On Searle’s Chinese Room Argument:
A Defense of the Systems Reply
Can computers think? This question has proved to be very divisive. On the one hand there are many in the field of Artificial Intelligence, which attempts to create machines that mimic a variety of human behaviors, who think they can. On the other hand there are many in the Philosophy of Mind, which aims to answer philosophical questions about what the mind is and how it relates to the physical world, who think they cannot. One of the most famous arguments that computers cannot think is John Searle’s Chinese Room Argument, which I will explain below. Though it may seem compelling at first sight, I will show that it cannot do the work Searle wants it to do.
2. The Argument Stated
John Searle offers the following argument—which is often called the Chinese Room Argument—against the idea that a digital computer could come to understand a language simply in virtue of implementing a computer program (Searle p. 279). Suppose there is a room, with you locked inside it. The room has an input slot and an output slot, and contains baskets of Chinese characters. In addition to these there are rule-books which specify which characters to put in the output slot in response to certain characters that come in through the input slot. The rules are an analogue of a computer program, and you are the analogue of the computer’s central processor. The rules mention only the shape and order of the characters—their syntax. Because of this Searle argues that you cannot come to understand what the characters mean—their semantics—by manipulating them according to such rules. Consequently, no digital computer could come to understand a language simply by manipulating symbols according to formal rules—rules which only concern the syntax of the symbols. And yet, to all outward appearances, you behave just as though you understand Chinese. Thus something can behave exactly as it would if it understood Chinese, and yet not understand a word of it. Searle thinks the conclusion generalizes: Syntax is not sufficient for semantics.
In order to get a clearer conception of how this argument is supposed to work, I will make it a bit more formal. Let us call the following principle the Indistinguishablity Principle (IP, for short):
IP: If x’s behavioral dispositions concerning some language or linguistic expression are indistinguishable from y’s, and y understands that language or linguistic expression, then x also understands that language or linguistic expression.
It is this principle, or something very close to it, that Searle seems intent on refuting. If we accept that IP is Searle’s target, the quasi-formalized Chinese Room Argument would go like this:
- If IP is true, then the person in the room understands Chinese.
- The person in the room does not understand Chinese.
- Hence, IP is not true.
3. The Argument Criticized
This reconstruction of Searle’s argument is clearly valid: if its premises are true, the conclusion must also be true. The problem is that the first premise is false, and because of that we cannot say whether the conclusion is true or not unless we have other arguments for or against it. I think (1) is false because, while I believe that IP is true, I do not think that IP implies that the person in the room understands Chinese. And I think that because the person in the room does not behave indistinguishably from someone who understands Chinese, contrary to what Searle says (See the bottom left column of p. 279). For one thing, a real Chinese speaker can understand Chinese even though they are not working in a room like that in Searle’s thought experiment. In order for Searle’s thought experiment to work, the person in the room must not understand any Chinese, and thus they could not generate correct responses to questions submitted to them in Chinese when they are working outside the room. So it is not really the behavioral dispositions of person in the room that match those of a Chinese speaker; it is rather the behavioral dispositions of the system consisting of the room, the person, the rule-books and the baskets of characters. So what IP really implies is that the system understands Chinese. Searle is right in insisting that manipulating Chinese characters according to formal rules does not allow the person in the room to understand Chinese, but one would not be justified in concluding from that that the system does not understand Chinese. In the literature, this response is known as the “systems reply”.
Searle replies to this very response (Searle, p. 280). He thinks that just as the person in the room has no way to figure out what the Chinese characters mean, the system has no way to figure out what the Chinese characters mean. But what is the person in the room or the system supposed to figure out? What is it to know or understand the meaning of an expression? Searle never says. I will concentrate my attention on this point because Searle’s replies to the other objections presuppose that his reply to this objection is successful.
For my part, I would wager that understanding any expression involves several things, such as—if the expression is a word—being able to define it, being able to use it in sentences, being able to recognize the objects or actions to which it applies (if it is a noun or a verb), and being able to teach others to employ it in the ways just listed. This account is basically what results when you take Ludwig Wittgenstein’s notion that the meaning of a word lies in its use in the language of which it is a part and apply it to understanding (Wittgenstein, Section 43, p. 20). If this account is correct then Searle’s reply will not work. If I want to know whether or not someone understands a word (or some other expression), I can ask them, and my estimation of their understanding will be based on what they go on to say or do. But how, in Searle’s thought experiment, should I determine whether the system understands a certain Chinese character (or a string of such)? Should I break into the room and ask the walls, or the baskets of characters, or the rule-books? I think not; that would be like trying to find out whether Searle understands a certain English expression by having a surgeon open Searle’s skull so I can query the various parts of his brain. For Searle can understand an expression quite well enough even if no part of his brain does. If I wanted to find out whether Searle understands an English expression, I would ask him as I normally would, face to face. Just so, if I wished to know whether the system understands a certain Chinese expression I would ask it, just as I normally ask it questions, by inserting the appropriate characters into the room’s input slot and evaluating the characters that come out of its output slot.
4. An Objection on Behalf of Searle
Searle could object that I have conflated a metaphysical issue with an epistemological one. For on the account of understanding I sketched above, understanding an expression consists in having certain behavioral dispositions. But while it is surely true that we gauge each other’s understanding of an expression by observing how we behave, Searle might say, this does nothing to show that our understanding is identical to these behavioral dispositions, nor even that our understanding and our behavioral dispositions necessarily accompany each other. We ought not to confuse evidence with what it is evidence of, and that is all our verbal behavior is: evidence of understanding. Now, acknowledging this distinction does not commit one to rejecting the idea that understanding consists in having behavioral dispositions, but it does mean that one must do more than merely appeal to our methods of determining whether someone understands an expression if one wants to show that idea to be true.
5. A Reply to the Objection
I concede the validity of this objection. I have not proved, and I cannot prove, that understanding consists in having behavioral dispositions, nor even that our understanding and our behavioral dispositions necessarily accompany each other. Nevertheless, I think it is a reasonable assumption to make. Of course, I would agree that verbal behavior is evidence of understanding. But we must go further and ask ourselves why we are justified in considering our verbal behavior to be evidence of understanding. In what follows, I will defend the view that our understanding and behavioral dispositions are necessarily connected. If it turns out that they are not only necessarily connected but in fact identical, that will only strengthen my position.
Let us suppose, for example, that our understanding and our dispositions for verbal behavior always occur together even though they have nothing to do with each other. If that were so, their constant conjunction would be a miracle. While we cannot rule out the hypothesis of such a miracle with complete certainty, it seems more reasonable to conclude that there is some real connection between them. This is a form of argument called “inference to the best explanation”. It works as follows. First, one observes a certain kind of phenomenon, pattern, or correlation, notes that it is highly improbable that the phenomenon, pattern or correlation is a matter of chance, and infers that there must be some explanation for it. Second, one examines various hypotheses and concludes that the hypothesis which best explains the observation is the correct one. In the case of our understanding of language and our verbal behavior, the best explanation I can think of is that one necessarily accompanies the other. But in which direction does the connection run? Is understanding sufficient for having certain behavioral dispositions? Or is having certain behavioral dispositions sufficient for understanding? Or does the connection perhaps run in both directions? Searle could not accept the second or third possibilities, for then having certain behavioral dispositions would, by itself, suffice for understanding—otherwise, having those behavioral dispositions would not really be sufficient for understanding–and in that case any computer which had the same behavioral dispositions as a Chinese speaker would really understand Chinese. Syntax would be sufficient for semantics, and Searle’s critique would fall apart. But if Searle’s view is incompatible with the second and third possibilities, it becomes utterly mysterious why, on his view, we should take anything’s dispositions for verbal behavior as evidence that it understands a language. His view, when followed to its logical conclusions, commits him to hold that dispositions for verbal behavior could occur in the complete absence of any linguistic understanding. After all, for the Chinese Room Argument to work the system must lack understanding even though it mimics a Chinese speaker’s dispositions for verbal behavior perfectly. If the mimicry was imperfect, or the system really did understand Chinese, Searle’s argument would not serve as a counterexample to the thesis he is trying to refute. But if dispositions for verbal behavior can occur without any real understanding, why should one believe, for example, that one’s fellow human beings understand anything they or others say? Searle does say that he thinks that our capacity to understand is based on our biological make-up (Searle, pp. 281-283), but aside from the arguments we examined above he seems to have no evidence to support this. The upshot is that accepting the conclusion of Searle’s argument leaves one in a very skeptical position, a position that we would do best to avoid.
In the above I have tried to show that we are justified in rejecting the premises that lead to Searle’s skeptical predicament. Of course, one can never completely quell skeptical worries of this sort. In this instance, to use Searle’s terminology, we cannot conclusively establish that syntax is sufficient for semantics. Be that as it may, we have some good reasons for thinking that it is, and no good reasons for thinking that it isn’t.
Searle, John. “Can Computers Think?” Analytic Philosophy: An Anthology, edited by A. P. Martinich and David Sosa. Blackwell Publishers Ltd, 2001
Wittgenstein, Ludwig. Philosophical Investigations, 3rd Edition. Translated by G. E. M. Anscombe. Engelwood Cliffs, New Jersey: Prentice Hall, 1958.
What do you think of this article? Discuss it on Scholardarity’s message board.
PARENT PAGE: Philosophy