ARTIFICIAL INTELLIGENCE: THE CHINESE ROOM ARGUMENT


INTRODUCTION
The Chinese Room argument has been about for a few years now, both in print and as the subject of more than one TV analysis program. In one  "Horizon" programme it was presented without any recognition of the many valid criticisms which had been made of it. This essay is an attempt to put the record straight for the benefit of those who may not be aware of these criticisms.  John Searle, an American academic philosopher who is its author, claims that the idea came to him while he was flying to a conference on Artificial Intelligence. He presented his new idea at the conference to a startled audience and since then he has made a kind of career out of the argument (in various forms). In general he claims to have shown that the whole concept and foundation of AI is false. The argument, it has to be admitted, has a superficial appeal. What I want to do here is to present first the Chinese Room argument and then to explain why it is invalid. First, however, we must explain what is meant by the Turing Test.

THE TURING TEST
Intelligence, as a concept, defies precise definition but that does not mean we cannot characterise it (see the essay on Artificial Intelligence on this site). Intelligence is not simply an ability to do things, it is an ability to do the kind of things humans do, and in a way that (superficially at least) resembles the way humans do it. That has to be the way of it, because if that proviso was not included, we would have to credit computers with superior intelligence because they can certainly beat us a some tasks (such as arithmetical calculation). Recognising this, the mathematician Alan Turing devised a simple definition of intelligent behaviour. It is "an operational definition" which means that it is defined in terms of observed behaviour rather than intrinsic or internal properties. He envisaged a computer, or whatever system we are testing, sitting in one room and a panel of human judges in another. A wire carries communications between the two rooms.  If, after chatting with the unseen system, the pnel of judges are unable to tell if it is a machine or another human, then the machine passes the test and is to be declared "intelligent". The definition has the merit of being practical while ducking the problem of defining the internal mechanism involved. A number of quite simple computer systems have, on some notable and well documented occasions, "passed" the Turing Test. The have done so, however,  only for a very short time and/or only within a very limited context. To pass the test, with flying colours, would require an extended period of testing coupled with a fairly wide remit on the permissible content of the conversation.

COMMENT ON THE TURING TEST
The Turing Test is simply a formal version of something we do with respect to each other on a daily basis. The workings of another person's mind are hidden from us. We assume the operation of an internal mechanisms similar to our own because it is the simplest explanation of what we observe and perhaps because we have an instinctive tendency to do so. Without that assumption, the behaviour we observe in other people would be completely unpredictable. If someone speaks to us in our language we assume that they understand the words they speak and hear. We have, however, no direct evidence for making such an assumption. Our justification is that the assumption seems to work successfully.

SEARLE'S FOLKSY TERMINOLOGY
A word of caution. One of John Searle's most disarming characteristics is his use of "folksy" terminology. He likes to sum up the rationale of the Turing Test, amusingly (but patronisingly) with these words - "If it walks like a duck and talks like a duck then it is a duck."  In one sense that is a correct summary but one should beware of the hidden innuendo. It suggests that the Turing Test is based on superficial criteria and that is not necessarily the case. The Turing Test can be as searching and as rigorous as we care to make it. It would, perhaps, have been better if Searle had described it thus - "If it walks like a duck, talks like a duck, has feathers like a duck, can swim like a duck, lays eggs like a duck, and if these eggs hatch out and grow into other duck-like objects which walk talk ... etc just like a duck and if when you kill it and eat it tastes like a duck, then we are entitled to call it a duck."  Couched in these terms the substance of Searle characterisation has not changed but the implcation of superficiality has been removed. The question Searle which then has to answer is this - is there an alternative technique he would recommend for duck recognition?

SEARLE'S THE CHINESE ROOM ARGUMENT
And now to Searle's argument. Imagine a room - a kind of "port-a-cabin" perhaps, which contains two things - a man and a filing cabinet. The man is a perfectly ordinary man with one special ability. He can open the drawers of the cabinet, extract files, read the contents and act on instructions he finds there at the speed of light. The other remarkable thing about this room is the filing cabinet which has a capacity which is virtually infinite. These acknowledged impossibilities are introduced deliberately because the argument is one of principle and not one of physical constraints. It is Searle's contention that no matter how big and powerful a computer we may build at some time in the future it still won't be able to do the job.

The filing cabinet contains a special store of information - nothing less than the rules about how to speak Chinese. The rules are written on filing cards and in English so that the man can read and understand them, but the cards also have Chinese characters written on them which the man can recognise only by shape and by comparison. He does not know what they mean because he does not speak Chinese. There are two things about this experimental design which are worthy of comment. To be fair to the argument we will allow the removal of physical limitations because, like Searle, we are arguing about general principles and not about practical limitations. The second point is that Searle's assumption that the rules for processing the Chinese room are (in his folksy words) "just dumb grammatical rules"  This is an assumption of a different type altogether. It is an assumption which is both false and of fundamental importance. We should not be fooled into thinking otherwise by the fact that he has buried it among oher, innocuous assumptions about removing physical restraints and dismissed it with a few casual words. We shall, at a later point, elaborate on that point. But to continue ...

The room is in communication with the outside world by means of an ordinary letterbox. Through this letterbox we (who are outside the room) can pass messages written in Chinese. The man, who is inside takes each message,  scans it, scans though the contents of the filing cabinet and compares the symbols he finds there with the symbols in the message. He matches symbols, reads the instructions (in English), opens more drawers, reads more instructions, matches more symbols and, as he does so, he writes symbols on to a blank sheet of paper as instructed by the filing cabinet rules. He then posts this fresh sheet of paper back through the letterbox to the outside world. Those outside will find that the sheet of paper carries a message in Chinese which is an answer to the original message. In brief, the "room" is able to carry on a conversation in Chinese in a way indistinguishable from a fluent human speaker of Chinese. The room "speaks Chinese".  But Searle claims that even although the room speaks Chinese, theroom does not "understand" Chinese. his justification for saying that is that the man does not understand Chinese. Neither does the filing cabinet. It follows, Searle argues, that while the room has passed the "Turing Test" it has done so without anyone (or anything) in the room having an understanding of Chinese. From that, he concludes, that the Turing Test is invalid as a test of understanding. And if the Turing Test goes, the whole foundation of Artificial Intelligence goes with it.

THE COUNTER ARGUMENT
According to Searle, we know that the Chinese room does not have an understanding of Chinese because none of its constituent parts has an understanding of Chinese. As Searle says - "Well the man does not understand Chinese and how can a set of dumb rules understand anything?". An interesting question. We shall put on one side the way Searle has introduced the term "understanding" without explanation of its meaning (lack of definition of his terms, or of characterisation, as we have suggested, is one of the features of Searle's style). For the present we shall concentrate instead on the assumption implicit to this argument, that understanding is a property (like the mass of an object) which is possessed by virtue of all its constituent parts having a some of the same thing. This is a false assumption. 

It is ironic that the idea for the Chinese room came to Searle when he was sitting in an aeroplane, for an aeroplane is a tangible contradiction of his argument.. An aeroplane has "the power of flight". But if an aeroplane is taken apart into its constituent parts no single part will be found to have that power. Yet each part does contribute something towards the behaviour we call flight. It may be aerodynamic shape, or propulsive power, or structural integrity to hold those aerodynamic parts in the correct position, or it may be the energy for propulsion (the fuel). But none has, by itself, the full power of flight.

Intelligence and understanding are complex concepts and like "the power of flight" they describes the co-operative performance of several disparate properties. There is no requirement for the individual components of the Chinese room to be able to speak and understand Chinese. It is only necessary that the various components provide all the necessary ingredients of that kind of intelligent understanding.

That point is sufficient to refute Searle's argument. It shows that his argument is founded upon an invalid assumption and we could rest our case at this point. There are however, additional objections which can be raised..

UNDERSTANDING
Let us return to that thorny word "understanding". By substituting (unobtrusively) the word "understanding" for the word "intelligence" Searle has perhaps felt that he was moving on to safer ground.  The term "intelligent behaviour" carries no connotations of internal mechanisms. It is something cold and remote, something to be observed from outside (in line with the Turing Test).  The term "understanding", however, implies something internal. It is something we associate with a visceral, emotional feeling. It has to do with the recognition of implications and with our emotional reaction to these implications. When we say that someone "didn't fully understand what he had done" we mean that he did not fully appreciate the implications of what he had done. For this reason it is difficult for us to conceive of something inanimate having "understanding". So by using this term it would appear that Searle has conceded that he has lost the argument on "intelligence" and has moved to (what is for him) the higher andsafer ground of "understanding". So be it. We shall show that even here his argument is founded on logical sand.

THE TURING TEST AGAIN
The issue is now about whether a computer system, which in physical terms is a collection of wires and transistors, could ever have what we would describe as understanding. Searle argues that he has given us a system which demonstrably does not have understanding and yet it passes the Turing Test. We argue that a system does not have what we would call understanding and has only "dumb grammatical rules" to guide it, CANNOT pass the Turing Test (if that test is applied with rigour). Our claim then is that his Chinese room is a theoretical as well as a physical impossibility.

DUMB GRAMMATICAL RULES
Consider the way Searle characterises the rules in the filing cabinet as "dumb grammatical rules". And now consider the following sentence.

"The councillors refused to hire the hall to the students
                because THEY disapproved of people who advocated revolution."

Here the crucial word is "THEY".  To what does it refer?. Without knowing that we cannot understand the sentence (and this argument is all about understanding). A simple application of "dumb grammatical rules" would have us assume that the referent is the last collective noun mentioned. In this case the referent would appear to "the students". Yet we all know that the referent is really the councillors. It is (stereotypical) councillors surely who would disapprove of people advocating revolution. It is (stereotypical) hot-headed students who would advocate revolution. Or is it? If this sentence was describing something which had happened in Communist China during the reign of Mao ts' Tung, the situation would be totally reversed. There and at that time, it was stereotypical councillors who advocated revolution and modernist/revisionist students who might (rebelliously) disapprove of revolution. To get a proper understanding of that sentence we need more than a few "dumb grammatical rules". We need a very dee understanding of the world, of people, of their attitudes and of the history of other times and other places. This is what we call  "deep knowledge" which Searle has excluded from the Chinese room. And yet he claims the room can pass the Turing Test. He is wrong. It cannot and Searle has founded his argument on a logical contradiction. There is a well known theorem in logic which shows that if you include even one contradiction in a logical argument then you can, with otherwise flawless logic, appear to "prove" absolutely anything.

PHYSICAL MATERIAL AND INFORMATION STRUCTURES
But there is more. With the Chinese room Searle has created a physical analogue of a computer system (the man is the CPU, the letterbox is the input/output mechanism and the filing cabinet is the ROM). The everyday ordinariness of this analogue and of the materials from which it is constructed, makes the idea of such a system having understanding seem absurd. And yet if we adopt that position we have assumed, as an initial premise, the proposition we are trying to contest. The argument is cyclic. If a paper and metal filing cabinet (and a man) cannot have understanding how can wires and transistors? How for that matter can cell membranes, axons, RNA molecules etc (ie: the constituent materials of the human central nervous system) have "understanding". Excise a cell membrane from your brain (any one of them) and look at it under the microscope. It has no understanding. How can it? It is just a dumb blob of cytoplasmic jelly surrounded by a lipid cell membrane. We can analyse all the constituent parts in a tes tube and they are all just "dumb" chemicals. It would seem then that by applying a Searle-style argument we have "proved" that human beings cannot have understanding.

The error is clear.  By introducing those homely materials Searle has, like a stage conjuror, focused our attention on the wrong thing.  The question is not whether the materials can have understanding. Understanding is not a property of physical materials. It is a property of the information structures which may be stored on or within those materials. It does not matter what those materials are, so long as they are capable of storing the necessary information in the appropriate structures and of translating them into physical reactions. An electric pulse travelling the length of an axon in the brain is very much the same as an electric pulse travelling in a wire. The conduction of a pulse through a synapse is physically very much like a pulse crossing a transistor junction. What is important is the interpretation placed upon those pulses and its consequences.  Pythagoras was supposed to have drawn his famous mathematical theorem on sand. How (as Searle might have put it)  can dumb sand know anything about mthematics? The sand, like those wires, brain cells and pieces of card in the filing cabinet, is simply the medium which carries the information. It is the information itself and its structure towards which we should direct our attention.

EMOTIONAL REACTIONS
Of course, in the Chinese room Searle has been careful to exclude anything which can feel emotion (or so he imagines). The man in the room does have the ability to feel emotions but we have been told pointedly that he does not understand Chinese and so he has no way of making a connection between what he is doing and his own emotional mechanisms. But consider this - How, in the ordinary world, do we know when another person understands something? How do we know when they are feeling emotion? The answer has to be that we observe their behaviour. Without being aware of what we are doing, we apply the Turing Test. And, as mentioned above, we can apply that test as rigorously as we want. Suppose then we were to pass a message into the Chinese room which says (in Chinese)

"If you don't get out of that room in 30 seconds I am going to set fire to it"

What is now required, if the room is to pass the test, is for something much more dramatic than a simply response written in Chinese. We require the man (and/or the filing cabinet if it has motorised wheels) to emerge from the room is a state of some confusion. If, on some far flung planet in space we were to encounter a strange room-shaped object which behaved like that, I submit we would have little hesitation in deciding that it had an intelligent understanding of Chinese. And that we would be justified in doing so.

CONCLUSION
Could we build a room like that? Could we ever devise a set of rules which could provide the system with that kind of response. I don't know. But then I am not trying to prove that full blown artificial intelligent understanding is a possibility. I happen to believe it is, but I don't pretend that I can prove it. What I have shown, however, is that Searle's argument is invalid.




Copyright Hugh Noble April 1999.