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: The Logic of the TTT (vs. Doin' the Hokey-Pokey) Message-ID: Date: 3 Mar 89 04:55:03 GMT References: <4298@pt.cs.cmu.edu> <9763@ihlpb.ATT.COM> Organization: Rutgers Univ., New Brunswick, N.J. Lines: 88 arm@ihlpb.ATT.COM (Macalalad) of AT&T Bell Laboratories writes: " Argument: In order to come to any resolution about what systems can " or cannot understand, we need an objective theory of understanding, " rather than the behaviorist "I'll know it when I see it" tests such " as the Linguistic Turing Test (LTT) or the Total Turing Test (TTT). Counterargument: To ascertain (beyond reasonable doubt) that a system CANNOT understand, you don't need a theory. Searle's argument is a case in point: If Searle (or you, or me) does exactly what the computer does but does not understand, then the computer does not understand. " you may be totally confident that you know when you understand " a language, but to conclude that you know when another entity understands " a language is a leap that I'm not quite ready to make. No need to make the leap. Just know when you yourself don't understand (in doing exactly what the symbol cruncher does) and infer that nothing/no-one else doing exactly the same thing can be understanding either. " As I understand it, TTT merely involves applying our practical solution " of the other minds problem to AI systems. In other words, TTT says, " "I'm not going to define how a system understands a language, but " I'll know it when I see it." No. The logic of the TTT is this: I have no other basis but the TTT for my confidence that other PEOPLE have minds, therefore it would be arbitrary of me to ask MORE of robots. This is only a practical, not a principled solution to the other-minds problem, however. Hence the same uncertainty remains, in both cases (human and robot). " I can turn around and say, "It sure acts like it understands, but look " at the underlying architecture. It's a machine, and MACHINES CANNOT " UNDERSTAND. (or it's a formal system, or it's non-biological, or it's " only a simulation, or any other hokey excuse)... Now we're at a " standoff, where one believes that the given system does understand " because it passed the TTT, and the other other just as firmly believes " that it does not understand because the underlying architecture is " incapable of believing. All because of the refusal to commit to a " definition. To say "machines can't understand" is to beg the question. (We don't even know what "machines" are -- and aren't -- yet.) "Wrong architecture" simpliciter is arbitrary too: What's the "right" architecture? No one knows what the brain's functional "architecture" is, or what aspects of it are necessary or sufficient for having a mind. To say it doesn't understand because its nonbiological is also to beg the question. To say it doesn't understand because it's just doing formal symbol manipulation calls for an ARGUMENT: Searle has given one. ("Simulation" is equivocal; even Searle's "simulated forest fires don't burn" argument is enough to handle that -- but it all boils down to whether symbol manipulation alone is enough not only to simulate the mind but to implement it.) So it still has nothing to do with definition; and there's clearly room for plenty of hokeyness on both sides. (You made a crucial error, by the way, in referring to the TTT above; you should have said the LTT. That's the one the two sides are disagreeing on, and that's the one Searle's argument is decisive against. The TTT is immune to Searle's argument and I've so far heard no non-hokey objection to it.) " I don't know about you, but I certainly don't include neuronal behavior " in any of my practical solutions to the other minds problem. I certainly don't either. That's why I don't accept "wrong internal functions" as an argument in itself: We have no idea what internal functions are "right" or why; and only the TTT can lead us to an answer. However, as I said, down the road a ways toward TTT utopia, brain "performance" may eventually provide a useful "fine tuning" variable on our near-asymptotic candidate. (The main problem with brain function when you're far from utopia -- besides the fact that we don't know what it is, and have no idea how we could find out by peeking and poking at the brain -- is that we don't even know what aspects of it are relevant and what aspects are irrelevant.) Ref: Harnad (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