marty@althea.UUCP (Martin B. Brilliant) (02/01/90)
In article <1990Jan27.004920.28355@agate.berkeley.edu> johnb@sandstorm.Berkeley.EDU (John L. Bergquist) writes: > >So what happens when you have the person in the chinese room internalize the >rules and the books? Meaning, suppose you have someone who had a photographic >memory and memorized all the chinese input symbols and their corresponding >output symbols? Imagine you are that person. Now the whole system is in your >own mind; in essence you ARE the system. Do you understand Chinese? You would >probably feel you do not, even though you could respond intelligently in >Chinese to any question that was asked. I think we are getting to the heart of the matter. Let me first dispose of some of Searle's nonsensical questions. He described a SYSTEM consisting of a person, some rulebooks, etc. The he asked whether the PERSON understood Chinese. No, the SYSTEM that might or might not understand Chinese is exactly the SYSTEM he described, not any part of it, such as the person, or the books, or the room. To insist that a part of the system must contain the properties of the whole is to create a homunculus argument. The SYSTEM is assembled from parts, and it is the fact of assembling it that gives it properties that the parts alone do not possess. That's probably what the term "imposed order" refers to. Now let's get to Searle's other question. Does success in the Turing test imply thinking? No, of course not. Turing's suggestion, as I understand it, was that his test was a necessary, not sufficient, condition. If it can't pass the test, it's not thinking. Whether it is thinking, depends on how long you keep on testing and how skillful the tester is. I think, however, that Searle and others underestimate the ingenuity of human testers. No stupid rulebook parser could pass. Marty M. B. Brilliant (201)946-8147 marty@althea.UUCP -- Marty M. B. Brilliant (201)946-8147 marty@althea.UUCP
cliff@delta.eecs.nwu.edu (Cliff Chaput) (02/04/90)
In article <466@althea.UUCP> marty@althea.UUCP (Martin B. Brilliant) writes: >No stupid rulebook parser could pass. > > Marty Ah! But according to Searle's statement of the criteria, the rules of the book are designed to be *indistinguishable* from a native Chinese speaker! This means that every response you would expect from a Chinese person you could get from this book. That's some book, if you ask me. So if you're looking for intelligence in the Chinese room, I'd examine the book. Given the current definition of the problem, they guy in the room doesn't know chinese, the symbols don't know chinese, but the BOOK DOES! Otherwise, how else could it be indistinguishable from a native Chinese speaker? This explains the anomalous "memorizing the book" problem: if the book knows Chinese, then a man memorizing a book does nothing more than change the medium. The information remains the same. I realize I'm not being that clear, be let me give an analogous example (god save me!). There's a program out there called Mathematica, which does an awful lot of mathematics from top to bottom (Integrations, Equation solving, all sorts of neat stuff). The program includes most of the known rules of mathematics. Is it safe to say that this program "knows" mathematics? I would think so. I can't expect any other person who knows mathematics to do much more that Mathematica. Now, this is not saying that people won't do more; some are better at math insights, some are better at passing math tests, some are better at showing their work. But I would claim that these are not central to the knowledge of mathematics proper. Even if we want to reduce it some and ask, "Does Mathematica know algebra?" one would surely be forced to answer positively. Now can I say that my NeXT knows mathematics? Kinda, but not in the same way. I know that through the computer I can access the mathematical information that Mathematica provides, but I am also aware that Mathematica is but a part of the NeXT. I cannot claim that the Interface Builder knows mathematics. Nor can I claim the same for Webster's Dictionary, WriteNow, etc. Now say I got a source listing of Mathematica. Can I now say that this source listing knows mathematics? Again, what is it to know mathematics? Will it perform the mathematical feats that a human would? Suppose you took ascii codes for the string "D[x^3,x]" and fed it to the input routines of the program. You could trace from there the operations that took place and followed it all the way to the output routines which gave you "3 x^2". Clearly this source listing has the same information the the running program has, or even a human (in this particular area). Okay. Now we take some guy with a remarkable memory, know-how of computer programming, and no knowledge of ascii codes. He memorizes the source code for Mathematica. Now, you take the string "D[x^3,x]" and convert it into ascii codes by hand, and tell this mnematic wonder to feed this code into the input routines of his memorized mathematics source, trace the program through completion, and report the code that is given by the output routine. Will this code, when translated back into text, be "3 x^2"? Most certainly! Though we can not say with any reassurance that the man knows mathematics. Even if he did, he wouldn't know to use it because he is being told ascii numbers which, as stated above, have no meaning to him. Why do we run up against this wall? We commonly misplace the first person. When we talk about ourselves, what do we mean? If I were to say, "I am injured," does this mean that my entire body injured? My vocal cords are certainly not injured, my mind seems to be relatively intact. No, certainly I must mean that a part of me is injured. If I say, "I am happy," it is not reasonable to say that my knee is happy, my eyes are happy, or my hair is happy, though it is clear that these are intrinsically parts of my whole. So when I say, "I know mathematics," which part of me do I mean then? Well, of course, I mean my brain. "My brain knows mathematics," would then be a sentence with identical meaning to "I know mathematics." Right? Well, what happens if I say, "My brain knows how to pump blood through my body," or "My brain knows how to replicate human life," or "My brain knows how to block out pain if it gets too intense." These are all true statements, but are they really the same as "I know how to pump blood through my body, replicate human life, and block out pain."? There seems to be a level of understanding that we are missing. Surely our brain can accomplish these tasks, but do we know how it's done? These are usually referred to as unconscious activities. So it is possible to say that a person knows how to do something without being conscious of the mechanism involved. As a result, it is possible that a person could know mathematics without being aware that he knows mathematics, or, even if he is aware, to understand the mathematics. Continuing on the same pattern, any body of knowledge could be said to have the same property, including Chinese. I hope this sheds a light or two on what I'm trying to get at. We should be careful when we say the computer knows Chinese, or the room knows Chinese. And we should understand that storing knowledge is not the same as knowing. So Searle hasn't done much but to point out this false assumption through contradiction. However, the knowledge is still there, no matter how it is accessed. Just because we are not conscious of the methods by which blood is pumped through our body does not imply that this doesn't happen. If WriteNow doesn't know mathematics, that doesn't mean a NeXT cannot add two and two. And if we do not know Chinese, we may very well be able to respond like a native Chinese speaker given the right information. Cliff Chaput Mneme Project, Northwestern University Psychology Dept. cliff@mneme.psych.nwu.edu, cliff@eecs.nwu.edu
gilham@Neon.Stanford.EDU (Fred Gilham) (02/06/90)
In article <3488@accuvax.nwu.edu> cliff@delta.eecs.nwu.edu (Cliff Chaput) writes: > >Ah! But according to Searle's statement of the criteria, the rules of the >book are designed to be *indistinguishable* from a native Chinese speaker! >This means that every response you would expect from a Chinese person you >could get from this book. That's some book, if you ask me. > >So if you're looking for intelligence in the Chinese room, I'd examine the >book. Given the current definition of the problem, they guy in the room >doesn't know chinese, the symbols don't know chinese, but the BOOK DOES! >Otherwise, how else could it be indistinguishable from a native Chinese >speaker? This explains the anomalous "memorizing the book" problem: if the >book knows Chinese, then a man memorizing a book does nothing more than >change the medium. The information remains the same. > >I realize I'm not being that clear, be let me give an analogous example >(god save me!). There's a program out there called Mathematica, which does >an awful lot of mathematics from top to bottom (Integrations, Equation >solving, all sorts of neat stuff). The program includes most of the known >rules of mathematics. Is it safe to say that this program "knows" >mathematics? I would think so. I can't expect any other person who knows >mathematics to do much more that Mathematica. Now, this is not saying that >people won't do more; some are better at math insights, some are better at >passing math tests, some are better at showing their work. But I would >claim that these are not central to the knowledge of mathematics proper. >Even if we want to reduce it some and ask, "Does Mathematica know algebra?" >one would surely be forced to answer positively. > No! Mathematica does not know algebra or mathematics. To see why this is so, we just have to look at the human analogy of Mathematica, namely "cookbook" math. Mathematica is a sophisticated system for doing cookbook math. Many people can do calculus from the integral tables without really knowing calculus. I remember taking a class where I was taught just that. Someone who knows calculus can go outside the cookbook. Such a person can apply his knowledge to problems that are not covered by any of the rules he already knows. Besides this, such a person can see applications for the rules that are not immediately obvious. To use a simpler example, take long division. Most of us know the standard algorithm for long division. I taught it to kids in elementary school. But I could tell that some of these kids were following the rules without knowing why they worked, or really, what the use of it all was. Yet they could do the algorithm reliably. I would say that they did not know division, even though they could execute the steps of the algorithm so as to produce the result. If I had asked them to think of some other algorithm for doing long division, they would have looked at me as if I were crazy. They probably would have thought "This IS long division." The "book", then, contains the steps of the algorithm. It is at least two steps removed from knowledge, in that something needs to execute the algorithm and some mind needs to interpret the results.. The whole trick behind the Chinese Room argument is that it is possible to handle symbols in two ways, mechanically and semantically. The Chinese Room argument claims that a mind can assign meaning to symbols that are produced by an entity to which the symbols have no meaning. Making the entity very complicated does not change this. -Fred Gilham gilham@csl.sri.com
cliff@delta.eecs.nwu.edu (Cliff Chaput) (02/07/90)
In article <1990Feb5.193530.13545@Neon.Stanford.EDU> gilham@Neon.Stanford.EDU (Fred Gilham) writes: >No! Mathematica does not know algebra or mathematics. To see why >this is so, we just have to look at the human analogy of Mathematica, >namely "cookbook" math. Mathematica is a sophisticated system for >doing cookbook math. Many people can do calculus from the integral >tables without really knowing calculus. I remember taking a class >where I was taught just that. > >Someone who knows calculus can go outside the cookbook. Such a person >can apply his knowledge to problems that are not covered by any of the >rules he already knows. Besides this, such a person can see >applications for the rules that are not immediately obvious. > >To use a simpler example, take long division. Most of us know the >standard algorithm for long division. I taught it to kids in >elementary school. But I could tell that some of these kids were >following the rules without knowing why they worked, or really, what >the use of it all was. Yet they could do the algorithm reliably. I >would say that they did not know division, even though they could >execute the steps of the algorithm so as to produce the result. If >I had asked them to think of some other algorithm for doing long >division, they would have looked at me as if I were crazy. They >probably would have thought "This IS long division." > >The "book", then, contains the steps of the algorithm. It is at least >two steps removed from knowledge, in that something needs to execute >the algorithm and some mind needs to interpret the results.. The >whole trick behind the Chinese Room argument is that it is possible to >handle symbols in two ways, mechanically and semantically. The >Chinese Room argument claims that a mind can assign meaning to symbols >that are produced by an entity to which the symbols have no meaning. >Making the entity very complicated does not change this. > >-Fred Gilham gilham@csl.sri.com Does one have to be aware of a fact to know it? It is clear that we know how to breathe. We are rarely aware of our breathing (though we would surely be aware if we stopped). We know how to talk, to speak in coherent sentences. But do we know the rules by which thought is produced? Planets circle the Sun according to the laws of physics. But do the planets know physics? The fact is that knowledge can exist without human awareness. Even if all people were to die, the Earth wouldn't stop revolving about the Sun. Even if you remove human knowledge from the human, the knowledge is still there. This is how we can store knowledge in history books, encyclopedias, and Mathematica programs. As for "a mind interpreting the results," this is in no way part of knowing "math." It is part of knowing how to interpret results. The two are very separable. Knowing how to interpret results, escaping the system, or "consciousness" is an act we can apply to any intellectual discipline. But that doesn't make it an intrinsic part of those areas, including math. Once again, it is possible to know something and not be aware of it, thus being unable to interpret it. This is something that Sartre called "transcendence." We can reflect on ideas that we are aware of, but most ideas we have are our "axioms," things we take for granted. So for a computer to do math as well as a human, yes, it will need "consciousness" of the math, or an ability to escape the system. But this is not the issue. This fact is already pre-supposed by the Chinese Room puzzle, stating that there is a book which can give responses just like a native Chinese speaker. This would require awareness. I feel, though, that my analogy still stands. Awareness is not a requirement of knowing. So knowing math doesn't imply being able to escape the system. Cliff Chaput Mneme Project -- Northwestern University Psychology Dept. cliff@mneme.psych.nwu.edu, cliff@eecs.nwu.edu
weeks@ssbell.IMD.Sterling.COM (John Weeks) (02/07/90)
In article <3567@accuvax.nwu.edu> cliff@delta.eecs.nwu.edu (Cliff Chaput) writes: >Does one have to be aware of a fact to know it? It is clear that we know >how to breathe. We are rarely aware of our breathing (though we would >surely be aware if we stopped). We know how to talk, to speak in coherent >sentences. But do we know the rules by which thought is produced? Planets >circle the Sun according to the laws of physics. But do the planets know >physics? > >The fact is that knowledge can exist without human awareness. Even if all >people were to die, the Earth wouldn't stop revolving about the Sun. Even >if you remove human knowledge from the human, the knowledge is still there. >This is how we can store knowledge in history books, encyclopedias, and >Mathematica programs. I guess I've got to get my 2 cents in here. I think that throughout this Chinese Room thread there has been a systematic ambiguity in the use of the terms "to know" and "knowledge." There are at least three uses involved, two major and one minor use. The minor one is of knowledge in the sense of a body of facts or wisdom. I don't think that this sense has much relevance to the CR. The other two senses have been distinguished at least since Plato: techne and episteme, techne being an ability - knowing how to breathe or how to ride a bicycle, episteme refering to a propositional attitude - knowing *that* such and such is the case. (This distinction is preserved in some languages other than English: kennen and wissen in German, for example.) There are arguments for saying that the Chineses Room knows, in the sense of having the ability to "speak", Chinese although these arguments would not support the thesis that the Chinese Room has any propositional attitudes at all (knowing, believing, doubting ...) -- John Weeks Phone: (402) 291-8300 Sterling Software FSG/IMD e-mail: uunet!btni!ugn!ssbell!weeks 1404 Ft. Crook Rd. South Bellevue, NE. 68005-2969 FAX: (402) 291-4362