Does the Chinese Room argument hold against AI?

Question

Back in college, I had a Complexity Theory teacher who stated that artificial intelligence was a contradiction in terms. If it could be calculated mechanically, he argued, it wasn't intelligence, it was math.

This seems to be a variant of the Chinese Room argument. This argument is a metaphor, where a person is put in a room full of Chinese books. This person doesn't understand a word of Chinese but is slipped messages in Chinese under the door. The person has to use the books, which contain transformation rules, to answer these messages. The person can apply the transformation rules but does not understand what (s)he is communicating.

Does the Chinese room argument hold? Can we argue that artificial intelligence is merely clever algorithmics?

Answer 1

It depends on the definition of (artificial) intelligence.

The position that Searle originally tried to refute with the Chinese room experiment was the so-called position of strong AI: An appropriately programmed computer would have a mind in the exact same sense as humans have minds.

Alan Turing tried to give a definition of artificial intelligence with the Turing Test, stating that a machine is intelligent if it can pass the test. The Turing Test is introduced here. I won't explain it in detail because it is not really relevant to the answer. If you define (artificial) intelligence as Turing did, then the Chinese room experiment is not valid.

So the point of the Chinese room experiment is to show that an appropriately programmed computer is not the same as a human mind, and therefore that Turing's Test is not a good one.

Answer 2

There are two broad types of responses to philosophical queries like this.

The first is to make analogies and refer to intuition; one could, for example, actually calculate the necessary size for such a Chinese room, and suggest that it exists outside the realm of intuition and thus any analogies using it are suspect.

The second is to try to define the terms more precisely. If by "intelligence" we mean not "the magic thing that humans do" but "information processing," then we can say "yes, obviously the Chinese Room involves successful information processing."

I tend to prefer the second because it forces conversations towards observable outcomes, and puts the difficulty of defining a term like "intelligence" on the person who wants to make claims about it. If "understanding" is allowed to have an amorphous definition, then any system could be said to have or not have understanding. But if "understand" is itself understood in terms of observable behavior, then it becomes increasingly difficult to construct an example of a system that "is not intelligent" and yet shares all the observable consequences of intelligence.

Answer 3

First of all, for a detailed view of the argument, check out the SEP entry on the Chinese Room.

I consider the CRA as an indicator of you definition of intelligence. If the argument holds, yes, the person in the room understands Chinese. However, let's sum up the three replies discussed in the SEP entry:

1. The man himself doesn't understand Chinese (he wouldn't be able to understand it when outside the room), but the system man+room understands it. Accepting that reply suggests that there can exist an intelligent system which parts aren't themselves intelligent (which can be argued of the human body itself).

2. The system doesn't understand Chinese, as it cannot interact with the world in the same way a robot or a human could (i.e. it cannot learn, is limited in the set of questions it can answer)

3. The system doesn't understand Chinese (depending on your definition of understanding), and you couldn't say a human performing the same feats as the Chinese room understands Chinese either.

So whether the argument, or a variant of it holds, depends on your definitions of intelligent, understanding, on how you define the system, etc. The point being that the thought experiment is a nice way to differentiate between the definitions (and many, many debates have been held about them), in order to avoid talking past each other endlessly.

Answer 4

Depends on who you ask! John Searle, who proposed this argument, would say "yes", but others would say it is irrelevant. The Turing Test does not stipulate that a machine must actually "understand" what it is doing, as long as it seems that way to a human. You could argue that our "thinking" is only a more sophisticated form of clever algorithmics.