Much of cognitive science and philosophy of mind is oriented around computational and functionalist theories of mind which hold, respectively, that the human mind is a digital computer (in a particular sense) and that what makes a mind a mind is not what it’s made of but rather how it functions (in which case, human minds and computers may function in the same way). Strong AI is the thesis that computers are (at least potentially) capable of human-like thought. John Searle’s Chinese Room thought experiment and related argument for the impossibility of Strong AI is commonly referenced in philosophical discussions of the mind and has been cited as central to cognitive science. If Searle is right and Strong AI is false, then much of the current efforts in cognitive science and philosophy of mind are futile. However, Searle’s reasoning contains a modality problem and an oversight of computers’ inherent semanticity.

Briefly, the thought experiment and argument run as follows¹. A person who understands only English sits alone in a room following English instructions for manipulating Chinese linguistic characters such that to those outside the room it appears that the person inside it understands Chinese. (The following argument is my own formulation.)

• 01) The person who merely manipulates characters of a natural language does not truly understand the language.
• 02) AI is analogous to the person who merely manipulates characters of a natural language.
• 03) If 01 and 02, then AI cannot truly understand natural language.
• 04) If AI cannot truly understand natural language, then Strong AI is false.
• 05) Therefore, Strong AI is false.

Many of Searle’s respondents have pointed out that Strong AI need not be oriented around a mere symbol-manipulating computer with no perceptual access to the referents of the symbols as would be analogous to the Chinese Room scenario. For instance, an android with such access and with such symbol-manipulating power may very well come to understand what the symbols mean in the same way that children come to understand language. We simply don’t yet know enough about how that happens to know if that process can be duplicated in silicon and until we do our tendency to impugn said android’s capability to understand language is based merely in untutored intuition. Therefore, like many others, I don’t believe we can assent to Premise 02.

The first premise of Searle’s argument is widely regarded as having brought to light another point, though, which is that syntax doesn’t seem to be sufficient for or constitutive of semantics. Incorporating that notion into the Chinese Room scenario has inspired this variation (again, my own formulation of others’ thoughts based on Searle’s 1984²):

• 06) Computers are merely syntactic.
• 07) Syntacticity is not a sufficient condition for semanticity.
• 08) Semanticity is a necessary condition for mind-hood.
• 09) If 06, 07, and 08; then computer-hood is not a sufficient condition for mind-hood.
• 10) If computer-hood is not a sufficient condition for mind-hood, then Strong AI is false.
• 11) Therefore, Strong AI is false.

Premise 08 is widely agreed upon and I can see no reason to dispute it. Such is also the case with 10 which is developed simply from the Strong AI thesis itself.

If our basis for evaluation is the current state of computer science we are prompted to assent to 06. Such a qualified version of 06 would run as follows:

• 06b) Computers as we know them are merely syntactic.

However, 06b is not strong enough to oppose Strong AI because it is a thesis concerned with both present and future manifestations of computer science. In other words, if the antecedent of 09 includes 06b, then 09 will be false. What is needed for 09 to be true is a modal version of 06:

• 06c) Computers are necessarily merely syntactic.

Now we are left with the task of evaluating 06c and 07. To my knowledge, there is no currently-known theoretical or empirical reason to assert either premise.

On the metaphysical/physical construal of 06c’s modality, one could make 06c true by restricting the referents of ‘computers’ to merely-syntactic machines, but, as the above discussion of 06b makes clear, that would not strike at the heart of the Strong AI thesis. Therefore, short of interpreting 06c in this bizarre fashion, we simply don’t know if computers are, by metaphysical/physical necessity, merely syntactic and therefore cannot judge 06c to be true. Furthermore, on the epistemological construal of 06c’s modality, the premise is false. There is neither a credible theory nor any empirical evidence our countenance of which forces upon us the necessary conclusion that computers are merely syntactic.

Moreover, I believe we have a reason to reject 06c. Executing programs causes computers to undergo state changes. The only way we can make any sense of this is to conclude that the executing computer understands the instructions for execution–an understanding that involves semantics. To make the point clear, consider a computer/robot in an automobile factory sticking out its screwdriver and rotating it in response to its programming. In order to understand the program’s instructions the computer must know that an expression in those instructions (let’s say the binary code or simply the voltage equivalent of ’screwdriver’) refers to one definite object (its screwdriver), that another expression refers to an action (rotating), and so on. Just as the computer in the auto factory acts on parts of itself that we can see and therefore must be capable of semanticity where the referential terms for those parts appear in a program it is executing, so all computers must be capable of semanticity where the referential terms for their microscopic parts appear in the programs they execute.

Note that my insistence on the semanticity of computers is compatible with both the internalistic and externalistic approaches to explaining semanticity by way of causation. Also note that this is not to say that computers are conscious in the sense of having second-order awareness. I am not claiming that the computer is aware that it is a causal agent, that it is aware that it is manipulating symbols, or that it knows what symbols are. Lastly, if my reasoning is correct thus far, then Premise 07 is irrelevant: computers cannot be merely syntactic so the sufficiency of mere syntacticity does not figure into the arguments for or against Strong AI.