ANALIZADO DESDE EL PUNTO DE VISTA JURIDICO

to its input. In such a case the system would, as it were, "talk gibberish" and would not be doing anything that would count as information processing, solving information processing problems, or extracting information.

In making these points I am not turning my back on the syntactic account of com putation that I advanced earlier. For though computation, by its very nature, involves the m anipulation of meaningful items and the generation of output that is semantically related to the input from which it is generated, computers and computational processes are individuated syntactically rather than semantically.

I feel tempted to claim that the notion of a computer is partly a teleological one; that nothing is a computer unless it is its function to process or extract information, solve problems, work things out, and such like. Computers don't just do these things; in addition they are used to do these things and in so being used are of a benefit to their user. With respect to the brain this suggests the following. A subsystem of the brain might satisfy all of the above described conditions but if it doesn't benefit the system in which it is housed in virtue of generating the information that it generates (if it isn't, so to speak, used to generate that information or if that information doesn't play a significant role in the life of the embedding system) then it isn't a computer.

This attempt to specify the conditions that a physical system must satisfy in order to be a computer could do with much in the way of elaboration. However, I think that I have done enough to indicate the outlines of an adequate answer to the question of how a physical system could be a computer, an answer which tells us in virtue of what those physical systems that are computers attain that status. What should be clear is that it is very hard to be a computer, contrary to what some philosophers would have us believe.

Suppose that physical system S is a computer that manipulates symbols of L. What computational capacities will S have, what symbolic / syntactic functions will it be able to compute? The answer is that it all depends upon w hat counterfactual-supporting generalisations concerning its internal state transitions are true of S (or its parts). Given the instantiation function, corresponding to each such generalisation will be a syntactic generalisation. Thus if it is true

of s that whenever it tokens an internal state of type I' it subsequently tokens an internal state of type I", then it will also be the case that whenever it tokens the symbol F' it will generate from it a token of the symbol F" (where F' and F" are the symbols that have, respectively, F and I" mapped onto them). Given the huge network of generalisations relating its internal physical states to one another, there will be a huge network of syntactic or symbol-manipulating generalisations true of S. Which such generalisations are true of S will determine which symbol-manipulating capacities it has, or which symbolic functions it is capable of computing. Suppose that it is claimed of S that it can compute the symbolic function SF, a function defined by the rule R. This claim will be true if and only if the syntactic generalisations that are true of S are such that whenever S tokens a symbol that is an argument of SF, that token causes S to token the symbol that is the value of SF for that argument (or, alternatively, S responds to the token by producing a token of the symbol that is the value of SF for that argument). If this condition is satisfied, some of S's (potential) symbol manipulating activity can be described as generating symbols of L from symbols of L by applying rule R.

We have seen that syntactic properties or syntactic types are multiply realisable. But is it true to say, as does Block (1989), that 'syntax is . . . a functional notion', and that 'it is having a certain

functional role that makes a state satisfy a syntactic description' (p.

142)? For syntactic properties to be functional properties it would have to be the case that a state of a system's having a given syntactic property was a matter of its bearing certain counterfactual-supporting causal relations to other states of the system. A consideration of symbols containing logical connectives would seem to support Block's view. Most powerful formal languages contain logical connectives, and such connectives are a primary way by means of which simple sentences are combined to build more complex

sentences. Consider the connective &. For a state of a system to have

the syntactic property of being a sentence of the form A & B certain

causal generalisations would have to be true of the system. Suppose that the state in question is state T of system S. For I' to have the

syntactic property of being a symbol of the form A & B it would have

causes the state that maps onto A , (or realises A in S), and the state

that maps onto B (or realises B in S). In other words, whenever S

tokens I', that token causes a token of the state that maps onto A, and

a token of the state that maps onto B. Second, I’ is jointly caused by

the state that maps onto A and the state that maps onto B. In other

words, whenever S tokens the state that maps onto A and tokens the

state that maps onto B , those two tokens jointly cause a tokening of

These generalisations correspond, respectively, to the familiar rules of &-elimination and «^-introduction. The same will hold,

mutatis mutandis, for all other connectives; that is, for any connective C, for a state of a system to have the syntactic property of being a symbol containing C as its main connective, generalisations corresponding to the introduction and elimination rules for C must be true of the system.

Indeed something like the above would also appear to apply to quantifiers. For a state of a system to have the syntactic property of containing a particular quantifier Q, that state must figure in causal generalisations that correspond to the introduction and elimination rules for Q.

However, we have to be careful, for it is certainly possible for two distinct computational systems, or the same system at different points in time, to employ the same language, yet to manipulate the symbols of that language differently. In such a case the causal relations between the symbolic states of one of the systems will diverge from the causal relations between the symbolic states of the other. Many of the syntactic generalisations that are true of a com putational system are contingent in the sense that those generalisations don’t have to be true of the system given the formal language that it employs; it is consistent w ith a com puter's employing a particular language that it manipulates the symbols of that language in many different ways. Indeed, what happens when one programs a computer is that one brings about changes in the way it manipulates the symbols of the language that it employs; in other

31 Strictly speaking this is a little too strong as S need only tend, or have a tendency