functionalism(1)

*The view that the physical realization of a functional component is not, in some sense, its essence. Rather, what makes a functional component the type it is, is characterized in terms of its role in relating inputs to outputs and its relations to other functional components. See multiple realizability, Turing machine, functionalism(2), causal functionalism.*

**Details:**

A popular way to pump the intuition that all that needs to be kept constant about a subject to keep the subject's mental properties constant is the set of causal relational properties of the subject is what has come to be known as "the silicon chip replacement" thought experiment.

Ned Block (1980) identifies three senses of functionalism. The first is simple decompositional functionalism. 'Functionalism' in this sense refers to a research strategy that relies on the decomposition of a system into its components; the whole system is then explained in terms of these functional parts. Second, computation-representation functionalism is a special case of decompositional functionalism which relies heavily on the 'computer-as-mind' analogy. Psychological explanation under computation-representation functionalism is "akin to providing a computer program for the mind" (Block 1980, p.171). Thus, mental processes are seen as being decomposable to a point where they can be thought of as processes which are as simple as those of a digital computer or, similarly, a Turing machine. Lastly, Block identifies metaphysical functionalism. This form of functionalism is a theory of mind that hypothesizes that mental states simply are functional states. The metaphysical functionalist claims that mental states are the types of mental state they are because of the causal relations between inputs, outputs and other mental (i.e. functional) states of the system, as in the Turing machine. The physical implementation of the set of functions which implement a mind are irrelevant to what makes something a mind – it's the functional relations that count.

In general, proponents of each of these forms of functionalism make the claim that the physical realization of a given function is not, in some sense, its essence. Metaphysical functionalism identifies causal structures with mental states which are realizable by "a vast variety of physical systems" (Block 1980, p. 173). Metaphysical functionalism often identifies mental states with Turing machine "table states" (Block 1980, p. 172). Similarly, computation-representation functionalism holds that the information processing mechanism necessary to run the 'mind' program could be hydraulic, electric, mechanical, or whatever. Lastly, decompositional functionalism is, superficially at least, more interested in the function of a system than the physical makeup of the system.

However, decompositional functionalism is seldom, if ever, completely divorced from the physical system it is used to explore. While it is true that in decompositional functionalism the function of the system being explored is often abstracted from its physical realization, it is also the case that that realization is continuously re-examined in order to better understand and characterize its function. Decomposition is not a simple matter of observing a system, noting the various functions, and constructing a model. It is a complex process that involves examining and re-examining and re-re-examining ad infinitum (or at least until the model is just about right) the original physical system. In other words, decomposition is severely constrained by the system being decomposed (Bechtel and Richardson 1993).

As a theory of mind, functionalism is quite appealing . It is based in mathematical proof and provides a means of constructing analogies to guide our understanding of the mind. Computation-representation functionalism is a clear example of such a 'rigorous' analogy. In fact, because both digital computers and people are presumably Turing machine describable, the computer/brain analogy can be supported by direct reference to computational theory. This is how Turing machine equivalence has played such a central role in supporting functionalist intuitions (Fodor 1981).

However, what the Turing machine really provides for cognitive scientists is not a notion of equivalence and a means for justifying purely functional descriptions of the mind, but rather a very general notion of computation. The Turing machine describes the whole space of possible computational systems, so anything computational lies in the Turing machine delimited space. Since any given real system is subject to many, possibly infinite, computational descriptions, any given real system will lie at many points in this space. Thus, the notion of Turing machine equivalence provides little information concerning what is 'the same' about any two real systems – it simply notes the fact the systems share one point in Turing machine space. Functionalists provide no means of distinguishing one computational description of the brain from another. It is highly unlikely that they are all important to having a mind. Thus, we need a better way of understanding the relationship between real computational systems than that provided by the Turing machine.

The realization of this under-specification of relevant functional properties also gives rise to the anti-functionalist argument first forwarded by Ryle (1949). Ryle claims that functionalism does not solve the problem of describing a complex system's behavior. Rather, it makes a system more complex by introducing more functions. What Ryle has noted is that there will be an infinite proliferation of unexplained functional terms since each upper-level function is decomposed into more lower-level functions under this methodology.

However, as Cummins (1983) remarks, Ryle has missed the important fact that each of the newly posited functions is simpler than the one it is intended to help explain. Thus, the decomposition will eventually be simple enough for us to understand and use in explaining the highest-level functioning – that of the whole system.

But, Cummins later makes a serious mistake in his attempt to circumvent 'Ryle's Regress'. Cummins is correct in noting that each newly posited function will be simpler, but he is incorrect in assuming, as he does, that these simpler functions may be implemented however we see fit (Cummins 1980). Cummins has erroneously assumed that preserving decompositional functionalism preserves computation-representation and metaphysical functionalism. As noted earlier, there is an important distinction to be made between these types of functionalism; the first is a methodology, where the latter two are theories of mind. As a methodology, decompositional functionalism provides a means of explaining the behavior of a system in simpler and simpler theoretical terms, so Cummins is right thus far. As a theory of mind, however, computation-representation and metaphysical functionalism wrongly entail that the implementation of a functional story is irrelevant to capturing the various aspects of behavior. Even though decompositional functionalism gets to simpler components, it only allows us to stop Ryle's regress at a sufficiently simple component if that component has a realization that preserves the functional description exactly. Once we have an implementable simplest function, there are no more questions as to what functions can be realized by that implementation, unlike in the case of a theoretical characterization. In other words, it is by no means clear that we have any consistent way of constraining the possible functions of a system without reference to an underlying physical mechanism.

Chris Eliasmith

__What's the logical relation between functionalism and multiple realizability__?

In much discussion in the philosophy of mind, the theses of functionalism and multiply realizability get mentioned in the same breath so often that it's not clear that they are separate theses, and, if so, what the logical relations between the two are. In this brief essay, I approach the issue by discussing what it takes to be functional and multiply realizable kinds in general. I will argue that the two theses are indeed different theses and further,that neither entails the other.

1. Functionalism does not entail multiple realizability.

Suppose we were to fix the extension of the predicate 'x is a blorg' as follows.

x is a blorg just in case when x collides with a hydrogen atom, then x is caused to move in a trajectory precisely perpendicular to the hydrogen atom's original trajectory.

Given that the reference of 'blorg' is fixed in the above way, it seems that blorgs comprise a functional kind: their defining features are causal relations. It might turn out though, that there's only one set of physical properties whose instantiation is sufficient to instantiate a blorg, say, being a molecule of PQR. So, even though blorgs are individuated by causal relational properties, it turns out that it is impossible to physically realize them in any way but one. Thus they are non-multiply realizable functional kinds.

2. Multiple realizability does not entail functionalism.

A kind can be multiply realizable even though it is not functionally individuated. For examples of multiply realizable but non-functional kinds, we need to look to kinds whose individuating relations are non-causal or to kinds whose individuating properties are not even relational. One example of the former might be: being outside of my light cone. Anything outside of my light cone is too far away from me to be enjoying any causal commerce with me, so the individuating relations in question are non-causal. Another example would be jade. Two chemically distinct materials are sufficient to realize jade: jadite and nephrite. But being a sample of jade isn't individuated by the kind of causal role it plays (in contrast to, say, a mouse-trap). Instead, something is a sample of jade just in case it has the chemical composition of jadite or nephrite. Sticking with examples from chemistry, we might say that being a chemical element is realized by diverse physical properties (at least as many as there are chemical elements) but it is not individuated by the kinds of causal relations that samples of chemicals bear to one another. For another class of examples, consider geometrically individuated kinds, like squares, cubes, circles, and spheres. Whether something counts as a sphere depends on whether it is a three dimensional object whose outermost points are equidistant from its center. Something can be a sphere without being causally related to anything else, so spheres are not functional kinds. And spheres can be realized by physically diverse sets of properties: spheres can be made of gold, protein, or wood; they can be solid or hollow; they can have any of a variety of masses and diameters; etc. Thus, we have another example of a multiply realizable kind.

It seems, then, that being functionally individuated and being multiply realizable are logically distinct in that neither entails the other.

Therefore, the thesis that mental states are functionally individuated is logically distinct from the thesis that mental states are multiply realizable.

Pete Mandik