Chinese Room

(File Last Modified Tues, Jan 28, 2003.)

• Summary of ``the Chinese Room''

Summary of ``the Chinese Room''

The ``Chinese Room'' is a theoretical experiment proposed by John Searle in 1980 as a refutation of Turing's Turing test for artificial intelligence. The experiment is as the following: you don't understand Chinese at all. You are in a room filled with unlimited Chinese Dictionaries and rule books which have Chinese answers for ANY questions in Chinese. People outside the room send in written questions in Chinese and you are supposed to answer them. Since you can find all the answer by looking up the shape of the characters in the question and match them in your unlimited resources of Chinese answers in your room, you are able to DRAW the answer onto a piece of a paper and send it out. Now, from the outside, people have reasons to believe that a perfectly knowledgeable Chinese Mr. know-it-all is sitting in the room because they can get answer for any questions. But in fact, you don't know Chinese at all. Under such circumstances, should we still say that you UNDERSTAND Chinese?

This is to be a refutation of the Turing Test because according to Turing, you are entitled as ``UNDERSTAND Chinese'' since your observed behavior is indistinguishable from a real Chinese answerer.

Searle also proposed two counter argument for the experiment and responded to them( - Searle's Minds, Brains, and Programs (1980)): the Systems Reply and the Robot Reply.

The ``System Reply'' argues that it's not the person who should be judged here, it's the system which is comprised with the person AND all the reference materials in the room, to be judged as an entity. Thus, we can say that the 'system' does UNDERSTAND. Searle's reply is that if the person memorize all the rules and data and do the lookup in his head, he can still give the answers to questions that he has absolute no idea of. He still doesn't qualify as ``understand'' Chinese.

The ``Robot Reply'' sets another scenario as we put all necessary knowledge into a robot and the robot can do fairly amount of human-like movements and action. When we ask the Robot to do grocery shopping. The robot may even hold a conversation to find out what we want, and then goes out and buys the stuff. Thus, the robot looks like it understands our commands, but does he know the meaning of grocery shopping as when we tell ourselves to do it?

Not only the Chinese Room experiment is debated by many researchers, but those two replies as well. For example, Jack Copeland (Senior Lecturer in philosophy and logic at the University of Canterbury, New Zealand) finds Searle's Reply inadequate for several reasons. (- Artificial Intelligence: A Philosophical Introduction (1993)).

I'm not sure if I find the original paper for the Chinese Room by John Searle, but I do find many reviews on it very useful. e.g. ``An Overview of The Chinese Room'' (http://www.ptproject.ilstu.edu/chinovrv.htm), ``JOHN SEARLE'S CHINESE ROOM ARGUMENT'' (http://www-formal.stanford.edu/jmc/chinese.html) etc.

Here's some of my thoughts after the reading. Not quite well organized yet, but some fragments to be pondered on:

1. Turing Test was proposed 42 years ago when no computers of the described specification existed. Today we already have several programs that can pass the Turing test but unfortunately we still can't say that we've found the solution for AI, as the 'Eliza's definitely do NOT qualify for A.I. I find the biggest discrepancy between the Turing Test and the Chinese Room is the definition of 'UNDERSTAND'. 'It walks like a duck, squacks like a duck', do we call it a duck? Or what if it even tastes like a duck? From a practicality point of view, we can say it's duck since it can satisfy us with anything a real duck can provide. But from a definition point of view, it's an imitation. I think this is part of the essence of the argument between Turing and Searle.

2. If Input-Output Imitation can't represent understanding, which overturns the legitimacy of Turing Tests by the way, what can? As a student, I believe we all have the experience in studying the text book just for passing a test. We can recite some of the typical answers in the test and sometimes do get a satisfying score. But we all know we don't really understand the subject. But from a teacher's point of view, we do. that is to say we have all experienced the Chinese Room somehow and in more than one occasion, it IS used to measure UNDERSTANDING already.

3. A question underlies is that what is the relationship between THINKING and UNDERSTANDING? Remember that Turing Test is to determine if the machine can THINK, while the Chinese Room judge if the person UNDERSTANDS. I take there's a big difference thus the two experiments cannot be used to refute each other symmetrically. A simple example, do we say that infant children can think? Well, no matter how annoying sometimes, since they are still human being, we do. But do they understand everything we say? Not exactly. Actually, most of the modern information systems in existence do ``understand'' in their specialty with databases, as the books in the Chinese Room, inside their storage. But still none of them can 'think'. I don't believe that the thinking process can evolve from any huge pile of data alone.

4. What is the essence of thinking? I'd say when an entity can process the information in a way not explicitly provoked by the questioner or master, there is thinking. Now suppose this, we have this know-it-all super machine. It can give appropriate answers to the questions we ask it. But when we are not demanding answers from it, it just sits there idling, well- maybe with a screen-saver. It is not a thinking machine. What if we re-program the system, build in a separate thread to logically link and summarize all the information it has and the thread is running constantly. Of course, the result of the summarization goes into its database as new knowledge. Then there is something revolutionary here: the machine has the ability of gaining new knowledge without human intervention. I'd say, that machine can THINK! If we have two identical robots A and B with equally complete database in biology and electronics, but we program B with the separate thread to 'THINK', what would you expect when you ask the two to do the same thing: ``Please turn your power OFF''? Hmmm, interesting...

5. Last point, maybe the organic THINK process of a human brain depends somewhat on a more 'random' or 'polymorphism' factors. And to simulate these factors are a definite weakness for current computers based on electron movements in man-made tunnels. Maybe a more suitable carrier can be found, like bio-chips, chemical-chip, quantum-chips... to perform intelligence. Well, I¡¯m not the expert, suit yourself with imagination on this one.

Thanks for all the patience. Maybe I should partition it into smaller segments. :)