Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!uwm.edu!bionet!ames!amdahl!kp From: kp@uts.amdahl.com (Ken Presting) Newsgroups: comp.ai Subject: Re: more Chinese Room Summary: Descartes, and how compilers contain semantics Keywords: Searle, Chinese Room, Descartes, Cartesian Message-ID: <0czk02el8b1301@amdahl.uts.amdahl.com> Date: 23 Feb 90 20:10:41 GMT References: <1990Feb13.225830.13432@wam.umd.edu> <53tT02R288oV01@amdahl.uts.amdahl.com> <1990Feb16.182519.18166@wam.umd.edu> <1990Feb16.220511.27647@Neon.Stanford.EDU> <857@wrs.wrs.com> Reply-To: kp@amdahl.uts.amdahl.com (Ken Presting) Distribution: usa Organization: Amdahl Corporation, Sunnyvale CA Lines: 91 In article <857@wrs.wrs.com> hwajin@wrs.wrs.com () writes: >The idea that one can definitely identify a corresponding >physical parts of anything that is non-physical seems to >be also based on the same old Cartesian model of world view; >everything has to be clearly categorized, labeled and analysed >to the littlest components. There is always a clear distinction >between things material and things spiritual. Let's lay off Descartes, OK? It was Aristotle who invented Categories, and Kant who is responsible for the (small) role they still have in current philosophy. Descartes was *opposed* to this view. Descartes' method involved pure deductions (read "calculations") from assertions such as "I'm Thinking", and had very little to do with objects of any sort (though he did draw the famous distinction, which I will not repeat, because I don't like it either). Note that Searle accuses the Strong AI'ers of Cartesianism. He is *right* that those who suppose pure calculation can constitute a mind have a position resembling Descartes. He is *wrong* to suppose that anyone holds such a position. For the purposes of the debate over the potential of AI, Descartes' most important contribution is the invention of analytic geometry. "That's completely irrelevant" you say? Quite right. >As I've mentioned several weeks back and some others more >recently, Searle's entire argument is based on the premise that >there are in fact two distinct components -- syntax and semantics >which are naturally separate and cannot be reconciled. This >assumption is taken without any investigation. I dare you to >find any attempt by Searle, in many of his redundant articles >on this tiresome Chinese Room nonsense, to re-evaluate this >basic condition of his argument -- Scientfic American, Jan. 1990, page 31, leftmost column. is it truely appropriate >for us to insist on this division between syntax and semantics >in our search for the understanding of the mind? Everything >is connected at a certain level. One cannot safely >assume any division of any abstractions. Quantum physists >certainly feel better at ease with this lack of >fine lines -- as in, "it seems to be both wave and particle at >the same time." > >The mind seems to be both syntax and semantics at the same time. Searle is not saying that syntactic information is ever encoded separately from semantic information in any natural or artificial language understanding system. He is *only* saying: IF you write a program that looks at input strings and nothing else, THEN that program will not contain enough information to figure out what the strings mean. Your objections actually *strengthen* Searle's position. He believes, as you do, that semantic and syntactic information are completely mixed in human brains. What he does not believe is that computers ever do anything other than look at strings. He concludes from this that computer programs can never contain any semantic information. Searle's problem is that he does not understand what compilers do. To put it better, he does not understand why they are useful. Compilers transform strings of symbols into other strings. Compilers contain both syntactic and semantic information about programming languages. Now suppose that you know lots about micrprocessors, but nothing about software. Would you be able to figure out the semantics of a programming language just by looking at the compiler? Well, if you knew the instruction set for the target processor, it would be tedious, but you could figure out that function references mean "push the stack and jump" and eventually you'd completely understand the source language. If you didn't know the instruction set, but you could watch the object programs run on a real computer, then it would be even more tedious to figure out the semantics of the programming langauge, but you could still do it eventually. But if all you had was the compiler code itself (let's suppose that it's a cross-compiler), and samples of source and object files, you'd be hard pressed to figure out some of the details like number representation formats and interrupt bit masks. In general, all the implementation- dependent headaches that are so carefully spelled out in compiler manuals are the places where it's not enough to know which symbols get changed how, you have to know how the program interacts with the real live hardware. So Searle is wrong that programs cannot contain semantic information, but he is right that if the occupant of the Chinese Room can only look at the program and not at what's goig on outside, he may never figure out the language he's "speaking".