You could mail me or go to my home page, or my main philosophy page.
Searle's Chinese Room argument has caused a great deal of stir of late. This originally appeared in 1980, in an article called "Minds, brains and programs" in Behavioral and Brain Sciences, volume 3. As far as I can tell, it is not available on line. Unfortunately, I have not seen anybody describe the basic fallacies of the argument, so here they are.
Back around 1950, Alan Turing proposed what became known as the "Turing test": suppose we have a properly programmed computer and a human. In interviewing them under suitable conditions, so that we see only the responses, could we tell the difference? Turing stopped there, but the question lingers. If we had such a computer (and we don't even come close today), would we consider it to think?
One obvious answer is that we would consider it to be a thinking being, since we conclude that humans are intelligent on the basis of conversation. Another is that it isn't; that thinking, whatever it is, is not just making intelligent conversation, and that what we take as indications of thought in people are just indications, and would not prove a machine intelligent. Some people think that current computers will not be developed to pass the Turing test, and don't worry about the consequences.
Searle is willing to consider that a computer might pass the Turing test, but considers that it will not think or possess intentionality, two things he seems to link. His formal argument is that computers operate on syntax (maybe), that thought is semantical (yes), that syntax does not produce semantics (not obvious), and therefore that computers don't think. This would not have drawn attention, except for his informal argument.
Searle attempts to show that something can seem to understand something, pass all tests on it, and still not understand it. My goal is to show that his arguments are worthless. I have no comment here on whether computers will ever seem intelligent, or whether there are other, possibly valid, arguments that computers cannot understand things.
Suppose we have a computer program that, when run on a suitable computer, can pass for human. We can presumably program it to simulate a Chinese human, so that it reads and writes Chinese fluently. Assuming that a computer can pass the Turing test, this is all correct.
Now, suppose that Searle prints out the program listings, arms himself with immense amounts of paper and pencils, and runs the program on himself, noting down all intermediate results on paper and referring to the program to see what to do with them. This is possible, although for any large program (and the Turing test program would be exceedingly large) it is excruciatingly slow and very error-prone. Searle is in the room, accepting Chinese characters that are meaningless to him, and sending out other meaningless characters.
The Chinese Room is alleged to understand Chinese. Searle doesn't understand Chinese. The paper and program listings obviously understand nothing. Since there is no component of the system that understands Chinese, certainly the system as a whole can't. This argument relies on a hidden assumption: that something that understands must have a part that understands. This denies that understanding, thought, and intentionality can be emergent phenomena.
The argument depends on the hidden assumption. If we don't agree that a system that understands something must have a component that understands, the argument falls to the ground. We say, "Yes, the paper understands nothing, Searle doesn't understand Chinese, the program understands nothing, the pencils understand nothing...so what's your point? This doesn't mean that the room as a whole can't understand Chinese." If we embrace the assumption, things start looking very, very strange.
Searle claims that intentionality and thought are fundamentally biological. This means that biological organisms think and understand things. Now, all biological organisms are made of organic chemicals, and these are made of atoms, these of subatomic particles, and these of leptons and quarks. If we assume that consciousness or thought cannot be emergent phenomena, then we must assume that, in every philosopher, there is a sentient quark or lepton, which is part of an intelligent atom, which is part of an organic molecule that possesses a mind. This is absurd, and certainly Searle would not admit it.
Searle anticipates that someone will claim that the Chinese room is intelligent as a system, even though its components do not understand what is happening. He calls this the "systems reply", and proposes the following counterexample.
Searle is to memorize the entire program and carry out the operation entirely mentally. At this point, he assumes that his opponent will claim that there are two separate and different sorts of minds in his brain. Searle replies that one mind is different from the other in that one mind knows what hamburgers are, in a sense, and the other merely can say intelligent things about hamburgers. This is a possible argument against the Turing test, but Searle does not go into it. His main argument is that there is not two minds, since his original argument has shown that such computation is insufficient to produce a mind.
Let's consider this carefully. It is completely impossible to memorize and execute even a relatively small program in that fashion. To assume that Searle can do this with a large program is ridiculous. We can allow this experiment for the sake of argument, with caveats.
First, we insist that Searle demonstrate some sort of logical contradiction to make his point, not just odd results. If I base an argument on the premise that I swallow the Atlantic ocean, I cannot create a reductio ad absurdam by showing that I no longer fit into my house. If we allow Searle's assumption, we are bound to find strange and counter-intuitive results, because the assumption is flatly impossible.
Second, we will disregard Searle's claim to have a mind of infinite power. If Searle can internalize a Chinese-speaking version of Searle, then presumably that Chinese speaker can internalize Basque, and we suddenly find no limits to what Searle's brain can hold.
Once Searle has swallowed the ocean, it seems reasonable that he has two minds in his head, one based on his brain in normal fashion, and one based on the program he has memorized. If we ask him a question in English, he will presumably reply normally. If we hand him a question written in Chinese, he will presumably stand still for a long time in deep thought, and write out a meaningful answer. Aside from the delay (which could be decades), and the fact that he reads and writes Chinese but speaks none of it, we cannot tell the difference between the two minds.
Searle claims that the Chinese mind is not a true mind, since it is simply symbol manipulation, and in his initial argument he has shown that that is insufficient. The trouble is that, if we are attempting to refute him, we are obviously assuming that his initial argument is flawed, and therefore we can refuse to accept it. Searle is begging the question and arguing in a circle, to use logician's terms.
Searle continues further in his paper, but introduces no new arguments.
Searle makes no valid arguments. His initial argument denies that understanding can be a property of a system that has no components that are understanding, which is absurd when considered carefully. His follow-up arguments are all based on his initial argument, and thus are invalid.
It is conceivable that we cannot build a computer that will pass the Turing test, although I think we will in the next few centuries. It is conceivable that such a computer will not be human in various senses, but Searle has provided no support for that conclusion.
All contents of these pages Copyright 1997 by David H. Thornley.