fransvo@htsa.uucp (Frans van Otten) (03/02/89)
1. Confusion about the Chinese Room Argument. Noone ever told me or wrote what the Chinese Room Argument is exactly about. So I had to make something up from the various postings. First I understood this: --> Searle, a human being knowing the English language, does not know the Chinese language or its graphical representation (the characters). This man is in a room from which he can read Chinese characters from and write Chinese characters to the "external world". Now, if there are (English) books on the Chinese language and characters, Searle would be able to translate the incoming characters to English. Then he could understand the message, and think of a reply. He formulates his reply in English, and translates it to Chinese. Finally he outputs this, represented by Chinese characters, to his pen-pal outside the room. However, after reading some more articles in this newsgroup, I started getting confused. From these postings I understood the following (essentially different) meaning of the Chinese Room Argument: --> Th same setting as above, except for the books in the room. These don't contain rules for translating Chinese characters to English, but merely "when you get these-and-these characters as input, you should produce these-and-these characters as output". This seems a highly unlikable meaning. If all the output depends on the input and a set of written rules, (almost) any computer can take the place of Searle, leaving the entire Argument without any meaning whatsoever. So I must assume that my original understanding was correct. 2. About "understanding" in general and "understanding a language". Th arguments brought up in this discussion made me wonder. E.g.: What is the essential difference between a simulation and an implementation ? Can an implementation not be viewed as a simulation by the physical world ? And: What is the essential difference between the "Teletype Turing Test" and the "Total Turing Test" ? What makes sensor input and motor output essentially different from character (ASCII) I/O ? Also, on the outside (result side) of human beings, you can not tell whether they "understand" (whatever that would be) or not. It is enough if we can pass the Turing Test. Then the question emerges: Why should we, human beings, have developed "understanding" in the course of our evolution if it can not be a selection criterion ? Viewing the human brain as a collection of simple elements (neurons, molecules, atoms, quarks, name it), each of which obeys the rules of physics and chemistry, every state of it and each process in it must be understandable and (exactly) explainable (definable). So it also must be possible to formulate a definition of what "understanding" is about. After some thinking I reached this theory: --> The process or state which we call "understanding" is simply the representation of the understood concept in some "internal symbols" (some internal state of neurons/groups of neurons/transistors/logical gates/...). Now let us see what a (natural) language is. It is a (complex) code, designed to transmit messages (concepts/...) with. A language consists of a vocabulary (a set of "words"). A word is a symbol with some aspects: - It represents a (class of) meanings; - It belongs to one or more syntactical classes (verbs, nouns, ...); - It has a "vocal" (and mostly also a graphical) representation (i.e., how to pronounce (write) the word). A language also needs a "grammar". This is a set of rules, defining things like: - The ordering of words in a sentence; - The modifications to the meanings and representations of words in different syntactical classes and subclasses (or rather, a specification of the contextual meaning of the word insofar as the meaning of the word depends on its syntactical state). Then, "understanding" a language means: "Having internalized (most of) the grammar rules and (a substantial part of) the vocabulary that constitute that language". 3. What is understood in the Chinese Room. Inside the Chinese Room, the person of Searle understands the English language. The Chinese language is also understood; this understanding is due to the fact that the Room has internalized the grammar rules and the words of the Chinese language. The "internal symbols" of the Room are English words. So it is not Searle who is doing the understanding of the Chinese language: it is the Room as a system. As the internal symbols of the Room are English words, a code which is understood by Searle, he can understand the message that was codified in Chinese. He can think of a reply, formulate it in English, which can be translated into Chinese, which can be output using Chinese characters. 4. My point. Don't confuse "understanding a language" with "understanding a message (coded in this language)" ! The Turing Test (both types) tests on the content of the message, NOT on the representation of it ! I believe many posters don't have this difference clear; this misunderstanding is responsible for quite some megabytes of articles. The confusion is very understandable, though. Most of us think of "understanding" as "the human way to understand". With this I mean: The internal symbols/states are required to be "human brain symbols/states" (i.e. neural states). This causes the feeling of "implausity of the systems theory" Searle has. The meaning of "understanding" as I have defined it is neccesary for AI you want a computer to understand. It also means the problem is solved, for each computer program has its internal state and symbols, meaning that it understands the issue that is assigned to it. 5. What next. This gives rise to some more interesting issues: What to do with the concept that is understood ? You would need some function which generates conclusions. How would this function be defined, what would be the design rules for this function ? Does this function change during the lifetime of the system ? Would this be what learning is all about ? What would be the rules for changing the "conclusion-function" ? Would system-goals be incorporated into the above (like "surviving" is a system goal of human beings, in the sense of the individual and in the sense of the species; these two may very well generate contradictionary conclusion-functions, how to deal with this ?) ? Other interesting (though probably not important) related issues are: (self-)awareness, consciousness. By the way, the ideas I explained above fit remarkably well into certain models of the "psyche" which emerged from psychological and psychotherapeutical research; specifically, models using keywords like "basic self", "functional self" and "subpersonalities". If there is any interest, I might see if I can explain these models and how my ideas fit into them. -- Frans van Otten Algemene Hogeschool Amsterdam Technische en Maritieme Faculteit fransvo@htsa.uucp