A problem with Searle's Chinese room argument

I used to be a big fan of John Searle's "Chinese room argument" against "strong A.I.". I still agree with Searle in rejecting strong A.I., but I now have my doubts about the effectiveness of the Chinese room argument. In this post, I explain the problem that I see with the argument.

"Strong A.I." is a term that Searle coined. It refers to a claim made by certain philosophers and scientists who investigate artificial intelligence (A.I.). Roughly speaking, strong A.I. is the theory that a computer with the right programming  is a genuine mind simply by virtue of its programming. More precisely, strong A.I. says that a system (e.g. a computer) has mental state M as long as it follows a set of rules that cause it to behave as if it has M.

The Chinese room argument is one of Searle's most famous ideas, and literature on it abounds. I won't bother to hunt down the specific references to it in Searle's writings. Basically, the argument is as follows. Imagine that Searle is sitting inside a room dubbed the Chinese room. Outside the room, a Chinese speaker writes Chinese questions onto slips of paper and passes the slips into the Chinese room through a slot. Searle writes replies to the questions and passes them out through the slot. Searle doesn't know Chinese and can't read the questions that he receives. However, he has a rulebook next to him that tells him what to write in response to each combination of Chinese characters. Thus, to the Chinese speaker who receives Searle's replies, it looks as if Searle knows Chinese. In other words, Searle is following rules that cause him to behave as if he knows Chinese. Yet he does not in fact know Chinese. Therefore, Searle concludes, strong A.I. is false.

The most common objection to the Chinese room argument is the "system" reply. The system reply rightly notes that the Chinese room argument is misleading in its portrayal of strong A.I. On a charitable interpretation, strong A.I. would not entail that Searle himself knows Chinese while sitting inside the Chinese room. It is not Searle himself but, rather, the whole room that behaves as if it knows Chinese. (Searle himself is constantly looking at his rulebook to find out what to write, and that is not the way someone who knows Chinese would behave.) Thus, the system reply argues, it is the room as a whole (the combination of Searle, his rulebook, his writing instruments, and the walls of the room) that knows Chinese—and that remains true even if Searle himself doesn't know Chinese.

Searle has a response to the system reply. Imagine, he says, that he simply memorizes all the rules in the rulebook. If he does so, then he can write responses to Chinese questions without being part of a larger system, and yet he still doesn't know Chinese. Thus, Searle concludes, a human being can follow rules that cause him to behave as if he knows Chinese, and yet not know Chinese; ergo, strong A.I. is false.

And it is here, I think, that Searle goes wrong.

Those who support the Turing test may believe that the ability to answer Chinese questions is sufficient for knowing Chinese. But there's no reason why strong A.I. proponents in general must accept that belief. After all, normal Chinese speakers do a lot more with their knowledge of Chinese than answer a bunch of questions. They can, for example, ask for a glass of water when they are thirsty. For Searle to really behave as if he knew Chinese, he would need to know how to write "Please give me a glass of water" in Chinese, and he would have to know to write this if and only if he wants a glass of water. It's not clear to me how this knowledge would differ from knowledge of the meaning of "Please give me a glass of water" in Chinese. In short, if Searle memorized a set of rules that really allowed him to act as if he knew Chinese, then it's not clear to me how Searle could fail to know Chinese.

Of course, Searle could construct a more plausible scenario in which an entity behaves as if it knows Chinese without knowing Chinese. Imagine an electronic robot that is programmed to behave as if it knows Chinese, that is also programmed to behave as if it is thirsty, and that is programmed to say "Please give me a glass of water" in Chinese if and only if its thirst-behavior has kicked in. Surely, Searle might say, this robot does not know Chinese, even though it follows a set of rules (i.e. its programming) that cause it to behave like a Chinese speaker. I would be inclined to agree with Searle about this. Yet, in saying this, Searle would simply be begging the question against strong A.I., which is precisely the claim that such a robot would know Chinese. More importantly, by changing his hypothetical rule-follower from a human being to a robot, Searle would effectively be abandoning the Chinese room argument in any recognizable form.

Thus, I think the Chinese room argument fails. It simply does not demonstrate that strong A.I. is false.

Slightly amended on November 21, 2011.