1. The Chinese Room thought experiment: Searle (who understands no Chinese) is locked in a room with an instruction book (written in English), plenty of paper, and slots for incoming and outgoing paper. Searle receives paper with "squiggle squoggles" on it through the in-slot and, by following the instructions in the book, draws "squoggle squiggles" on paper, and sends it through the out-slot. The operation of the Chinese Room would appear to pass a "Turing Test" conducted in Chinese: a native Chinese speaker outside the room would be deceived into thinking he was communicating with someone who genuinely understands Chinese inside the room.

2. This is set up as an argument from analogy. Just as Searle doesn’t understand Chinese by dint of following instructions in a book, we shouldn’t believe that a computer could genuinely understand English (or any other language), merely by virtue of following a computer program.

A. The Systems reply: Inside the room, Searle might lack an understanding of Chinese. However, he’s part of an overall system (including the room and all its contents) which does understand Chinese.

Searle’s response: Have Searle "internalize" the entire system (commit the book to memory, etc.). Still, he would lack any genuine understanding of Chinese.

B. The Robot reply: Perhaps the problem is that the original Chinese room doesn’t possess the full range of interaction with the world that genuine speakers of a language do. So why don’t we add input from sensory channels and output that corresponds to muscle commands. In effect, let’s imagine the man locked in the room as controlling a robot.

Searle’s response: That changes nothing. Even with the additional input and output channels, the man locked inside the Chinese room (or robot) would still be unable to understand Chinese.

C. Brain Simulator reply: Suppose instead that the instructions in the book in effect has the man simulate all of the neural activity of a genuine speaker of Chinese. Since the responses of the Chinese room would then be generated in roughly the same way as the responses of a genuine speaker, wouldn’t that mean that there’d be genuine understanding of Chinese in the Chinese room?

Searle’s response: Not at all! Simulations need not possess the same powers or properties as that which they simulate. A simulated hurricane, for instance, lacks the capacities of a real hurricane; it can only flatten unreal, simulated cities.

D. While they might not succeed singly, perhaps some combination of the replies will nevertheless undermine the conclusion of the Chinese Room thought-experiment.

Searle’s response: Given that each of the responses fails individually, it’s hard to see why they would nevertheless succeed in combination.

4. Is Searle being too glib, especially in this last response? Consider a combination of the robot reply and the system reply: We don’t attribute understanding to entities inside the head. Rather, we attribute understanding to functioning human beings interacting in the world.

Our judgments about understanding do not obviously depend upon assessments of what exactly is going on inside. As another thought experiment, suppose we were to discover a little man (or a CPU) inside the head of someone we earlier took to be a normal human being. Would that undermine our earlier beliefs that they understood language? Should it?