Path: utzoo!attcan!uunet!aplcen!uakari.primate.wisc.edu!zaphod.mps.ohio-state.edu!swrinde!emory!mephisto!mcnc!rti!ntpdvp1!kenp From: kenp@ntpdvp1.UUCP (Ken Presting) Newsgroups: comp.ai Subject: Re: Searle and Radical Translation (was: Re: Searle and Biology) Summary: I am NOT defending Searle. He is wrong, but so is the Systems Reply! Message-ID: <609@ntpdvp1.UUCP> Date: 23 Jul 90 16:00:26 GMT References: <14265@enera.isi.edu> <602@ntpdvp1.UUCP> <1595@oravax.UUCP> <1606@oravax.UUCP> Organization: SNA Solutions Inc., Contract Programming Group Lines: 160 In article <1606@oravax.UUCP>, daryl@oravax.UUCP (Steven Daryl McCullough) writes: > In article <607@ntpdvp1.UUCP>, kenp@ntpdvp1.UUCP (Ken Presting) writes: > > > (Daryl McCullough) writes: > > > > > >The only idea that I am attributing to Searle is that he "denies the > > >validity of using behavior as a test for intelligence". > > > > He does NOT deny this, and neither do I! > > Ken, you have been defending Searle for too long; you are starting to > confuse your own beliefs with his. The whole *point* of the Chinese > room argument was that behavioral tests for intelligence could *never* > be enough. He assumed at the outset that the Chinese room was able to > pass the Turing Test; that behaviorally it was indistinguishable from > a human being. I'M NOT DEFENDING SEARLE! Not only do I believe that it is possible for implementing the right program to constitute intelligence, I am confident we will figure out how to write it. I believe that the Systems Reply is mistaken, but that does not imply that I agree with any (other) point Searle is trying to make. More to the point, passing the Turing Test does not entail that the system which passes is "behaviorally indistinguishable from a human being". I agree that the linguistic behavior which is observed in the Turing test is crucial to intelligence. But there are certainly many other observable parts of behavior (such as perception and motion), which the Turing test ignores. Furthermore, the Turing test does not specify in any great detail how the questioning is to be conducted, which IMO is much more important than, eg, the duration of the test. Now, from Searle's argument it does follow that the Turing test is not adequate to verify the production of intelligence. That, however, is not the point of the CR. The "point" of an argument is surely its conclusion, and the conclusion of the CR argument is that implementing a program cannot constitute intelligence. > > Don't claim to speak for Searle unless you are willing to give > evidence: where, in anything that Searle has said, did he indicate > that he accepted using behavior as a test for intelligence? Where did > he give any indication that he believed in your Radical Translation > stuff? To me, these two positions seem diametrically opposed to what > Searle was saying. Searle mentions behavior-based ascription of intelligence in Minds, Brains, and Programs, under "The Combination Reply": If we could build a robot whose behavior was indistinguisahble over a large range from human behavior, we would [rationally] attribute intentionality to it, pending some reason not to. . . . But I really don't see that this is any help to the claims of Strong AI, and here's why: According to Strong AI, instantiating a formal program with the right input and output is a sufficient condition ... of intentionality. Searle, is *not* diametrically opposed to behavioral evidence for intelligence. He doesn't care about it (much). > > > (I happen to think the Turing Test is too simplistic, but the > > Radical Translation procedure advocate is every bit as behavioral). > > WHOEVER SAID THAT SEARLE BELIEVED IN THE RADICAL TRANSLATION > PROCEDURE??? > > The way you described it, radical translation is a special case of the > Turing Test. Any behavioral test you want to perform is perfectly > permissable in the Turing Test. If by performing the radical > translation one can determine that the entity behind the curtain is a > thinking being, then said being passes the Turing test. As Dave Chalmers pointed out, Searle is opposed (at least) to the use of radical translation as an exhaustive technique for determining the meaning of words, because it leads to indeterminacy. Intrinsic intentionality would avoid such indeterminacy, and Searle explicitly advocates intrinsic intentions. This is splitting hairs, but there is a difference between determining *what a word means* and determining *whether it is meaningful*. Those who disavow indeterminacy in the first case need not be opposed to it in the second case. (since no claim regarding specific meaning need be made) The CR depends only on meaningfulness, as I read it. I tend to agree that radical translation *could* be part of the Turing Test, but not as part of the usual scenario of blind Q & A. Radical translation depends on (eg) knowing about the environment of the questionee, and comparing its description of its suroundings to one's own observations. It may be possible to treat the teletype interface itself as a "shared environment", and ask questions like "How many seconds did it take to type this question?". (Note that no "mere TM implementation" could answer such a question, because TM's have no clock. Note especially that reading a clock is not just "symbol manipulation".) My main objection to Turing's proposal is that it needs to be sharpened up quite a bit. Radical translation is a specific method, which could be applied in some interpretations of what Turing said. I claim that no other behavior based method would give much support at all to a claim of intelligence. As I understand the Turing test, it does not *require* radical translation, and therefore is too lenient. > > > Searle assumes "Minds have semantics" and "Programs > > can't generate semantics". It follows immediately that "Programs can't > > generate minds." This has nothing to do with the question of how you decide > > whether a specific system has a mind, or any semantics. > > Ken, I think you are distorting Searle's already dubious argument. I > thought the Chinese room argument was supposed to *prove* that > "programs can't generate semantics". If he assumed it at the start, > then what was his argument supposed to show? I just combined Searle's Axioms 1 and 3 from Sci. Am. p.27 (these are verbatim quotes): Axiom 1: Computer programs are formal (syntactic) Axiom 2: Human minds have mental content (semantics) Axiom 3: Syntax by itself is neither constitutive of nor sufficient for semantics. Conclusion 1: Programs are neither constitutive of nor sufficient for minds. Anybody who claims that Searle is making a trivial logical error should look again at this argument. It is a very simple syllogism, which has nothing to do with consciousness, collecting behavioral evidence, or anything other than programs and semantics. I think Axiom 1 is false, and I can see how some people would object to Axiom 2 (it is rather vague). If I follow you, Daryl, you would say that even if the argument were OK, the conclusion is irrelevant, because a system consists of interface documentation as well as programmed hardware. The problem is that I'm not sure you actually disagree with Searle - both you and he insist that something besides the program be added to the hardware. > > Anyway, the Chinese room argument certainly *was* about "how you > decide whether a specific system has a mind". The system in question > was the Chinese room itself. Does it have a mind separate from > Searle's or not? Searle claimed not, but my question is, on what > basis? On the basis that it contained no semantics. If you accept Axiom 2, then that's all he needs. One of the things that intrigues me personally about the CR is that it *avoids* discussing any particular evidence-gathering procedure for deciding the presence of thought. It is an _a priori_ objection to an _a priori_ claim - "Running this program guarantees thought". The Turing test gets mentioned but is never the object of much attention in Searle's articles. What it's all about is "What does a program say about the system that implements it?" According to Searle, the program has nothing to say. You seem to agree that programs have an "operational semantics", but I think that topic also needs some additional detail. Since semantics is a relation between words and things, the program must specify just such a relationship. But that's another thread. Ken Presting ("And I don't mean Speech Acts")