Causal theories of mental content determination have dominated philosophy for nearly half a century. They hold that representing vehicles are contentful in virtue of being (actually, nomologically, or historically) caused by their represented objects (Devitt & Sterelny 1987; Fodor 1984, 1987, 1990; Stampe 1977, 1986). Perhaps the most well-known causal theory in all of the literature has been developed, through a number of iterations, by Dretske (1981, 1988, 1995). What makes Dretske’s account particularly apposite in the current context is that it has been fashioned, at least in its later iterations, to address explicitly the account of content intruded by the triadic analysis of mental representation (though Dretske doesn’t use this terminology). At one point in his discussion, for example, Dretske states that he approves of Armstrong’s (1973) description of beliefs as “maps by which we steer”, and goes on to observe that “beliefs are representational structures that acquire their meaning, their maplike quality, by actually using the information it is their function to carry in steering the system of which they are part” (Dretske 1988, p. 81). This, for Dretske, motivates the very desideratum we extracted from the triadic analysis in the last section:
It will not be enough merely to have a C [inner state of some cognitive system] that indicates F [i.e., causally covaries with some external condition] cause M [some observable behaviour]. What needs to be done […] is to show how the existence of one relationship, the relationship underlying C’s semantic character, can explain the existence of another relationship, the causal relationship (between C and M) comprising the behaviour in question. (1988, p. 84)
Dretske’s response to this problem, famously, is to appeal to teleology. It is only when an inner state, which causally covaries with some bit of the external environment, is “recruited” by the cognitive system (either by an evolutionary design process or through individual development) to cause appropriate behaviour, that the state acquires the function of indicating that part of the environment, and thereby comes to represent it (Dretske 1988, pp. 84–89).
On the face of it, Dretske’s theory seems to represent a promising solution to the content causation problem. From the perspective of the triadic analysis, a solution to this problem requires an explanation of how certain relations between mental vehicles and their objects can dispose cognitive subjects to behave selectively towards those represented objects. Dretske’s elegant proposal is that reliable causal relations between inner states and environmental conditions (i.e., when the latter reliably cause the former to be tokened) can endow cognitive systems with these dispositions when the former states are conscripted by design processes to cause behaviour that is in some way relevant to the latter conditions. When this happens, the inner states are elevated to the status of representing vehicles, and their subsequent activity in bringing about behaviour directed towards their represented objects are examples of content causation.
Unfortunately, a closer inspection of Dretske’s suggestion reveals a fundamental flaw. Contrary to what he contends, the relations at the core of his proposal are powerless to explain the required behavioural dispositions. Rather than describing this problem in the abstract, let me illustrate it using one of Dretske’s favourite examples of a very simple representation-using system:
A drop in room temperature causes a bimetallic strip in [a thermostat] to bend. Depending on the position of an adjustable contact, the bending strip eventually closes an electrical circuit. Current flows to the furnace and ignition occurs. The thermostat’s behaviour, its turning the furnace on, is the bringing about of furnace ignition by events occurring in the thermostat—in this case […] the closure of a switch by the movement of a temperature-sensitive strip […].
The bi-metallic string is given a job to do, made part of an electrical switch for the furnace, because of what it indicates about room temperature. Since this is so, it thereby acquires the function of indicating what the temperature is […]. We can speak of […] representation here. (Dretske 1988, pp. 86–87)
There is a subtle sleight of hand at work here, however. It is Dretske’s contention that the bi-metallic strip is recruited (by the manufacturer) to play a causal role in the thermostat because of what it indicates about ambient temperature. But that’s not the full story. Bi-metallic strips have an additional property that appeals to the manufacturers of thermostats: their degree of curvature corresponds in an orderly fashion with ambient air temperature, such that it can be configured to complete a circuit when the temperature drops to a pre-set level.
In Dretske’s thermostat example, therefore, there are two distinct relations between representing vehicles and represented objects: a systematic correspondence relation (wherein variations in ambient air temperature are mirrored by orderly variations in the bi-metallic strip’s shape) and an indication relation (wherein variations in ambient air temperature cause variations in the bi-metallic strip).[5] These two relations are not independent of one another, of course, as the former is mediated by the latter. But we can still consider which of these relations is doing the work, insofar as the capacity for the thermostat to control the behaviour of the furnace is concerned. And here the answer is clear: it is the fact that the curvature of the bi-metallic strip systematically mirrors the temperature, and not the causal covariation per se, that explains its capacity to operate the furnace in an appropriate manner. Consider the counterfactuals: curvature correspondence without causal covariation (e.g., where a mere correlation exists) would still generate the appropriate behaviour, but causal covariation without curvature correspondence (e.g., where the bi-metallic strip heats up but maintains its shape) wouldn’t. The important point is that while the causal relation plays an important role in mediating the correspondence relation, it is the latter, not the former, that explains the thermostat’s capacity to bring about the desired behaviour.[6]
So Dretske’s own example fails to satisfy the desideratum that he set for himself: the obtaining of a reliable causal connection between ambient air temperature and the bi-metallic strip doesn’t explain the thermostat’s capacity to control the behaviour of the furnace. Moreover, this example illustrates a fundamental problem with all causal theories of mental content determination: there is a fatal disconnect between world-mind causal relations, on the one hand, and the mind’s behavioural dispositions on the other. This disconnect exists because any (actual, nomological, or historical) causal relations that might exist between external conditions and inner vehicles do not explain, in their own right, how a cognitive system inherits the capacity of behaving sensitively to the former. Whether cognitive systems have this capacity is determined by the properties of their inner vehicles in concert with their organizational, architectural, and motoric properties. And while external conditions can cause tokenings of and alterations to inner vehicles, the mere obtaining of such causal relations can’t explain how the tokened or altered vehicles are capable of interacting with these multifarious systemic properties such that they bestow the appropriate behavioural dispositions.[7] This is why manufacturers are very choosy about the materials from which they construct thermostats. Engineering a causal covariation relation between ambient air temperature and the innards of a thermostat is easy; engineering these innards such that they possess the requisite causal capacities is a great deal harder.
Ultimately, therefore, Dretske’s ingenious attempt to solve the content causation problem doesn’t succeed. Dretske holds that the internal states of cognitive systems are elevated to representing roles when they are recruited by design processes to regulate behaviour towards the external conditions they indicate. He takes this to be a case of genuine content causation because he thinks that the causal relations between represented objects and representing vehicles can explain the causal activity in which the vehicles subsequently engage. But Dretske has over-estimated the explanatory power of world–mind causal relations. And he has done so because he has illicitly smuggled into his story a quite distinct form of content determination—one that exploits systematic correspondence relations between representing vehicles and their represented objects. Such systematic correspondences are, of course, a species of resemblance relation. The failure of Dretske’s proposal is thus instructive, since it suggests that this alternative world–mind relation offers some prospect of a solution to the content causation problem.
Resemblance theories of content determination hold that representing vehicles are contentful in virtue of resembling their represented objects. The most obvious and straightforward application of this idea can be found in many public forms of representation, from photographs and paintings to sculptures and maps. But what is most significant about this approach for our purposes is that when vehicles resemble their objects, the former actually replicate the latter in some way, either by reproducing their properties or their relational organisation (more about which in the next section). And this affords a relatively straightforward way of explaining how a physical device, in virtue of incorporating vehicles that bear resemblance relations to the world, acquires a capacity to behave selectively towards particular elements of the environment. The thermostat’s bi-metallic strip reproduces—in the variations in its degree of curvature—the diachronic pattern of magnitude relations between ambient air temperature. Once this bi-metallic strip is incorporated into the thermostat, therefore, this device has a set of internal vehicles that dynamically replicates the external temperature. It is then simply a matter of rigging the innards of the thermostat so that its operation of the furnace is regulated by these internalised surrogates (Swoyer 1991).
Dretske is correct to judge this an example of content causation. It is a case in which the exploitation of a relation between environmental conditions and inner vehicles explains how the latter are capable of modifying a device’s behavioural dispositions towards particular aspects of the world. But what is seldom acknowledged about this much-used example is that it demonstrates the causal efficacy of content fixed by resemblance. Despite this virtue, resemblance theories of mental content determination are unfashionable in contemporary philosophy, largely because they are widely thought to suffer from a number of fatal flaws. Before we end, therefore, it would be wise to engage in a degree of resemblance rehabilitation. This turns out to be easier than one might expect, however, once we adopt the perspective of the triadic conception of representation.