Fine-grained supervenience, cognitive neuroscience, and the future of functionalism
Philosophy and Cognitive Science, William Paterson University, New Jersey
The majority of contemporary philosophers of mind are physicalists. The majority of physicalists, however, are non-reductive physicalists. As nonreductive physicalists, these philosophers hold that a system's mental properties are different from a system's physical properties, that is, they hold that the sum total of mental facts about some system is a different set of facts than the sum total of physical facts about the same system. As physicalists, however, these nonreductivists hold that mental facts are nonetheless determined by physical facts, that is, they subscribe to the supervenience thesis, i.e., the thesis that no mental differences can obtain without physical differences obtaining. In this paper I take up the issue of how best to understand the notion of supervenience, especially in the light of recent advances in the neurosciences.
Advocates of the supervenience thesis, when supplying a formulation of the thesis that is more detailed than the "no differences" slogan, supply formulations that conform to the following schema:(S)
If two situations differ in the mental properties they instantiate, then they must differ in the physical properties that they instantiate.
Construing "situations" as individuals, spatial regions, or entire possible worlds yields what have come to be known as "local," "regional," and "global" supervenience, respectively. Construing S as applying to situations in a single possible world or across possible worlds yields "weak" and "strong" supervenience, respectively.<1>
To date, advocates of the supervenience thesis have overlooked another way of construing the "no differences" slogan. The version that I have in mind is more fine-grained than S, and I shall refer to it as "Fine-Grained Supervenience". Versions of supervenience of this latter type conform to the following schema:(FS)
If a single situation instantiates two or more different mental properties, it must instantiate two or more different physical properties.
In this paper I argue that Fine-Grained Supervenience must not remain overlooked by philosophers of mind interested in the notion of supervenience. First and foremost, Fine-Grained Supervenience has quite a bit of empirical warrant, a small sample of which I present in this paper. I present considerations from cognitive neuroscience that lend evidential support to the thesis of Fine-Grained Supervenience. A second reason that Fine-Grained Supervenience deserves the attention of philosophers of mind is that it has serious ramifications for the widely held thesis of functionalism. In this paper I argue that if the Fine-Grained Supervenience thesis is true, then functionalism must be false.
The notion of Fine-Grained Supervenience, then, is doubly important for philosophers of mind interested in current trends in the neurosciences. Fine-grained supervenience not only gathers its plausibility from neuroscientific considerations, but it can serve as a powerful weapon against those philosophers who have argued against the relevance of neuroscience for theorizing about the mind. Functionalism entails the thesis of the multiple realizability of the mental by the physical.<2> And some philosophers, most notably Jerry Fodor<3>, have argued against the relevance of the neurosciences for psychology based on considerations of multiple realizability. But, if functionalism turns out to be false, then Fodorian considerations against the relevance of neuroscience are rendered largely ineffectual.
My paper is organized as follows. In section 1 I further unpack the general notion of supervenience. In section 2 I discuss the empirical warrant for Fine-Grained Supervenience. In section 3 I present arguments for the incompatibility of Fine-Grained Supervenience and functionalism.
1. Concepts of Supervenience
A set of properties or facts M supervenes on a set of properties or facts P if and only if there can be no differences in M without there being differences in P.
An example of supervenience is given in the relations between the acceleration, velocity, and position of an object in space. An object cannot change it's acceleration without changing it's velocity, and in turn, cannot change it's velocity without changing it's position. Thus, facts about an object's acceleration supervene on facts about an object's velocity which in turn supervene on facts about an object's position. The example of moving objects illustrates an important feature about supervenience, namely, that M can supervene on P even though M is not identical to P For example, an object's acceleration supervenes on its velocity even though acceleration is not identical to velocity. That supervenience allows property determination without property identity explains its wide-spread appeal to nonreductive physicalists who hold that a system's mental properties are not identical to a system's physical properties. Nonetheless, supervenience is compatible with the identity of super- and subvenient properties, thus providing a common ground for reductive and nonreductive physicalists alike. The example of moving objects illustrates another important feature of supervenience, namely, that supervenience is transitive.<4> I exploit the transitivity of supervenience in my argument against functionalism in 3. But first I turn to consider the supervenience thesis.
The supervenience thesis states that mental properties and facts supervene on physical properties and facts. The supervenience thesis can be further unpacked as the following three theses about objects and their properties. (Please note that I use the term "object" as a place holder for the unwieldy "object, event, or state of affairs" and the term "properties" to denote extrinsic as well as intrinsic properties.)
(i) No two objects can differ in their mental properties without differing in their physical properties.
(ii) A single object cannot change its mental properties without changing its physical properties.<5>
(iii) If, at a given time t, a single object has two different subsets of mental properties, it must have two different subsets of physical properties.
The first two theses above are common formulations of psychophysical supervenience.<6> The third has not previously appeared in the literature and is the thesis that I have been calling "Fine-Grained Supervenience". I shall hereafter use "the supervenience thesis" to refer to all three of the above corollaries, coarse- and fine-grained alike.
I turn now to consider the empirical warrant for corollary (iii), i.e. Fine-Grained Supervenience .
2. Cognitive Neuroscience: Empirical Warrant for Fine-Grained Supervenience
Cognitive neuroscience is comprised of a set of research strategies for determining how the central nervous system functions to give rise to psychological phenomena. Two aspects of cognitive neuroscience that are especially important for the purposes of this paper are the following. First, much of cognitive neuroscientific research involves the localization of different psychological functions to different regions of the brain. Second, cognitive neuroscientific explanations of how individual brain regions give rise to psychological functioning appeal to computational or information processing models of how neurons interact. These models are typically, if not universally, connectionist or parallel-distributed-processing models of computation. These two different aspects of cognitive neuroscientific research are 'localizationist' and 'connectionist', respectively, and both lend support to the thesis of Fine-Grained Supervenience. The localization of different psychological functions to different brain regions provides a very obvious way in which psychological differences of a single individual are given rise to in virtue of physical differences, since the differences between different regions of the brain are physical differences par excellence. I take it that being a different region of a brain is sufficient for being a different physical property of a brain. It is not necessary, however. There are instances in which different mental properties--say, two different mental representations--will be associated with a single region of the brain. That such a situation may obtain is no threat to the thesis of Fine-Grained Supervenience, however, since our best current ideas about how a single brain region might instantiate several mental representations are connectionist. And as I shall argue below, connectionism supplies all the resources needed to show physical differences for all representational differences that might obtain in a single network.
In the next section, I briefly describe an example of localization to give the flavor of what such a research strategy involves. In the section after that I discuss how connectionist models of neural computation show that even when different mental properties are instantiated in the same brain region, they are nonetheless accompanied by physical differences.
2.1. Localization of Object Vision and Location Vision
A classic example of localization of psychological function in different brain regions is found in Mishkin et al's (1983) localization of 'object' and 'location' vision in inferior temporal (IT) and posterior parietal (PP) cortex, respectively.
Monkeys with only IT lesions and monkeys with only PP lesions were tested on the following pair of tasks. The first task--the object task--involved learning under which of two different objects a food reward was hidden. The objects differed in their shape and color. The locations of the objects were changed randomly from trial to trial, but the food would always be located under an object that the monkey had not seen previously (this is an example of a delayed non-match to sample task). The second task--the location task--involved learning which of two locations a food reward was hidden at. In this second task, the objects that covered the food wells differed in no way other than their locations. Monkeys with lesions only IT lesions performed significantly more poorly on the object task than on the location task, whereas monkeys with only PP lesions performed significantly more poorly on the location task than on the object task.
According to the theses of Fine-Grained Supervenience (FS), if an individual instantiates two different mental properties, then that individual must instantiate two different physical properties. That there are different brain regions responsible for processing 'what' and 'where' visual information supports FS as follows. A visual perceptual mental event, such as seeing a rose on the table, is comprised of the visual recognition of the rose and the visual identification of the location of the rose as being on the table. And these separate perceptual capabilities--recognizing a particular object and seeing it in a different location--depend on the intact neural mechanisms in separate regions of cortex: regions in the temporal and parietal lobes, respectively.
2.2. Connectionism: Distributed Representations
I have been offering instances of the localization of different psychological functions in different brain regions as evidence for Fine-Grained Supervenience, i.e., the claim that an individual's instantiating different mental properties entails instantiating different physical properties. We can imagine an objection to the use of this evidence on the grounds that if a brain region is a single network of neurons that computes in accordance with connectionist principles, then different mental properties, e.g., different mental representations, would be instantiated by the same physical property, i.e., that single neural network. However, such an objection has no force against Fine-Grained Supervenience. I argue below that connectionism provides everything needed to show that different representations in a single network will nonetheless be accompanied by different physical properties of that network.
Assuming that connectionist networks have representations (as my imagined objector does) connectionist networks have both of two different kinds of representations: (i) patterns of activation across neural units of the network and (ii) patterns of connection weights between the units.<7> These two kinds are the occurrent and abeyant representations in a network, respectively.<8> Occurrent representations, like thoughts and percepts, are typically tokened only for relatively short periods of time. On the other hand, abeyant representations are the long-term memory and stored knowledge of the network. These latter representations are the products of training and are tokened or present for a relatively long duration.
In the case of different occurrent mental representations instantiated in a connectionist net, it is quite easy to see that the different physical properties that instantiate the occurrent representations just are the different patterns of activation that the network is capable of having. Verifying that there is a one-to-one mapping between a network's possible occurrent representations and its possible states of activation is a relatively simple exercise. However, things may not appear to be so simple in the case of the network's abeyant representations.
A network's abeyant representations are stored representations that are distributed throughout the sum total of the connection weights between neural units. The problem of telling how many abeyant representations are stored in a set of connection weights may seem intractable at first, since the pattern connection weights is spatially distributed through the net. However, this intractability is merely apparent. The key to determining what abeyant representations are stored in a set of connection weights is to realize that the physical properties of a network that instantiate its different abeyant representations are that network's different causal dispositions. And differences in causal disposition are physical differences par excellence regardless of whether they are spatially distributed. For examples of distributed yet different dispositional differences, consider the dispositional properties of a sugar cube. It is both soluble and breakable, and it's solubility and fragility are distributed throughout its cubic volume. Nonetheless, that the sugar cube's solubility and fragility are distributed is no bar to their being different physical properties. For another example, consider a lump of steel. The lump instantiates two different physical properties: having gravitational mass and being ferro-magnetic. Both properties are totally distributed throughout the volume of the lump. But they are nonetheless different properties. And further, they are causal dispositional properties. The difference between being gravitationally massive and being ferro-magnetic is just the difference being disposed to act one way in a gravitational field and another way in an electromagnetic field.
Returning to the abeyant representations of connectionist networks, we can see that they are individuated dispositionally as follows. Suppose that we attribute to a network like NETtalk the following separate bits of knowledge: it knows how to pronounce "through"; it knows how to pronounce "trough"; it knows how to pronounce "abeyance"; and so on. The attribution to the network of each of these separate bits of knowledge depends on the network's having corresponding causal dispositions. For instance, NETtalk knows how to pronounce "abeyance" just in case the activation pattern at that input layer that constitutes the orthographic input of "abeyance" causes NETtalk to token the pattern of activation in its output layer that constitutes its pronouncing "abeyance". Similarly for a trained up Rock/Mine network: it knows a rock from a mine just in case it is causally disposed to output "rock" in response to sonar profiles of rocks and "mine" in response to sonar profiles of mines.
If my above remarks about abeyant and occurrent representations in connectionist nets are correct, then it turns out that connectionism supports, instead of providing counter examples to, Fine-Grained Supervenience.
Thus concludes my discussion of the empirical warrant for fine-grained supervenience. I turn now to consider some ramifications that fine-grained supervenience has for a prominent topic in the philosophies of mind and psychology: functionalism.
3. Fine-Grained Supervenience and Functionalism
Functionalism is the view that the only properties essential to the instantiation of some mental state are the causal relations that state bears to other states.<9> Internalist versions of functionalism hold that the other relata need only be states internal to the subject of the mental state, e.g., states such as sensory input states, motor output states, and other mental states. Externalist versions of functionalism, including versions known as teleological functionalism, add certain environmental states to the list of essential relata.<10>
Distinctive of all versions of functionalism is an insistence upon the psychological irrelevance of certain lower levels of physical organization . For instance, a functionalist may hold as irrelevant to mentality facts peculiar to biochemistry, or, at a slightly higher level of grain, facts peculiar to neurobiology. While functionalists may differ as to what levels of grain they hold to be psychologically irrelevant, I will, for the sake of simplicity, assume a version of functionalism whereby only levels lower than the neuronal are taken to be irrelevant. Thus the version of functionalism with which I am concerned is one that holds that two entities that differ in their biochemistry can nonetheless token qualitatively identical mental states so long as they have qualitatively identical functional organizations at the neuronal level.
It is worth repeating that the versions of the multiple realizability thesis and functionalism I discuss are such that functionalism entails multiple realizability. Since I'm construing functionalism as the thesis that only causal relational properties at or above the neuronal level are necessary for the instantiation of any mental property, it follows that properties at lower levels of organization are psychologically inessential and thus replaceable without detriment to the instantiation of the mental.<11> I further assume, but shall not argue for, the wide-spread but debatable view that the multiple realizability of the mental by the physical entails the nonreducibility of the mental to the physical.<12> Thus, functionalism, by entailing multiple realizability, also entails nonreductivism.
In the argument that follows, I show that the supervenience thesis entails the falsity of functionalism. Since functionalism entails multiple realizability, the following argument removes a major reason for adhering to the multiple realizability thesis. It am not claiming, however, that the falsity of functionalism entails the falsity of multiple realizability. That would be denying the antecedent, to be sure. I am only claiming that without functionalism, the advocate of multiple realizability is deprived of one of her most compelling arguments for her position.
Philosophers opposing functionalism have offered thought experiments in which the causal relations alleged to be sufficient for the instantiation of some mental event are instantiated in ways supposedly too bizarre to succeed. Block's Chinese Nation (Block 1978) and Searle's Chinese Room (Searle 1980) arguments are two examples. In each, the requisite functional relations are instantiated by the activities of people--in Block's case, a whole nation of them, and in Searle's case, just one. The authors of these anti-functionalist arguments claim that it is implausible to suppose that the activities of these people suffice to instantiate mental properties other than those had by the individual people in the first place. In Searle's famous case, Searle himself is imagined to have memorized and run through all the steps of a Chinese-understanding computer program without himself understanding Chinese. Thus, says Searle, he instantiates the alleged functional structure of the mental event of understanding Chinese without actually instantiating the mental event in question. Likewise in Block's case, a whole nation of people, each obeying instructions relayed via walkie-talkie, instantiate the functional structure in question without, Block asserts, instantiating the mental event in question.
A popular functionalist response to these sorts of attacks has been to bite the bullet and maintain, contrary to intuitions such as Block's and Searle's, that such systems would instantiate the mental events in question.<13> Thus, according to the bullet-biting functionalist, Searle's rote rule following does give rise to genuine Chinese-understanding even though Searle himself seems not to understand Chinese. Likewise, the functionalist claims, Block's nation's walkie-talkie facilitated activities instantiate a mental event over and above the mental events of the individual citizens.
Such functionalist responses constitute an advocacy of mental-mental supervenience, which is to say that they allow for the possibility of a situation in which one mind or set of mental facts supervenes on another. In the Chinese Room scenario, the Chinese understanding mind supervenes on John Searle's mind. In the Nation case, a high level "group mind" supervenes on the conjunction of activity of each of the citizens, which includes the mental activity of those citizens. The unitary group mind supervenes on the minds of the individual citizens because in order to successfully instantiate the higher mind, it is important that the citizens are sufficiently intelligent to understand the directions they convey to each other over walkie-talkie. And for the instantiated group mind to, say, change it's mind, some difference of the activity of the citizens would be required. Similarly, in Searle's case, the difference between the instantiated Chinese understanding mind tokening one thought rather than another is a difference that entails a difference in what rule governed activities Searle is engaged in.
Functionalist advocacy of mental-mental supervenience is not limited to considerations of fanciful thought experiments such as Block's and Searle's, but is also a prevalent feature of functionalist theorizing about the way human minds are typically instantiated. For example, Homuncular Functionalism (AKA Homunctionalism) as advocated by Lycan (1987) and Dennett (1978) , also involves mental-mental supervenience. According to the Homunctionalists, a human mind taken at the personal level is decomposable into a handful of sub-personal homunculi each of which are decomposable into further homunculi. At each level of decomposition the units at that level have genuinely mental properties, but of a stupider, simpler sort than those found at the levels above. The recursive decomposition bottoms out with units so simple and stupid as to succumb to wholly mechanistic and non-psychological explanations. But any two adjacent levels that both contain homunculi offer examples of mental-mental supervenience. Consider the personal level and the first sub-personal level of homuncular decomposition right below it. The personal level, which contains one mind, supervenes on the next level down, which consists of many homuncular minds and the interactions between them. The Homuncular Functionalist view of the mind is analogous to the bullet biting functionalist construal of Block's Chinese Nation insofar as both hold that a high level "group" mind can supervene on a multitude of lower level minds. Mental-mental supervenience surfaces in the functionalist literature whenever Homunctionalism or the aforementioned bullet biting does.
The possibility of mental-mental supervenience follows from functionalism. According to functionalism, all that matters to the instantiation of a mental event is that causal relations obtain between parts of the realizing system. It follows then that (i) the parts of the realizing system can have their own mental properties and (ii) the causal relations between parts of that system can be instances of mental causation.
The possibility of mental-mental supervenience, however, poses a serious threat to theorists subscribing to the conjunction of the supervenience thesis and functionalism, because the possibility of mental-mental supervenience leads to a reductio ad absurdum of that conjunction. The key to the reductio is the fact that the possibility of mental-mental supervenience contradicts the supervenience thesis. Briefly, the supervenience thesis states that no mental differences can obtain without physical differences obtaining, but the possibility of mental-mental supervenience is the possibility of mental differences obtaining without physical differences obtaining.
Instances of mental-mental supervenience are instances in which two different sets of mental facts share a supervenience base. To see this, consider the following. We can describe an instance of mental-mental supervenience as a case in which the sum of the mental activity, M, occurring at some level supervenes on the sum of lower level mental activity, M'. Given the supervenience thesis, M' ultimately supervenes on the sum of activity, P, occurring at some wholly physical level. And given the transitivity of supervenience, both M and M' supervene on P.
However, to posit that M and M' both supervene on P is to contradict the supervenience thesis. According to the supervenience thesis, no mental differences can obtain without physical differences obtaining. Therefore, if M and M' are different sets of mental facts and M supervenes on P then M' cannot supervene on P but must instead supervene on some other set of physical facts, P'. Thus, mental-mental supervenience, which is entailed by functionalism, is forbidden by the supervenience thesis, especially by corollary (iii): Fine-Grained Supervenience..
In this paper I have presented empirical warrant for a previously unnoticed formulation of psychophysical supervenience: Fine-Grained Supervenience. I have also sketched an argument against functionalism that takes Fine-Grained Supervenience as one of its premises. Since Fine-Grained Supervenience draws support from the results and techniques of neuroscience, and also casts doubt on a popular philosophical argument against the relevance of the neurosciences, Fine-Grained supervenience is a thesis that deserves the attention of philosophers of mind interested in the neurosciences.<14>
<1> See Chalmers 1996, and Kim 1993 for discussions
of the different types of supervenience mentioned in this paragraph.
<2> See, for example, Baker 1995.
<3> Fodor 1974.
<4> See Lycan 1987: 103 and Kim 1993: 67 for
mention of the transitivity of supervenience.
<5> That is, if an object at time t has one
subset of M properties (or M facts true of it) and at time t+n has a different
subset of M properties (or M facts true of it), then that object must have
two different subsets of P properties at those two different times.
<6> (i) and (ii) are adapted from Davidson 1970.
<7> Haugeland 1991: 84.
<8> Churchland and Sejnowski 1992: 165.
<9> Some versions of functionalism require that
the relevant causal relations be individuated computationally and/or teleologically.
I take everything I say in this paper about functionalism to apply to these
versions as well.
<10> For the sake of simplicity I restrict
my discussion here to internalist functionalism. For longer versions of
the present argument in which I discuss externalist functionalism see Mandik
1996 and Mandik 1997.
<11> Other formulations of these theses may
not retain the aforementioned implication relation, but I think it safe
to say that versions of functionalism that are compatible with the falsity
of the multiple realizability thesis are indistinguishable in principle
and practice from their historical physicalistic opponent, the psychoneural
type-identity theory of defended by U. T. Place 1956 and J .J. C. Smart
<12> For the view that it does, see Fodor 1974
. For the view that it does not, see Kim 1992 .
<13> See Chalmers 1996 for a recent example.
<14> I am grateful to Tom Polger, John Post,
and Jesse Prinz for helpful discussions of the material in this paper.
Baker, L. R. (1995). Functionalism. The Cambridge Dictionary of Philosophy. R. Audi, Cambridge University Press: 288-9.
Block, N. (1978). Troubles with functionalism. Minnesota Studies in the Philosophy of Science. 9: 261-325.
Chalmers, D. J. (1996). The Conscious Mind: Toward a Theory of Consciousness. Oxford, Oxford University Press.
Cuda, T. (1985). Against neural chauvinism. Philosophical Studies 48: 111-127.
Davidson, D. (1970). Mental events. Experience and Theory. L. Foster and J. Swanson. Amherst, Mass., University of Massachusetts Press: 79-101.
Dennett, D. C. (1978). Brainstorms. Cambridge, MA, MIT Press.
Fodor, J. A. (1974). Special sciences, or the disunity of science as a working hypothesis. Synthese 28: 97-115.
Kim, J. (1992). Multiple realization and the metaphysics of reduction. Philosophy and Phenomenological Research 52: 1-26.
Kim, J. (1993). Supervenience and Mind. Cambridge, Cambridge University Press.
Lycan, W. G. (1987). Consciousness. Cambridge, MA, MIT Press.
Mandik, P. (1996). Consciousness, content, and causal structure: An essay in the metaphysics of mind. Unpublished
Mandik, P. (1997). Supervenience, self-reference, and anti-functionalism. Unpublished.
Mishkin, M., Ungerleider, L. G., and Macko, K. A. (1983). Object vision and spatial vision: Two cortical pathways. Trends in Neurosciences, 6, 414-417.
Place, U. T. (1956). Is consciousness a brain process? British Journal of Psychology 47: 44-50.
Putnam, H. (1975). The meaning of `meaning'. Minnesota Studies in the Philosophy of Science 7: 131-193.
Pylyshyn, Z. (1980). The `causal power' of machines. Behavioral and Brain Sciences 3: 442-444.
Savitt, S. (1982). Searle's demon and the brain simulator reply. Behavioral and Brain Sciences 5: 342-343.
Searle, J. R. (1980). Minds, brains, and programs. Behavioral and Brain Sciences 3: 417-24.
Searle, J. R. (1992). The Rediscovery of the Mind, MIT Press.
Smart, J. J. C. (1959). Sensations and brain processes. Philosophical Review 68: 141-56.
Tye, M. (1995). Ten Problems of Consciousness. Cambridge, Mass., MIT Press.