Here is the outline of a presentation I am going to give tonight in my Philosophy of Mind seminar:
1. The Chinese Room
a. The thought experiment: Searle is trapped in a room equipped with a book (written in English) which contains the proper “program” for manipulating Chinese symbols in a way which allows him to pass the Chinese Turing Test. Thus, he behaves as if he understood Chinese without actually understanding Chinese.
b. What does the computer have which Searle in his Chinese Room lacks? Nothing.
c. What does a competent Chinese speaker have that computer, therefore, lacks? Understanding.
2. Searle’s conclusions
a. The syntax of a program by itself is neither constitutive of nor sufficient for semantic content while minds do have semantic content.
b. The purely abstract, formal syntax of a program is causally inert while minds do have causal efficacy.
3. Objections:
a. There actually is understanding in Searle’s Room.
i. The Systems reply: the entire system understands Chinese.
1. Searle: Allow me to internalize all the rules, there is still no understanding.
ii. The Virtual mind reply: Searle confuses the syntax of representation with the representation of syntax. The shuffling of represented syntax at the level of the man creates a second mind (this one for the entire system) at an entirely different level.
1. Searle: This is a bit speculative and metaphysically extravagant.
b. The right program is not being instantiated.
i. What makes a program the “right” one?
1. Behavior: The proper input/output relations make a program the right one.
2. Function: Both the input/output relations as well as the formal relations which hold between the two make a program the right one.
3. Searle: I can have whatever formal program you like and the point still stands.
ii. The Robot reply: If we allowed the Chinese Room some form of sensory input, it could attach meaning to its symbols.
1. Searle: The sensory input would just be more, meaningless syntax (The Matrix).
iii. The Brain simulator reply: If we simulate every neuron firing on a Turing Machine it will understand.
1. Searle: Simulation is not duplication. The Turing Machine only models rather than instantiates the proper causal properties.
c. The program is not being instantiated in the right way. (Searle’s point.)
i. The Instantiation reply: The symbols need to be the efficient causes which bring the machine from one computational state to another. What is supposed to matter is the causal dynamics among the symbols not between the symbols and some operator.
1. Searle: This is to (at least) partially concede that causal powers matter. Beyond this, however, the symbols are still without content.
ii. The Neural-Net reply: A properly trained neural-net is the way which the right program must be instantiated.
1. Searle: Any neural-net program can be run on a serial machine, taking us back to the original problem.
2. Searle: Consider an enormous Chinese gym in which a vast army of people are performing the proper program. There is still no understanding of Chinese.
4. The Biggest Problems:
a. If one recognizes the difference between the syntax of representation and the representation of syntax, it is difficult to see how the argument gets off the ground at all. Perhaps a non-linguistic thought experiment might be more compelling?
b. It is not clear that strong AI is as committed to the total isolation of syntax from all causal considerations as Searle suggests it is.
c. It is unclear what role, if any, causal powers are supposed to play in the Chinese Room Argument. Is it just an ad hoc addition aimed at preventing counterexamples? What connection, if any, is supposed to exist between causal powers and semantic content?
