Two fundamental flaws in Searle’s argument for his Chinese Room thought experiment, showing that, as a consequence, it fails to raise significant doubts about functionalism.
First, Searle assumes that the rulebook in his postulated scenario, complex enough to answer any Chinese query convincingly, but not so complex so as to have intentional states could actually be written, without giving any reason why this should be so.
Second, closer observation of his argument reveals an entirely unsupported double-standard: the fundamental assumption that the nature of humans is somehow intrinsically greater than that of machines is a crucial and unjustified part of his theory. Regardless of whether or not functionalism is an acceptable a theory of the mind, it is clear that Searle's argument is, by itself, insufficient to trigger its downfall.
For the purpose of this argument, functionalism will be taken to mean the theory in which mental states are defined by the causal relations they have with inputs, other mental states, and behaviour1. If a potential mind in question receives certain inputs and produces certain outputs indistinguishable from those that we would associate with our own mental states, then the theory of functionalism would indeed ascribe to it mental states. A central consequence of this theory is the concept of Multiple Realisability. This would contend that, since it is only an entity's function (and not its composition) that may credit it with thought, humans are not the only entities that can have cognitive states. Functionalism thus allows for the emergence of what Searle calls strong artificial intelligence2, the notion that a computer with the right program can be said to actually understand something, and hence be able to hold mental states.
In Minds, Brains, and Science, Searle directly confronts functionalism by strongly objecting to the idea held in Multiple Realisability that a program might attain mental states; he insists that any mental states one might ascribe to such a program are mere “illusions”3. In support of his objection, Searle provides a nomologically possible thought experiment: explained briefly, he imagines a man who has no understanding of Chinese4 characters locked in a room and fed written Chinese stories, and returning written Chinese responses according to a highly complex instruction manual.
It must be stressed that, though his responses are deemed to be indistinguishable from those of native Chinese speakers - and the system could easily pass the Turing Test - the man merely manipulates formal symbols, of which he has no comprehension, according to the rulebook. Searle rightly asserts that the man does not “understand a word of Chinese stories”5. However, he then goes on to state that, analogously, a computer program can never by itself hold understanding, or any other cognitive state.
The first flaw in Searle's argument is that it revolves around two simultaneous assumptions about his thought experiment's rulebook: that such a rulebook could actually fool a native Chinese speaker, and that such a rulebook does not hold intentional states. (Intentionality can be defined as an act or state that is directed at objects and characterised by the objects at which it is directed6.) Though both of these assumptions are theoretically possible, it is highly doubtable that they could co-exist. It is entirely feasible that a rulebook, or program, could be created with no intentional states.
Indeed, we already have countless rulebooks like this today; a cookbook (or cooking computer program), a French-English dictionary (or online translator), or a Physics textbook (or pre-programmed calculator) are excellent examples. However, were we to try to converse with a Physics textbook (or even with a pre-programmed calculator) we could never be fooled into believing it were human. It is, however, a necessary branch of Searle's argument that such a program could be created. Any program that successfully produces satisfactory answers to the infinite number of possible questions that could be asked of it and yet does not hold any intentional states, would, however, necessarily be an infinitely long program.
The very nature of language, with unlimited possible expressions, necessitates that such a program, if it were non-intentional, could never be written if all. Searle himself even implies that such a program could never be written, suggesting that it is implausible to hold “the supposition that we can construct a program that will have the same inputs and outputs as native speakers.”7 Searle can give no reason why his program could ever be possible, precisely because it could never be.
Were this program to contain some elements of intentionality within it, and thus satisfy the input/output requirement without taking an infinite period of time to do so, Searle does not object to the fact that it would in this case hold cognitive states; in fact, he writes that a digital computer with “the level of description where it can correctly be described as the instantiation of a computer program… can think.”8
