[1]
This, for example, is one way of construing Dennett’s instrumentalism (1978, 1987).
[2]
There has been a some discussion in the literature about whether the relational properties of brain states are implicated in the causation of behaviour. The standard way of defending this claim is by individuating behaviour broadly, so as to incorporate factors beyond bodily movements (Burge 1986; Wilson 1994). But many philosophers think this form of individuation does great violence to scientific practice in general and to neuroscience in particular, and hence this way of resolving the problem of content causation is thought to seem very unpromising (Fodor 1987).
[3]
Perhaps the best we can do, according to some of these, is to accept that the representational properties of mental phenomena are causally inert, but to argue that there is enough room between explanation and causation for representational properties to be explanatorily relevant—despite their inertness (Baker 1993; Block 1989; Fodor 1986, 1989; Heil & Mele 1991; Jackson & Pettit 1990a, 1990b; LePore & Loewer 1989). A more radical response is to opt out of representation-based explanation altogether, as advocated originally by eliminativists (Churchland 1981; Stich 1983), and more recently by anti-representationalists (Brooks 1991). Finally, note that another radical position currently fashionable in philosophy—the extended-mind hypothesis (Menary 2010)—doesn’t represent a solution to the content causation problem, since it signally fails to align mental phenomena with the brain-based causation of behaviour.
[4]
Von Eckardt actually uses the terms “representation bearer”, “representational object”, and “interpretant” to describe the three relata implicated in representation. I prefer the terminology I have used here because it is more consistent with the philosophical literature on mental representation.
[5]
Dretske scholars will cry foul at this point, of course. This is because Dretske claims that while indication is mostly founded on causal relations, it need not be. Indeed, he goes as far as to suggest that indication obtains whenever there is a non-coincidental covariation between vehicle and object (Dretske 1988, pp. 56–57). But this characterisation of indication transforms Dretske’s proposal into something close to a resemblance theory (the approach to be examined in the next section), since it privileges systematic correspondence relations over causal relations. Consequently, insofar as Dretske’s position is to be understood as a causal theory of content determination (as is widely assumed in the literature), it is essential that indication is interpreted as a relation of causal covariation. I adopt this interpretation in what follows.
[6]
One would expect to find causal relations mediating systematic correspondence relations between the representing vehicles of biological systems and aspects of the world. But, as Dretske is well aware, this is not always the case. Nature will make do with what works, and some kind of systematic correspondence in the absence of causal commerce will do just as well. This can be illustrated by another of Dretske’s favourite examples: the evolutionary recruitment of magnetosomes in anaerobic bacteria to steer them towards deoxygenated water (1986). According to Dretske, evolutionary forces operating on these bacteria have selected magnetosomes because they are indicators of anaerobic water capable of influencing the direction in which the bacteria swim. But as Millikan has pointed out, the connection between the orientation of magnetosomes and anaerobic water is merely correlational, not causal (2004, Ch. 3). Magnetosomes indicate and steer northern hemisphere anaerobic bacteria in the direction of magnetic north, which results in these bacteria swimming into deeper (and hence deoxygenated) water. But there is no causal connection between magnetic north and deoxygenated water. In this case, therefore, magnetosomes have been selected because their alignment systematically corresponds with the direction of anaerobic water, not in virtue of any causal covariation between them.
[7]
Cummins reaches a similar conclusion, though via a somewhat different route (1996, p. 74).
[8]
I am here adapting terminology used by Shepard & Chipman (1970).
[9]
To be more precise, suppose SV = (V, V ) is a system comprising a set V of objects, and a set V of relations defined on the members of V. The objects in V may be conceptual or concrete; the relations in V may be spatial, causal, structural, or inferential, and so on. For example, V might be a set of features on a map, with various geometric and part–whole relations defined on them. Or V might be set of well formed formulae in first-order logic falling under relations such as identity and consistency. There is a second-order resemblance between two systems SV = (V, V ) and SO = (O, O ) if, for at least some objects in V and some relations in V , there is a one-to-one mapping from V to O and a one-to-one mapping from V to O such that when a relation in V holds of objects in V, the corresponding relation in O holds of the corresponding objects in O. In other words, the two systems resemble each other with regard to their abstract relational organisation. As already stressed, resemblance of this kind is independent of first-order resemblance, in the sense that two systems can resemble each other at second-order without sharing properties. Second-order resemblance comes in weaker and stronger forms. As defined it is relatively weak, but if we insist on a mapping that takes every element of V onto some element of O, and, in addition, preserves all the relations defined on V, then we get a strong form of resemblance known as a homomorphism. Stronger still is an isomorphism, which is a one-to-one relation-preserving mapping such that every element of V corresponds to some element of O, and every element of O corresponds to some element of V. When two systems are isomorphic their relational organisation is identical. In the literature on second-order resemblance the focus is often placed on isomorphism (see e.g., Cummins 1996, pp. 85–111), but where representation is concerned, the kind of correspondence between systems that is likely to be relevant will generally be weaker than isomorphism. For a much fuller discussion of second-order resemblance, see O’Brien & Opie (2004).
[10]
Two early theorists who sought to apply second-order resemblance to mental representation are Palmer (1978) and Shepard (Shepard & Chipman 1970; Shepard & Metzler 1971). More recently, Blachowicz (1997), Cummins (1996), Gardenfors (1996), O’Brien & O’Brien (1999), O’Brien & Opie (2004), and Swoyer (1991), have all defended second-order resemblance theories.
[11]
Notice that in this case, the second-order resemblance is sustained structural relations among the set of representing vehicles (i.e., the set of bi-metallic strip curvatures). This is an example of what Palmer (1978) calls natural isomorphism, since the second-order resemblance relations are sustained by constraints inherent in the vehicles, rather than being imposed extrinsically. Elsewhere I have used the term structural resemblance to describe this kind of second-order relationship and to distinguish it from functional resemblance, where the second-order resemblance relations are sustained by causal relations among the vehicles—see O’Brien & Opie (2004).