDU> Organization: Rutgers Univ., New Brunswick, N.J. Lines: 163 geddis@polya.Stanford.EDU (Donald F. Geddis) of Stanford University wrote: " you are defining the "computer" to be the dumb processor " that interprets the rules. No one claimed that the processor (by itself) " *did* understand. I still haven't heard a satisfactory rebuttal to the " "Systems Reply", namely that (Searle + Rules) understands, whereas just " (Searle) doesn't. You've heard it. You just haven't understood it, because of a far-fetched and circular notion to which you and many others have become committed so strongly that the ability to un-commit in the face of logic and counter-evidence seems to have been lost: There's no other entity, no other eligible candidate for having a mind in the Chinese Room; nobody home! "Searle + rules" is a piece of cog-sci-fi. Do you believe that I could fail to understand, and alpha centauri could fail to understand, but "I + alpha centauri" could compositely understand? Could we compositely feel an itch that neither of us feels singly? If you don't PRESUPPOSE the far-fetched notion that the Chinese Room Argument set out to debunk in the first place, then you're less inclined to come back with it by way of a reasoned rebuttal! It's not that there CAN'T be a systems reply in principle: Searle COULD have been a non-understanding part of an understanding system. He could have been standing in, say, for the input and output of one neuron in a real brain. Then the system WOULD have understood and Searle would not have. But then neither would Searle have been performing ALL THE FUNCTIONS that were the substrate of the understanding. So it would be no surprise that he didn't understand. One of Searle's premises is that he himself must do EVERYTHING the candidate mental model does, yet not understand. (This is why my "robotic functionalist" counterargument works, and why the TTT is immune to Searle's Argument.) So you see, what Searle has really shown is not that no "systems reply" is tenable, but that a systems reply is untenable in the case of pure symbol-crunching, where Searle CAN do everything the system does. Symbol-crunchers are the WRONG KIND OF SYSTEM for having a mind (understanding, being intelligent [in the mental sense], etc.). " But we *can't* imagine what it would be like to process the Chinese Room " rules, because any set of rules that passed the Turing Test would be far " to large and complex by orders upon orders of magnitudes for a person to " actually (as opposed to simply in the thought experiment) process them. See earlier replies on "speed and complexity." This is just hand-waving. It's equivalent to taking a dumb toy model and saying "Just more of the same will pass the TT and will have a mind." I think the gap is not one of speed and complexity but missing, yet-to-be-discovered substantive functional concepts (and not just symbolic ones!). " [To solve the practical "other-minds" problem] I can use a biological " argument with humans [not just the TTT]... Robots don't get quite that " close an argument in their favor. We did many rounds on this on the Net 2 years ago: To synopsize: (1) The predictive power of biological facts about organisms that pass the TTT is parasitic on the fact that they pass the TTT. (Consider the problems we start to have as we move to species further and further than our own. -- Though let me hasten to add that I for one am fully inclined to give other species the benefit of the doubt about having minds, feeling pain, etc.) However, biological facts do, of course, have a secondary corroborative power. (2) We don't understand biological organisms or their brains functionally, hence we couldn't even answer the question, in principle, which one was a real organism and which was just a robotic look-alike! Another reason why biology and brain function can't help us much in our practical, everyday Turing Testing. " I don't think that those who argue for the Turing Test (your LTT) would " disagree that your TTT works as well, since it encompases the original " test. It is harder to pass. I'm not convinced that that is an " advantage, though. You've missed the point. Of course the Lingustic Turing Test is a subset of the Total Turing Test, and of course the whole Test would be harder to pass. But my point (and it had supporting arguments) was that it may well be that the only kind of device that could pass the LTT in the first place would have to be a device that could likewise pass the TTT, and that IN BOTH CASES it would have to draw essentially on nonsymbolic internal functions. " ["The TTT, unlike the LTT, is immune to Searle's Argument"] " Go through this again, please? I think that Searle doesn't have an argument " at all, but I fail to see how your TTT test makes any difference at all to " his analysis. At any rate, it certainly is not obvious that Searle's " argument is decisive, and that your reformulation is immune. But I'd be " interested in hearing your justifications. Here are points 3-7 from the summary and conclusions (p. 20 -21) of Harnad (1989) Minds, Machines and Searle. Journal of Experimental and Theoretical Artificial Intelligence" 1: 5-25 -- with apologies to those who have seen them before (they have already been posted once in their entirety at a poster's request). (3) The Convergence Argument: Searle fails to take underdetermination into account. All scientific theories are underdetermined by their data; i.e., the data are compatible with more than one theory. But as the data domain grows, the degrees of freedom for alternative (equiparametric) theories shrink. This "convergence" constraint applies to AI's "toy" linguistic and robotic models too, as they approach the capacity to pass the Total (asymptotic) Turing Test. Toy models are not modules. (4) Brain Modeling versus Mind Modeling: Searle also fails to appreciate that the brain itself can be understood only through theoretical modeling, and that the boundary between brain performance and body performance becomes arbitrary as one converges on an asymptotic model of total human performance capacity. (5) The Modularity Assumption: Searle implicitly adopts a strong, untested "modularity" assumption to the effect that certain functional parts of human cognitive performance capacity (such as language) can be be successfully modeled independently of the rest (such as perceptuomotor or "robotic" capacity). This assumption may be false for models approaching the power and generality needed to pass the Turing Test. (6) The Linguistic Turing Test versus the Robot Turing Test: Foundational issues in cognitive science depend critically on the truth or falsity of such modularity assumptions. For example, the "teletype" (linguistic) version of the Turing Test could in principle (though not necessarily in practice) be implemented by formal symbol-manipulation alone (symbols in, symbols out), whereas the robot version necessarily calls for full causal powers of interaction with the outside world (seeing, doing AND linguistic competence). (7) The Transducer/Effector Argument: Prior "robot" replies to Searle have not been principled ones. They have added on robotic requirements as an arbitrary extra constraint. A principled "transducer/effector" counterargument, however, can be based on the logical fact that transduction is necessarily nonsymbolic, drawing on analog and analog-to-digital functions that can only be simulated, but not implemented, symbolically. (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. -- 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