Below is the unedited penultimate draft of:

This is the unedited penultimate draft of a BBS target article that has been accepted for publication (Copyright 1998: Cambridge University Press -- publication date provisional) and is currently being circulated for Open Peer Commentary. This preprint is for inspection only, to help prospective commentators decide whether or not they wish to prepare a formal commentary. Please do not prepare a commentary unless you have received the hard copy, invitation, instructions and deadline information.

For information about subscribing or purchasing offprints of the published version, with commentaries and author's response, write to: journals_subscriptions@cup.org (North America) or journals_marketing@cup.cam.ac.uk (All other countries).

A NEURON DOCTRINE IN THE PHILOSOPHY OF NEUROSCIENCE

Keywords

Churchlands, classical conditioning, cognitive neuroscience, Kandel, learning, materialism, mind, naturalism, neurobiology, neurophilosophy, philosophy of neuroscience, psychology, reduction, theoretical unification

Abstract

1. Introduction

Among those who reflect on the nature of neuroscience there is a view about its scope and limits which we will call, with a certain amount of historical license, the neuron doctrine.[1] Roughly, the neuron doctrine is the view that the framework within which the science of the mind will be developed is the framework provided by neuroscience; or, as we shall put it, that a successful theory of the mind will be a solely neuroscientific theory.[2] The idea is not of course that neuroscience will explain everything about the mind; perhaps there are aspects of the mind we will never explain. The idea is rather that, to the extent that we will achieve a scientific understanding of mental or psychological phenomena at all, neuroscience will be the science that achieves it. According to the neuron doctrine, in the race to explain the mind, smart money is on neuroscience.

There are at least three reasons for thinking that the neuron doctrine so described is important. First, to claim that it will one day explain the mind is obviously one of the more fundamental and ambitious claims that could be made about any science. If, as the neuron doctrine alleges, neuroscience will explain the mind, that would make it tremendously important to scientists and non-scientists alike. Any discussion of the foundations of neuroscience, therefore, must involve an assessment of the neuron doctrine.

Second, the neuron doctrine seems to be strongly supported by science and philosophy. This might be brought out in the following way.[3] Many scientists and philosophers adhere to the methodological view known as naturalism. According to naturalism, to the extent that we will be able to understand the world, it will be empirical science (and not, say, religion or philosophy) that provides that understanding. Similarly, many scientists and philosophers adhere to the metaphysical view sometimes known as materialism. Roughly, materialism holds that psychological events, states and processes are nothing more than events, states and processes of the brain.[4] Given these two views, and treating neuroscience by definition as the science of the brain, it seems inevitable that the neuron doctrine is true: if the mind is the brain, and if neuroscience is the science of the brain, then it is practically[5] a fact of logic that neuroscience is the science of the mind, and that it alone will explain what can be explained about the mind. Indeed, from this point of view, it is difficult to deny the neuron doctrine without sounding - as the philosopher Frank Jackson (1982) has put it in a related context - like someone who believes in fairies.

Finally, proponents of the neuron doctrine often suggest that their view apparently has important, and potentially devastating, consequences for our current practice of attempting to construct a scientific understanding of the mind, and perhaps even for intellectual domains further afield, such as the structure of scientific theories, the correct approach to the understanding of the social world, and the proper conception of morality, art and the self.[6] For on the face of it, neuroscience is not the science of the mind, as the neuron doctrine suggests, for the simple reason that it is not the only science of the mind. On the contrary, there are plenty of apparently non-neuroscientific disciplines, such as psychology, psychophysics, linguistics, ethology, and the like - sciences we will group together as the psychological sciences. If the neuron doctrine is true, what are we to make of them? For proponents of the neuron doctrine, the inevitable result of tracing out its consequences is that the psychological sciences must be relegated to a second-rate, or place-holder, status. Of course, one might fail, or simply refuse, to draw this consequence, but, from the point of view of the neuron doctrine, this could only constitute a failure of intellectual nerve and, anyway, does nothing to undermine the importance of the doctrine for the status of these fields. In short, the neuron doctrine seems to be a remarkable thesis, one with solid intuitive foundations but with a strikingly counter-intuitive result.

This paper is a critical examination of the neuron doctrine and the philosophy of neuroscience on which it is based, with a particular focus on the consequence for the psychological sciences that it apparently entails. Our central claim is that the doctrine suffers from a fatal ambiguity. Interpreted one way, the neuron doctrine is highly plausible and does find strong support in science and philosophy. But, on this interpretation, it fails to have the revolutionary consequence for the psychological sciences suggested by its proponents. Interpreted another way, the neuron doctrine is extremely interesting and would have this consequence, but we argue that there is little evidence that, on this interpretation, the doctrine is true. The problem with the neuron doctrine, we will claim, is that there is no way for it to be made both plausible and interesting.

We begin our examination of the neuron doctrine by asking who holds it and by considering in some detail the revolutionary consequence of doing so. In section 2, we distinguish two versions of the doctrine, one trivial and one radical, and we argue that these versions have not been clearly distinguished in the literature on neuroscience. We then turn to three arguments for the radical version of the doctrine, each prompted by scientific claims or views in the philosophy of science. The first argument, which we call the argument from naturalism and materialism (section 3), is based on a small number of highly plausible claims that we take most neuroscientists and philosophers to accept. The second argument, which we call the argument from unification (section 4), attempts to defend the neuron doctrine by appealing to considerations about the development of scientific theories. The final argument, which we call the argument from exemplars (section 5), looks to neuroscience itself for support of the radical version of the neuron doctrine. In order to evaluate this argument, we consider one case of neuroscientific theory at some length, namely, the theory of elementary learning in Aplysia due to Eric Kandel and his colleagues. These arguments do not exhaust all the ones that might be offered in defense of the neuron doctrine, but we have chosen to discuss these because they are all highly plausible, because they appeal to principles respected by neuroscientists and philosophers alike, and because they point up important conceptual features of neuroscience and its place among the sciences of the mind. We conclude, in section 6, with some remarks about the morals one might draw from our argument.

1.1. Who holds the neuron doctrine?

For the purposes of our discussion, we will take the chief proponents of the neuron doctrine to be the philosopher-neuroscientists Patricia and Paul Churchland. There are two reasons for this. First, as we shall see, the Churchlands are particularly clear and knowledgeable advocates of the doctrine.[7] Second, because neuroscientists themselves tend to be reticent about expressing meta-scientific commitments in anything other than popular or quasi-popular publications (some of which we canvass below), it has largely been left to the Churchlands to articulate in a technical way the status and commitments of neuroscience. Indeed, more than anybody else on the contemporary scene, the Churchlands are responsible for painting the portrait of neuroscience and for rightly drawing attention to its many successes. Their advocacy of the doctrine can therefore be reasonably taken as a reflection of a central, perhaps dominant, intellectual trend in the field as a whole.[8]

A clear statement of the neuron doctrine can be found at the beginning of Patricia Churchland and Terrence Sejnowski's (1992) book The Computational Brain:

One can also see a commitment to the neuron doctrine in the joint work of the Churchlands. In a discussion of how the relation of psychology to neuroscience is likely to compare to significant historical cases of intertheoretic reduction - the reduction of Kepler's planetary laws to Newton's laws of motion, the reduction of temperature to mean molecular kinetic energy, and of light to electromagnetic waves - the Churchlands (1994) say that in the neuroscience-psychology case,

Although our primary focus is the Churchlands, commitment to the neuron doctrine is by no means limited to them. In A Vision of the Brain, for example, Semir Zeki says:

1.2. A consequence of the neuron doctrine

According to some of its proponents, the importance of the neuron doctrine lies in what it implies about the various sciences of the mind. In Neurophilosophy, Patricia Churchland (1986) writes:

A similar suggestion is made by Hubel (1974):

It is important to see just what a strong consequence this is. If the neuron doctrine implies that any psychological theory of the mind is second-grade, or place-holder, science, we are faced with the problem of what to say about the many developed psychological theories we now have. Linguistics, to take one example, is among the most advanced sciences of any mental phenomenon. According to most linguists, the fact that every normal human being is linguistically competent is to be explained by our (largely unconscious and innate) knowledge of a system of rules and principles that assign semantic and syntactic interpretations to physical forms, whether those forms are heard, seen, or touched. Linguists and psychologists are in the process of describing this system of rules, its development in childhood, and its interaction with other cognitive systems. But according to the neuron doctrine, there is something misguided about this enterprise. For it is far from clear that the basic concepts of linguistic theory as it is currently understood can be reduced to the basic concepts of neuroscience as it is currently understood; indeed, many linguists believe they can not (Higginbotham 1990). So linguistics faces a dilemma: either it must be reformulated in neuroscientific terms, or else it must be judged a place-holder science. In Churchland's terminology, if the neuron doctrine is true, linguistics does not tell us what it really means to have knowledge of language; and in Hubel's terminology, linguistics is in the position of providing only an indirect account of the phenomena it seeks to explain. Given the relative maturity and complexity of linguistic theory, this is no trivial result.[13] 

Strong as this consequence is, the Churchlands quite explicitly accept it and regard it, in fact, as a major selling point of their view. Paul Churchland (1992, p. 109; see also P.M. Churchland 1995), for example, writes that the position he advocates holds out "the possibility of an alternative to, or potential reduction of, the familiar Chomskyan picture." Patricia Churchland is even more explicit. Churchland and Sejnowski (1992) write:

1.3. Evidence for the neuron doctrine

Of course, one would have to accept this consequence, revolutionary though it is, if the neuron doctrine were backed up by the best empirical results. If we currently had a mature neuroscience that could explain a wide range of mental phenomena, then we would have to admit that the interpretation of linguistics and the other psychological sciences was settled. Is it the case then that the facts are in, but nobody has bothered to tell the linguists and psychologists?

Of course the answer is "no." While we have a great deal of knowledge about the basic biology of the brain, it is only a slight exaggeration to say that we are almost completely ignorant about how the brain produces mental life. As we remarked above, Paul Churchland's optimism about current neuroscience should really be understood as hyperbole, and other neuroscientists are far more cautious. Hubel (1988), for example, says that

So we have a problem. On the one hand, the neuron doctrine has widespread support, in part because it seems to follow from widespread views in both philosophy and science. On the other hand, the neuron doctrine has a consequence concerning the effect of neuroscience on the psychological sciences that is not only radical but unsupported by neuroscience itself. The problem is how to explain the existence of a single doctrine that is simultaneously radical - and radical in the absence of evidence - as well as widespread.

1.4. Two questions

Our first question, therefore, is this: Given that the facts required to evaluate the neuron doctrine are not in, why do informed people believe it? Our answer to this question (set out in section 2) is that the neuron doctrine suffers from an ambiguity. Interpreted one way, the doctrine is very plausible but fails to have the radical consequence for the psychological sciences that we have discussed. Interpreted another way, the neuron doctrine is an empirical conjecture that does have this substantive consequence but is highly controversial and currently unsupported by the evidence. Our suggestion is that proponents of the neuron doctrine sometimes conflate these two very different interpretations of their view and believe as a result that there is a single view that is both obvious and revolutionary.

Our second question is this: Is there any scientific reason to believe the radical and interesting form of the neuron doctrine? We address this question in sections 3-5 by considering three arguments that might be offered in its defense.

2. Two versions of the neuron doctrine

We have expressed the neuron doctrine as the view that a successful theory of the mind will be a solely neuroscientific theory. What exactly does that mean? In the first part of this section, we argue that attention to this question leads to a distinction between two conceptions of neuroscience which in turn makes it possible to distinguish two versions of the neuron doctrine. We then provide evidence that proponents of the doctrine frequently conflate these two versions.

2.1. Two conceptions of neuroscience

According to one conception of neuroscience, perhaps the more traditional conception, neuroscience is to be understood as the science we will call biological neuroscience, the concern of which is the investigation of the structure and function of individual neurons, neuronal ensembles and neuronal structures. For simplicity, we will stipulate that biological neuroscience includes only neurophysiology, neuroanatomy, and neurochemistry, and we will take it to be synonymous with neurobiology.

According to another conception, neuroscience is taken to be what is often called cognitive neuroscience (see Gazzaniga 1995; see also Kosslyn and Andersen 1992, and Kosslyn and Koenig 1995), and we will adopt that name here. Cognitive neuroscience is an interdisciplinary approach to the study of the mind the concern of which is the integration of the biological and physical sciences - including in particular biological neuroscience - with the psychological sciences to provide an explanation of mental phenomena. Whereas biological neuroscience is interested in understanding the biology of the brain, cognitive neuroscience attempts to synthesize biology and psychology in order to understand the mind. Cognitive neuroscience thus includes biological neuroscience as a proper part but is not exhausted by it.

2.2. Versions of the neuron doctrine

With these conceptions of neuroscience before us, we can now distinguish two versions of the neuron doctrine. The two doctrines are generated by replacing 'neuroscience' in our general statement of the neuron doctrine above by 'cognitive neuroscience' and 'biological neuroscience.' We will call these versions the trivial neuron doctrine and the radical neuron doctrine, respectively.

2.2.1. The trivial doctrine. The trivial neuron doctrine is the view that a successful theory of the mind will be a solely cognitive neuroscientific theory. According to this doctrine, to the extent that psychological phenomena will be explained at all, the science that will do so is cognitive neuroscience. And since cognitive neuroscience includes any concept from the psychological or biological sciences (including any of the branches of physical science that might be relevant to describing the brain), the theory of the mind will turn out to involve any one of a very large number of possible combinations of scientific concepts. For example, the future of research could see the psychological sciences providing the functional description of the phenomena to be explained and biological neuroscience providing the mechanistic account of how function is implemented in the brain.[14]

The essential feature of this version of the neuron doctrine is that it does not have the radical consequence for linguistics and the other psychological sciences discussed above. In the first place, this version does not entail that linguistics and the other psychological sciences will be superseded because it is consistent with the trivial neuron doctrine that psychological science will be part of the successful explanation of the mind. In the second place, and more importantly, the view entails nothing about the concepts that will be used in a successful theory. The claim that the theory of the mind will be expressed in cognitive neuroscientific terms expresses nothing more, therefore, than an ecumenism in the development of the theory and an agnosticism about its content.

The trivial neuron doctrine is thus a very weak doctrine indeed. The picture that emerges from it has three components. First, the trivial neuron doctrine holds that the mind is a biological phenomenon; in other words, the trivial doctrine adheres to the thesis of materialism, the thesis that mental phenomena are neural phenomena. Second, the doctrine insists that the understanding of this phenomenon will derive from science; that is, the trivial doctrine adheres to the thesis of naturalism. Finally, however, the doctrine also holds that this understanding may not be provided by means of biological concepts alone but that psychological concepts may be required as well. Indeed, the trivial doctrine leaves it in principle open which concepts will feature in the successful theory of the mind. Since a joint commitment to materialism and naturalism is a scientific commonplace, and since it has no radical consequences, we have called this version of the neuron doctrine the trivial doctrine.[15]

As one might expect, the trivial neuron doctrine is widely held by cognitive scientists. James Higginbotham (1990), for example, says that while many cognitive scientists follow Descartes in supposing that

In saying that the trivial neuron doctrine is trivial, however, we do not mean to suggest that it is compatible with any approach whatever to the study of the mind. In general, any theory of the mind that denies either (a) that the mind is a biological phenomenon, or (b) that the study of the mind is a part of natural science, or (c) that at least psychology or neurobiology is in principle relevant for the explanation of the mind, is incompatible with the trivial neuron doctrine. An example of the first (and perhaps the second) kind of theory is the version of dualism according to which the mind is an object wholly distinct from the brain and body. An example of the second kind of theory is the version of social constructivism according to which the mind is a social construct in principle isolated from natural science. An example of the third kind of theory would be a certain version of artificial intelligence program according to which both neurobiology and the details of psychology are in principle irrelevant to the construction of theories of mentality in the most abstract sense. All such views are clearly incompatible with the trivial neuron doctrine.[16]

2.2.2. The radical doctrine. Since the trivial neuron doctrine amounts only to the claim that a successful theory of the mind will be a theory of the brain, it is uncontroversial and deserves to be as widespread as the neuron doctrine is. But this version of the neuron doctrine is uninteresting because no radical, or even moderately substantive, consequences follow from it.

However, there is a reading of the doctrine that does make it interesting. By substituting 'biological neuroscience' for 'neuroscience' in our formulation of the neuron doctrine, we get a radical doctrine according to which a successful theory of the mind will be a solely biological neuroscientific theory. Since, as we have said, we stipulate that biological neuroscience includes only neurophysiology, neuroanatomy, and neurochemistry, the radical neuron doctrine holds that neurophysiology, neuroanatomy, and neurochemistry will by themselves eventually have the conceptual resources to understand the mind and, as a consequence, that a successful theory of the mind will make no reference to anything like the concepts of linguistics or the psychological sciences as we currently understand them.[17]

According to the radical neuron doctrine, a successful theory of the mind will be a theory of the brain expressed in terms of the basic structural and functional properties of neurons, ensembles, or structures. As a result, the radical neuron doctrine is substantive in having as its essential feature the consequence that the intellectual project pursued by Higginbotham and many others is doomed from the beginning. After all, Higginbotham assumes - and assumes that most cognitive scientists assume - that the rules posited by linguists and psychologists cannot be reduced to neurobiological notions. If Higginbotham and others are right about this, and if the radical neuron doctrine is true, then the psychological sciences will yield in the fullness of time to better biological neuroscientific theories. To adopt again the phraseology of Patricia Churchland and David Hubel, these sciences produce only indirect theories of the phenomena they seek to explain and do not tell us what these phenomena are really like.

2.2.3. Evidence of commitment to the radical doctrine. Once we have the distinction between the radical and trivial neuron doctrines clearly before us, the crucial question is whether the proponents of the doctrine intend to defend the radical or only the trivial version. It seems clear that at least some of the passages cited above expressing commitment to the neuron doctrine should in fact be taken as expressions of commitment to the radical version of the doctrine.

Recall, for example, that in their discussion of the relation between psychological phenomena and neuroscience, the Churchlands (1994) write:

While support for the neuron doctrine - and, in some cases, the radical doctrine - appears to be widespread, in our view supporters of the doctrine do not always distinguish between the two versions we have identified. Indeed, a closer examination of some of the texts we've considered reveals a tendency to conflate the two versions of the doctrine.

From the passages cited above, for example, it seems clear that the Churchlands' official view is the radical neuron doctrine. Nevertheless, they sometimes present the radical doctrine as equivalent to the trivial doctrine. Consider again the passage from Churchland and Sejnowski (1992) cited above:

To take a different example, consider the passage from the Churchlands (1994) that we quoted earlier:

This conflation is also evident if one examines central trends in the Churchlands' work as a whole. One such trend is the methodological idea that neurobiology is, or should be, relevant to the task of explaining cognitive or psychological phenomena. Patricia Churchland (1986), for example, emphasizes a "co-evolutionary strategy" in developing theories of mental function that would explicitly take results in neurobiology into account (see also Churchland and Sejnowski 1992, p. 11). And Paul Churchland (1990) writes of the need for "empirical and theoretical research into brain function in order to answer the question of what are the most important forms of representation and computation within cognitive creatures" (p. 158). He goes on to say that "the long-standing disinterest in the neurosciences, both within AI and in cognitive psychology...has been most unfortunate" (p. 200). And the Churchlands describe the central fact that divides them from their critics as their rejection of the what they call the "autonomy" of psychology (see, e.g., Churchland and Churchland 1996, p. 220) according to which the development of psychology is conceptually isolated from results in neurobiology.

Now claims of this kind are unobjectionable because they entail only that in developing a functional theory of a particular psychological phenomenon, one does well to keep an eye on whether the theory is likely to have some neurobiological instantiation. Functional theories, as it is sometimes put, ought to be neurobiologically realistic. But what follows from this? Only that some approaches in psychology and artificial intelligence are mistaken. While the Churchlands may be quite right in criticizing these approaches, it does not follow from these claims - contrary to what Paul Churchland (1990) goes on to say - that "fundamental insights into the general nature of cognition are likely to be found by examining the microstructure and microactivity of biological brains" (p. 225). Nor does it follow, as Patricia Churchland (1986) writes, that

One can find a similar pattern of argument in other writers as well. In a passage we cited above, Edelman (1989) says that his aim is to

Finally, in a passage quoted above, Zeki says:

2.4. The importance of the ambiguity

The fact that the neuron doctrine is ambiguous between at least the two claims we have identified is enormously important for understanding and evaluating the doctrine. What the ambiguity explains is why a view that is apparently radical and controversial is so widespread. The neuron doctrine is both widespread and controversial because it has one interpretation that renders it very plausible but unsubstantive, and one interpretation that renders it radical but unsupported by the scientific evidence. Our first question was why informed people believe the neuron doctrine in the face of inadequate evidence. Our answer to this question is that the doctrine is ambiguous, and running together two different versions of the doctrine gives it the illusory appearance of having the important features of both.

But there is also another reason why the ambiguity is important. When confronted with the ambiguity, proponents of the neuron doctrine face two options: they can either say that they endorse only the trivial version of the doctrine or they can stick their necks out and endorse the radical version. For the Churchlands, however, the first of these options is out of the question because the trivial doctrine has none of the consequences that the Churchlands clearly want to defend. As we have seen, the trivial neuron doctrine is no more than scientific common sense. Indeed, if the Churchlands, or any one else, did intend to defend the trivial doctrine that intention would be entirely mysterious. Why would one bother to defend explicitly a doctrine that is trivial or, at any rate, that just about everyone in the field accepts? To say that smart money is on neuroscience, in our trivial sense of the claim, no more than to bet that some science of the mind or brain will win the race to understand the mind. But that is not a bet that any rational bookie would take.

On pain of triviality, then, proponents of the neuron doctrine must adopt the radical version of the doctrine. But this raises the second of the two questions we asked earlier, whether the radical - and interesting - version of the doctrine is true. Our goal in the remainder of the paper is to try to answer this question by considering three arguments for this version of the doctrine. In each case, we will suggest that the argument does not support the radical neuron doctrine. If we are right, the choice between the two versions of the neuron doctrine constitutes a destructive dilemma for its proponents: either to hold a view for which no scientific defense has been given, or to hold a view that requires none.

3. The argument from naturalism and materialism

The first argument we will consider is one that we have already mentioned in passing a number of times and which we will call the argument from naturalism and materialism. The attraction of this argument is that its premises are widely accepted both by neuroscientists and philosophers.[21] The first premise of the argument expresses commitment to what we have identified as naturalism. For our purposes, naturalism can taken be to be expressed by the following claim:

3.1. The limitations of the argument

One objection that could be brought against it concerns the inference to (4). We have already seen that the fact that As are made up of Bs does not entail that the explanation of the As is to be given in terms of the Bs. In particular, the fact that mental phenomena are identical to neural phenomena does not entail that the science of mental phenomena is the science of neural phenomena.

But there is also a second objection to this argument that concerns the distinction between the two conceptions of neuroscience that we have introduced. Even if the argument is successful, it is not an argument for the view of interest because (4') is the ambiguous formulation of the neuron doctrine rather than the radical formulation. The conclusion we need is:

In response, one might claim that our definition of the science of the brain is incorrect. It is often assumed that the brain is studied by neurobiology so that the innocent premise actually amounts to this:

Once we have the ambiguity of the neuron doctrine clearly before us, therefore, a plausible argument for the doctrine turns out to be only an argument for its trivial version. As we claimed in section 1, we suspect that many scientists and philosophers accept the neuron doctrine because of their commitment to some form of naturalism and materialism. Given that the trivial neuron doctrine is, as we have suggested, essentially an expression of those commitments, this is not surprising. However, while it can seem irresistible from these premises to draw the conclusion that the science of the mind will be solely neurobiological, with the two versions of the neuron doctrine in mind, it becomes clear why this line of argument is unpersuasive.

3.2. The trivial doctrine, naturalism, and materialism

We should emphasize that in our discussion of the argument above, we have not taken issue with the doctrines of naturalism and materialism. For the purposes of this paper, we agree with the naturalism of the first premise, and we agree also that mental phenomena are identical to neural phenomena. In addition, we take it as obviously true that the mind is to be explained by appealing to the structure and function of the grapefruit-sized things in our skulls.[23] Our objection is only to the view that the best description of that thing will be entirely neurobiological and to the idea that naturalism and materialism provide support for such a view.

Indeed, if there is any position that sits uneasily with naturalism, it is not ours but the position of those who support the radical neuron doctrine. The only way to infer the radical doctrine from naturalism is to make a prediction about the explanatory force of neurobiology in the fullness of time - a prediction we presented in the form of (3*). But naturalism itself cannot support such a prediction. On the contrary, in deferring to science for judgments about what there is and what it is like, the naturalist ought to avoid predictions of this sort.

4. The argument from unification

We turn next to an argument for the neuron doctrine that appeals to the notion of unity in science as a marker of successful theories.

It is often suggested that the fact that a science can be unified or integrated with other sciences is a virtue, and that, as a consequence, science itself is tending toward unity (Oppenheim and Putnam 1958; Sellars 1963; but see Dupré 1993 for an extended argument to the contrary).[24] One reason for this view is that science attempts to construct a picture of the world, and the more coherent, seamless and simple that picture is, the better. Another reason is the apparent connection between unity and explanation. The history of science seems to reveal that sciences or theories that unify previously disparate domains tend simultaneously to provide highly successful explanations of those domains; in unification, one finds explanation. The unification of electricity and magnetism in special relativity, the unification of gravity and inertia in general relativity, and the unification of evolutionary theory and genetics in neo-Darwinism are familiar examples. (We leave aside the question of whether unification itself constitutes a form of explanation, or whether successful explanations tend to co-occur with unification.)

Let us suppose then that unity is an important general tendency of science. How is does this occur? A natural suggestion is that global unity in science will be the product of many local theoretical unifications in the various branches of science: think globally, act locally! And it is here that the argument from unification for the radical neuron doctrine begins to take shape. The general idea is that, within the cluster of sciences that deal with mental phenomena, there is one science that is best placed to support the global tendency toward unity, and that is neurobiology. Because of its obvious connections to biology[25] and thereby to the rest of science, a neurobiological theory of the mind would contribute most to the overall goal of unity, and this means that our best bet is to regard neurobiology as the eventual science of the mind.

We can make this argument more precise as follows. Its first premise is simply the presumed historical fact about science:

Now in this formulation, the argument from unification is obviously open to a number of different objections. First, the idea that unification is a general trend of science is, as we have indicated, controversial in some quarters (Dupré 1993). Second, it is certainly not obvious - contrary to what the argument assumes - that the best way of achieving global unity is by means of local unifications.

We have some sympathy with both of these objections. But we think there is another objection which cuts deeper than either. The central problem with the argument from unification is that it does not distinguish among different conceptions of unification. Following Maudlin (1996), we distinguish three different relations or processes that might be denoted by 'unification.'[26] This in turn suggests that the idea behind premise (2) has at least three different formulations. We will argue that, on none of the formulations, is the argument persuasive.

4.1. Unification as dissolution

On the first interpretation, 'unification' denotes a process we shall call dissolution. When dissolution occurs, the distinction between two theoretical domains is dissolved by a conceptual advance. That advance reveals the two domains to be features or manifestations of a single theoretical domain or to be derivable from that domain. In his paper, Maudlin cites special and general relativity as two paradigms of dissolution in this sense.

In the sciences of the mind, dissolution is also possible. It would require the discovery or development of a family of concepts that would reveal the biological features of the brain and the psychological features of the mind to be manifestations of, or derivable from, some third set of things. This view has traditionally been called double-aspect theory in philosophy, and it is surprising that it has so few contemporary supporters. As Maudlin notes, the likelihood of a unification of the four forces of nature in a "theory of everything" is a dogma in physics. But the parallel view in the philosophy of mind is thought to be eccentric.

One philosophical view that might be classified as a dissolution proposal is Wilfrid Sellars's (1963; 1971) theory of color. Sellars argued that the common sense picture of the world (the manifest image) must be mistaken in its commitment to the existence of color because science has revealed that the world is made of atoms, and atoms cannot be colored. Sellars eliminated color as it is normally conceived from his manifest ontology, but he argued that in the completed scientific description of the world (the scientific image), colors would be reintroduced in the form of entities he called pure processes. Sellars thought that pure processes would be discovered by physics, and for this reason color would be scientifically explicable. Being pure processes, however, colors would not have to be properties of colorless atoms. Pure processes would thus dissolve the distinction between the psychology of color and color physics, producing a third unified domain.

Sellars's account appeals to a future physics for a dissolution of the boundary between the mental and the physical, but one might be inclined to think it more likely that neurobiology itself will produce radically new concepts that will dissolve that boundary. Indeed, this is the claim behind premise (2) of the argument from unification, if by 'unification' one means dissolution. If the premise is interpreted this way, and the argument is persuasive, then the final theory of the mind would be neurobiological, and the radical neuron doctrine would be vindicated.

In our view, however, the argument is not persuasive when the premise is so interpreted, and for two reasons. First, suppose that future science discovers a radically new family of concepts that dissolves the distinction between neurobiology and psychology. It seems completely arbitrary to count these concepts as part of neurobiology rather than psychology; after all, the family of concepts is radically different from both. Moreover, once the concepts are in place, there will be no difference between these disciplines. So dissolution undermines the radical neuron doctrine by doing away with neurobiology at the same time as it does away with psychology.

There is also a simpler reason why this version of the unification argument is unpersuasive. As Maudlin (1996) argues, dissolution is very hard to come by even in physics.[27] The theory of everything, like the radical neuron doctrine, is a dogma without much scientific support. Since physical theories rarely achieve dissolution, and since there is doubt about whether fundamental physics will ever do so, even though a deep understanding of the phenomena is already available there, we conclude that there is that much less reason to think that the distinction between the neural and the mental will be dissolved by future theoretical advances. Since we doubt that dissolution will ever be achieved at all, we doubt a fortiori that neurobiology will produce it.

4.2. Unification as reduction

The second and most obvious possibility is that by 'unification' one means intertheoretic reduction. Reductionism is the view that the concepts and the laws of a more basic theory - the reducing theory - can be used to capture and explain the phenomena described in a less basic theory - the reduced theory. In cases of reduction, the reduced theory is derived from, and exhibited as a proper part of, the reducing theory (for the locus classicus, see Nagel 1961; and for a more recent discussion, see the papers in Kim 1993).[28] In the case of psychology and neurobiology, this means that a science recognizable as neurobiology will eventually produce concepts that reduce psychology, just as physics reduced the concept of temperature to the concept of mean kinetic energy. Since neurobiology is taken to be more fundamental, the explanatory power of the theory would lie with neurobiology and would justify the claim that the successful theory of the mind was solely neurobiological.

A variant on this position is eliminativism, usually associated with the Churchlands (see, for example, P.M. Churchland 1981). Eliminativism envisages a replacement of psychology by neurobiology rather than a process by which psychology is exhibited as a proper part of neurobiology. Physics did not reduce the mistaken concept of phlogiston to some more fundamental physical concept; it disposed of the concept altogether. Similarly, on the eliminativist picture, a future theory of the mind will dispose of psychological or mental concepts. Although reduction and elimination appear at first blush to be rather different views, they are in fact two ends of a continuum defined by the extent to which one believes that psychology is correct in its description of the mind. At one end of the continuum is the view that psychology is entirely correct but that its description is not given in fundamental terms. In this case, psychology must be reduced to the level of neurobiology. At the other end of the continuum is the view that psychology is entirely mistaken about the mind and must be replaced by a neurobiology that starts from scratch with a new set of concepts. A more likely outcome of scientific progress is that some of psychology will be reduced and some eliminated as neurobiology develops. For our purposes, the class of views on the reduction-eliminativism continuum can be considered together. For simplicity, we will call this class of views 'reduction.'

Reduction differs from dissolution in the degree of radicalness envisaged in the future of neurobiology. Whereas dissolution imagines that neurobiology will undergo an Einsteinian or Sellarsian revolution and obliterate the distinction between what we currently take to be neurobiology and psychology, reduction envisages a less radical option in which neurobiology goes on much as it is but gradually develops the resources to flesh out or replace psychology. On the reductionist view, future neurobiology will be recognizable as neurobiology but will have greater explanatory power.[29]

There is an enormous literature dealing with reductionism, and it is not our concern to evaluate all the arguments for and against it here.[30] Rather, we are interested only in asking whether the argument from unification is persuasive if in premise (2) one reads 'reduction' for 'unification.' As we have noted, the support for the premise is that considerations of unification privilege neurobiology over other sciences when it comes to explaining psychological phenomena. If by 'unification' one means 'reduction,' this becomes the idea that considerations of reduction privilege biological neuroscience when it comes to explaining psychological phenomena. But do they?

In our view, the answer to this question is "no." If reductionism is a constraint in science at all, then it is a general or global constraint. That is to say, if reductionism is true, then anything that is in principle reducible, reduces to the most basic science there is, namely, physics.[31] But this means that considerations of reduction do not privilege neurobiology over the other sciences but only physics, and reductionism implies that a successful theory of the mind will be solely physical and not solely neurobiological. The relation between neurobiology and psychology is left entirely open by reductionism and must be regarded as an empirical question about the local relations among the sciences. In short, then, the appeal to reductionism does too much for the proponent of the neuron doctrine. If one is going to be a reductionist, one has to take the train of reduction to the terminus of physics. In the absence of a further argument privileging neurobiology as well as physics, neurobiology represents for psychology nothing more than - in Jerry Fodor's (1981) phrase - a local stop. If there is an argument privileging neurobiology in this way, it is not an argument that derives from reductionism itself, and this is enough to defeat the argument from unification on the interpretation we are considering.

4.3. Unification as conjunction

The final possibility we will consider is that by 'unification' one means what we shall call conjunction. Conjunction is the process whereby two theories are unified simply by being joined together into a single larger theory. Although conjunction represents an extremely weak version of unification, it is not empty. For one thing, to conjoin two theories, they must at least be mutually consistent, and their consistency may not be a simple matter to establish. For another, the notion of conjunction must somehow be made interesting enough so that, as Maudlin (1996) puts it, merely showing that a theory of embryonic development and a theory of the formation of the rings of Saturn are not inconsistent when conjoined does not count as unifying them. However these issues are resolved, the important issue for us is that the claim of conjunction with respect to the mind is that a successful theory of the mind will be neurobiology-conjoined-with-psychology.

It is not necessary for us to decide here whether current neurobiology and psychology are consistent or, if not, whether they will one day be made consistent. It suffices for us to note that if, in premise (2), one means 'conjunction' by 'unification,' then the argument from unification will not support the radical neuron doctrine because, unlike dissolution and reduction, conjunction unifies without doing away with anything, including psychology. Since psychology would continue to be part of a successful theory of the mind on this view, conjunction supports only the trivial, but not the radical, neuron doctrine.

5. The argument from exemplars

The two arguments for the radical neuron doctrine that we have considered thus far proceed from philosophical considerations. The final argument we discuss makes use instead of the details of neuroscience itself. We call this argument the argument from exemplars.

5.1. An inductive strategy

It might seem that any argument of this kind is doomed from the start since, as we noted at the beginning of this paper, neuroscience is at an early stage of development. How then can an argument based on an embryonic neuroscience to be developed?

The following passage from Hubel (1988) suggests a way in which this might be done:

5.2. Kandel's theory of learning in Aplysia

In order to evaluate the argument from exemplars, one would have to consider a number of exemplars in some detail.[32] In this paper we will consider one - the neuroscientific theory of elementary learning developed by Eric Kandel and his colleagues (Pinsker et al. 1970; Castellucci and Kandel 1976; Walters, Carew, and Kandel 1981; Kandel and Schwartz 1982; Walters and Byrne 1983; Hawkins et al. 1983; Hawkins and Kandel 1984; Hawkins, Kandel, and Siegelbaum 1992; Walters and Byrne 1983; Bailey and Kandel 1995; for a summary see Shepherd 1994 and Kandel, Schwartz, and Jessel 1995).

Although we consider only one exemplar, we take the upshot of the analysis of this case to be highly significant for the following reasons. First and foremost, this case appears to offer an account of some forms of naturally-occurring learning in terms of neurons and their properties alone, and for this reason seems ideally suited to support the radical neuron doctrine. In addition, the theory we consider has a number of other virtues that make it a worthy case to consider. It is a very successful bit of neuroscientific theory in the sense that it appears to offer a relatively complete account of a set of behaviors. It is successful also in the sense that it manages to integrate the behaviors in question not only with cellular neurobiology but also with the biochemical and molecular events that are crucial to those cellular processes. It thus traces learning to its most basic biology. In addition, Kandel's theory makes use of a central neuroscientific notion - the notion of neural plasticity,[33] according to which the structure and function of adult neurons can be altered - that plays an important role in learning theory quite generally and is related to other areas of neuroscience such as neural development. It is the sort of account, therefore, that might be expected to generalize at least to other forms of learning and possibly beyond.[34] Finally, we take it to be a sociological fact that Kandel's theory is widely regarded in the neuroscientific community as the best that neuroscience can now offer in the way of explanation of behavior or the mind in fundamental neuroscientific terms. If our critique of the radical neuron doctrine makes sense in the context of this bit of neuroscience, therefore, that is strong evidence for the account. Other exemplars would have to be examined to make the case complete, but this is a good place to start.[35]

5.2.1. Simple and associative learning. Kandel's theory deals with a cluster of elementary forms of learning called simple learning and associative learning. The theory attempts to explain these behaviours by appealing to the properties of a small family of neural circuits composed primarily of a sensory neuron-interneuron-motor neuron pathway, and the process of learning is hypothesized to be identical to a change in strength of the sensory neuron-motor neuron synapse. This change in synaptic strength occurs as a result of an alteration in the extent of exocytosis - the production of neurotransmitter - in the sensory neuron. This basic model can be adapted or modified to account for a number of forms of learning including habituation, sensitization, and classical conditioning, and the features of the model may be able to account for other aspects of elementary learning as well (Hawkins and Kandel 1984).

Habituation is the process whereby a neutral stimulus is gradually ignored by an animal when it leads neither to harmful nor rewarding consequences. Sensitization is the process whereby a harmless stimulus that produces no aversive response comes to be experienced as noxious following a stimulus that naturally produces aversion. Both habituation and sensitization are classified as forms of simple learning in contrast to associative learning, the paradigm of which is classical conditioning - the learned association between two stimuli such as the ringing of a bell and the presence of food, in the well-known Pavlovian case. In classical conditioning the spontaneous response to a non-neutral stimulus (the unconditioned stimulus or US) is transferred to the neutral stimulus (the conditioned stimulus or CS) in virtue of their contiguity, or repeated pairing in time.[36]

5.2.2. The neurophysiology of simple and associative learning. Because the neurophysiology of processes even as simple as these elementary forms of learning would be enormously difficult to study in complex organisms, Kandel and his colleagues use the marine snail Aplysia californica as a model. The Aplysia is an organism with a very simple central nervous system made up of about 20,000 cells, but it exhibits an innate gill-withdrawal reflex that can be modified in simple or associative learning paradigms.

For example, a neutral tactile stimulus to the tail of an Aplysia, initially producing a weak gill-withdrawal reflex, can be habituated. Habituation of this reflex occurs as a result of a decrease in the quantity of neurotransmitter released at the synapses made by sensory neurons on inter- and motor neurons. Sensitization in Aplysia occurs when a noxious stimulus causes a weak gill-withdrawal reflex to be strengthened. This process involves a change at the same neural locus as habituation, but in sensitization there is an enhancement of neurotransmitter release by the sensory neurons on their target cells. The process by which this occurs, however, is more complex. A mild stimulus to the siphon of an Aplysia produces a weak gill-withdrawal reflex. A shock to the tail activates facilitator interneurons which synapse near the synapse formed by the siphon sensory neurons on the motor neurons. The activity of the interneurons at this locus causes a molecular process to be initiated within the siphon sensory neuron, the upshot of which is that the sensory neuron is disposed to produce more neurotransmitter than prior to stimulation. When the siphon is weakly stimulated again, therefore, more neurotransmitter is produced, and a stronger gill-withdrawal reflex occurs. This process is called presynaptic facilitation because the sensory neuron (the presynaptic neuron) is made more effective, or facilitated, by means of the activation of the facilitator interneurons (Figure 1).

The cellular mechanism of classical conditioning in Aplysia, is, on Kandel's model, an elaboration of the cellular mechanism of sensitization. In classical conditioning, a US, such as a tail-shock, is contiguous with a CS, such as a weak tactile stimulus to the siphon. Initially, only the US causes robust gill-withdrawal, but repeated pairings of the US and the CS eventually cause the gill-withdrawal to occur in response to the CS as well. At the cellular level, the process of presynaptic facilitation in the pathway responding to the CS is enhanced by activity in the neural pathway responding to the US which occurs just afterwards. The action potentials generated by the US cause a greater facilitation of the sensory neurons responding to the CS as a result of the activation of the facilitator interneurons. This change makes the sensory neuron more effective in causing the motor neuron to fire and bring about gill-withdrawal in response to the CS. The primary difference between conditioning and sensitization, therefore, is the temporal coincidence of the activity of the facilitating and facilitated pathways. As a result, Kandel calls the process activity-dependent presynaptic facilitation (Figure 1).

Figure 1: A partial circuit for sensitization and classical conditioning in Aplysia. In sensitization, a shock to the tail acts on siphon sensory neurons to bring about presynaptic facilitation. In classical conditioning, presynaptic facilitation is dependent upon the temporal pairing of CS (siphon) pathway activity and US (tail) pathway activity. The CS pathway is active just prior to activation of the US pathway, thereby enhancing presynaptic facilitation of the siphon sensory neuron. (Adapted from Hawkins and Kandel 1984.)   Before moving on to the philosophical questions at issue, it is important to observe that Kandel's model not only provides a neural description of the processes that underlie some forms of elementary learning, it also makes important conceptual contributions to the study of learning in general (see Kandel, Schwartz, and Jessel 1995). Kandel argues that studies of learning in Aplysia reveal that short-term and long-term memory are part of a continuum and not two utterly distinct processes (Frost et al. 1985). The model also provides evidence that memory function in elementary learning is not a function of a neural network but of individual cells (Bailey and Kandel 1995).[37] Kandel's theory also holds that it is possible to produce a model of classical conditioning in terms of the process of sensitization, thus suggesting that the psychological distinction between simple and associative forms of learning may not be as hard and fast as one might suppose.[38] All of these are substantial claims about learning and not just its neurobiology.

5.3. Kandel's theory and the radical neuron doctrine

Clearly, then, Kandel's theory provides an admirable exemplar of neurobiological research. However, our aim here is not to evaluate the theory but to ask whether it supports the view that a successful theory of the mind will be solely neurobiological. It is important to be clear about what this question means. We are not asking whether Kandel's theory is true, let alone whether it constitutes a complete neuroscientific theory of learning, and still less of the mind! We are assuming that Kandel's theory is a successful explanation of the neurophysiology of elementary learning. What we want to know is whether, in accordance with the strategy of the argument from exemplars, Kandel's theory provides inductive support for the view that a future account of mental phenomena will be solely neurobiological. If we assume that Kandel's theory gives us a window onto a successful theory of the mind, can we infer inductively that the theory will be neurobiological?

Our answer to this question is "no." Whatever the virtues of Kandel's theory, we will argue that it is not solely neurobiological, and for this reason, that the view through the window of that theory reveals the future theory of the mind not to be solely neurobiological either. The instance of the argument from exemplars that appeals to Kandel's theory, therefore, does not succeed. In what follows, we focus on the case of classical conditioning.

5.3.1. "Pure" neurobiology. We have claimed that Kandel's account is not purely neurobiological. In saying this, we mean that the notions involved in the description of classical conditioning in Aplysia include substantive psychological concepts and not merely the concepts of neurobiology. The history of classical conditioning, its roots in the associationism of Hume and Mill, and its incarnation in behaviorism is well-known (Boring 1950; Pinker 1991).[39] Kandel's theory of classical conditioning is not developed in a vacuum but, as Kandel and his co-workers acknowledge, makes use of recent work in psychology that is part of that tradition. This includes, in particular, the primary psychological model of classical conditioning - that of Rescorla and Wagner (1972; see also Gluck and Thompson 1987). In modelling Aplysia neurophysiology, that is, Kandel's theory appeals explicitly to psychological concepts.[40] And it is this fact that leads us to say that his model of conditioning is not solely neurobiological.

The fact that Kandel's account is developed within an explicit and highly theoretical psychological framework means that the account does not provide a genuine alternative to psychological theory. It rather absorbs the required psychological ideas in order to provide a framework for understanding the behavior of Aplysia neurons and their role in conditioning. While Kandel's account changes our conception of conditioning somewhat (e.g. in providing evidence that it is constructed out of the mechanism of sensitization), it does not replace that theory. Its primary success is the fleshing out of a psychological story in neurobiological terms (Gluck and Thompson 1987). This should not be surprising because neurobiology has no concepts that can be used to describe the behavior of an animal so that the notion of a "pure" neurobiology actively in competition with psychology can only be a vision of some future science.

In order to make our argument clearer, consider an illustration from the theory of color. One of the salient features of color phenomenology is the fact that only certain color combinations are possible. We can, for example, see reddish-blue and reddish-yellow but not reddish-green. This aspect of color perception is called color opponency, and it led Ewald Hering in 1877 to propose the opponent process theory of color perception, later revived and developed by Leo Hurvich and Dorothea Jameson (see Hurvich 1981). According to this theory, the space of all perceivable colors can is organized along three axes of phenomenal difference - a red-green axis, a blue-yellow axis, and a black-white axis. The contemporary interpretation of opponent process theory appeals to opponent neural channels the function of which depends on color opponent cells. These neurons - different species of which exist in the retina, lateral geniculate, and cortex - behave in an opponent fashion, being excited, for example, by green light in the surround of the neuron's receptive field and inhibited by red light in its center (see Zeki 1993; cf. also Hardin 1988).

Now, in our view, in order to correctly model the function of opponent neurons, one has to appeal to opponent process theory. That is to say, in the absence of the psychological theory, one cannot understand what opponent neurons do. While the neurobiology of color opponency, therefore, might exhaust the mechanism of color opponency it does not provide a complete theory. A complete theory of color opponency must appeal to the function of opponent neurons and the psychophysical framework of opponent process theory. A purely neurobiological theory of color opponency, therefore, does not exist even if opponent neurons are all there is to the mechanism of opponent color vision.

Our claim about Kandel's theory of learning is analogous. Even if the synaptic mechanisms described in that theory are all there is to the mechanism of elementary learning, it does not follow that there is a purely neurobiological theory of elementary learning. Indeed, we claim that there isn't such a theory. In the sections that follow, we develop this argument by elaborating the idea that Kandel's theory is not purely neurobiological. We do this by considering a number of objections to the argument just described.

5.3.2. An objection concerning reduction. It might be objected that we have set an unreasonable standard for Kandel's theory, and, by implication, for any other putative neurobiological theory of a mental phenomenon. Let us suppose that Kandel's theory is genuinely explanatory of the neurophysiology of classical conditioning, and let us further suppose that while it appeals to psychological notions or theories, it can nonetheless explain conditioning in neural terms. Under these assumptions, shouldn't we say that the presence of psychology in the theory is harmless?

Consider a familiar analogy already mentioned. We know that temperature is mean kinetic energy, but we continue to refer to temperature nonetheless. The mere fact that we do so, however, does not mean that there is anything inadequate in the original identification. Similarly, one might argue that the presence of psychology does not affect the success of Kandel's theory at explaining conditioning in neurobiological terms. On this view, Kandel's theory is purely neurobiological in any sense that matters scientifically, and it counts therefore as support for the radical neuron doctrine.

5.3.3. Response: Reduction and implementation. We think this objection misconstrues the role of the psychology in Kandel's theory. In order to explain why, we distinguish two kinds of case in science in which an intuitively more basic theory is brought to bear on the phenomena explained by an intuitively less basic theory.

The first case is reduction. As we remarked above (section 4.2.), reduction is the process whereby the concepts and laws of the more basic reducing theory are used to explain the phenomena described in the less basic reduced theory. Because the reduced theory can be derived from the reducing theory, the latter is conceptually independent of the former. The reduction of temperature to mean kinetic energy is such a case. Because the phenomena explained by the concept of temperature and its associated laws (such as they are) can be derived from the concept of mean kinetic energy and its associated laws, the kinetic theory can explain at least as much as the temperature theory can. In principle, therefore, the concept of temperature and its laws can be ignored without any loss of explanatory power.[41]

The second case is implementation, whereby a more basic theory provides the mechanistic details of the system that instantiates the functions posited by the less basic theory. The best-known illustration of implementation is Marr's (1982) conception of the role of neurobiology in the theory of vision. (For a more general discussion, see Cummins 1983.) According to Marr, the neurobiology of vision describes how a particular psychological process, which Marr referred to as an algorithm, occurs in a particular neural system. In turn, the algorithm is a particular instantiation of a more abstract computational process. In cases of implementation, the conceptual work is done by the computational theory, and not by the algorithm or the implementation. However important the neurobiological story is, it remains conceptually parasitic on the higher-level theory. If the computational level of the theory were eliminated, the mechanistic details would no longer make sense. As Marr (1982) famously remarked, it is impossible to explain how a wing works simply by describing its feathers.

Implementation can hold between neurobiology and psychology whether or not the psychological story is computational; other psychological accounts will do (Cummins 1983). The case of color opponency discussed earlier can be used to illustrate this point as well. We suggested above that the properties of color opponent cells might constitute the mechanism that causes color vision to have an opponent character. One can therefore think of the physiology of color opponent cells as implementing opponent process theory because, as we suggested, in the absence of that theory, one cannot explain color opponency or what these neurons do.

Nor is the relation of implementation restricted to theories of psychological function. Suppose, for example, that someone were to produce an explanation of gene replication in terms of particle physics. It is plausible that while such a theory would explain what is happening at the particle level during replication, the explanation of the basic phenomena of replication would continue to reside at the molecular level. In such a case it would be appropriate to call the particle story an implementation of the molecular story rather than a reduction of it.[42]

Given this distinction, the question is whether Kandel's theory is an instance of reduction or implementation. We think there is good reason to suppose it is an instance of the latter. Hawkins and Kandel (1984) themselves say:

5.3.4. Rejoinder: Interpretation aside, doesn't Kandel's theory in fact provide a conceptually independent description? In response to our claims above, one might argue as follows.[43] Suppose one gave a neuroscientist a description of activity-dependent presynaptic facilitation in the classical conditioning paradigm. Perhaps the story might go like this: stimulus 1 produces a moderate response in the sensory neuron; it releases a small quantity of neurotransmitter which fails to elicit the gill-withdrawal reflex. With repeated near-simultaneous activation of a connected pathway by stimulus 2, the response of the sensory neuron is enhanced, a greater quantity of neurotransmitter is eventually released and its capacity to fire the motor neuron is facilitated; and so on. Wouldn't it be fair to say that our neuroscientist understands classical conditioning as well as any psychologist without the need for psychological concepts? If so, then it looks as if the concept of activity-dependent presynaptic facilitation is not parasitic on the psychology of conditioning, and our claim that Kandel's theory is implementation rather than reduction is mistaken.

5.3.5. Response: The complexity of conditioning. The objection is tempting, on our view, only if one underestimates the psychological complexity even of the elementary forms of learning dealt with by the Kandel model. This becomes clear if we briefly examine the psychology of classical conditioning.

A pervasive view of classical conditioning, even in textbook presentations, is that it is a process whereby contiguity of the US and the CS transfers the response from the former to the latter. But, as Rescorla (1988) argues, while this view is adequate to the conception that Pavlov himself had - a conception that emerged from the reflex tradition - it is entirely inadequate to account for the data that have been accumulated over the last twenty years of research into conditioning.[44]

Consider, first, some relevant data. Contiguity is essential to the traditional theory and is supposed to be sufficient to elicit conditioning. In an early study, however, Rescorla (1968) compared two related learning situations. In the first situation, a rat is exposed to the CS, a tone, that is presented randomly during a two-minute interval. During the same interval, the US, an electric shock, is also randomly applied (Figure 2A). In the second situation, the shock occurs only when the tone is presented (Figure 2B). In the latter case, but not the former, an association between the tone and the shock is learned. Notice, however, that the CS and the US occur within the same overall period of tone and shock exposure, thus satisfying the requirement of contiguity. Nevertheless, only one pattern of CS and US pairing produces conditioning.

Figure 2: A classical conditioning paradigm demonstrating the insufficiency of contiguity for conditioning. In the first case (A), the CS is always paired with the US, but the US also occurs at other times during the training interval. In the second case (B), the US occurs only when the CS occurs. Although the requirement of contiguity is satisfied in both cases, conditioning is achieved only in the latter case. (Adapted from Rescorla 1988.)

  Of course, conditioning occurs in the latter case because the tone provides information about the occurrence of the shock. In order to explain this effect, therefore, one needs to appeal to some notion of information which is richer than the notion of "low-level mechanical process in which the control over a response is passed from one stimulus to another" (Rescorla 1988, p. 152).

Here is a second case (see Rescorla 1980). In a variation of classical conditioning called autoshaping, a bird learns to associate a red square, the CS, and food, the US, and will eventually come to peck at the square as if it were the food itself. The bird will also learn to peck at stimuli that are associated with the original CS - a form of learning known as second-order conditioning. Now consider two different conditions. In the first, an achromatic outline of a square is associated with the red square; in the second, an achromatic outline of a triangle is associated with the red square. In the first case but not the second, the second-order stimulus relates to the first-order stimulus as part to whole, and in the first case, conditioning occurs much more readily than in the second. A mental representation of a relation governing the stimuli thus has a differential effect on the course of learning (Figure 3). Once again, the traditional conception of contiguity will not explain the effect.

Figure 3: Learning an association between a second-order stimulus that bears a part-whole relation to a first-order stimulus. In the similar case the second-order stimulus bears the part-whole relation to the first-order stimulus, and in the dissimilar case it does not. (Adapted from Rescorla 1988.)

  In general, recent research has shown that classical conditioning is more complex in many respects than Pavlov and others believed (see Rescorla 1988). For example: (1) the context of learning is relevant; (2) the animal learns a variety of associations among many stimuli simultaneously in the learning situation; (3) associations exhibit a hierarchical organization; and (4) the response to the CS depends not only on the response to the US but on the properties of the CS itself: for instance, a tone signalling shock will cause a rat to freeze, but a prod signalling shock will cause the rat to try to cover the prod up. Classical conditioning is thus more than the transfer of response from US to CS. It requires positing, as Rescorla (1988) puts it, "the learning of relations among events that are complexly represented, a learning that can be exhibited in various ways" (p. 158). Thus,

The second horn of the dilemma is this. In response to the objection just made, one could argue that Kandel's model is neutral with respect to the reflex and the modern conceptions of classical conditioning, and indeed to any other conception. Kandel's theory could be interpreted as an instance of whichever notion turns out to be correct. On this line of argument, however, Kandel's model cannot be a reduction of the theory of conditioning because it is neutral with respect to incompatible theoretical notions. Since it does not tell for or against a particular theoretical conception, it cannot reduce any particular one. To take an analogy: a putative reduction of optics that didn't decide for or against the wave conception of light or the particle conception (or a mixed conception) could not successfully reduce optics.[45] Similarly, no putative reduction of classical conditioning that speaks neither for nor against some essential theoretical characterization of conditioning can be successful.

We conclude that the claim that Kandel's model is a reduction of classical conditioning, rather than an implementation of it, cannot be sustained because it is not possible to understand the full complexity of conditioning without recourse to some psychological theory or other. If Kandel's model is accurate, it can only be a representation of how the relevant psychological notions are instantiated in the neural machinery.

We have only considered classical conditioning here, but we suspect that similar remarks could be made about other forms of elementary learning (see, e.g., Wagner and Pfautz 1978). And if such remarks could not be made, this in itself would be problematic for Kandel since it would reintroduce theoretical variation (say, between conditioning and sensitization) where Kandel has argued for theoretical unity. What is true of conditioning had better be true of the other forms of elementary learning, or one of the significant virtues of the Kandel model is lost.

5.3.6. Summary. How does this discussion affect the question of the neuron doctrine? The radical neuron doctrine says that a successful theory of classical conditioning, among other phenomena, will be neurobiological. But if our argument is sound, then even if we consider Kandel's theory of elementary learning to be a correct neurobiological description of the phenomena, it does not follow that a successful theory of conditioning will be solely neurobiological, any more than it follows from the implementation of color vision by color opponent cells that a successful theory of vision will be neurobiological, or from the imaginary particle physics description of replication that a successful theory of replication will be a part of physics. What determines the form of a successful theory is where the best explanation is to be found, and in the present case, we take the best explanation to reside with psychology and not neurobiology because we take the latter to be parasitic on the former. We think that it is currently an open question what form a successful theory of learning will take and that the answer to this question will depend in part on whether neurobiology can produce conceptually independent accounts that are superior to those already to be had in psychology. Since Kandel's theory is not purely neurobiological, therefore, appealing to it in the context of the argument from exemplars does not support the radical neuron doctrine.

5.4. Is Kandel's model a special case?

Our interest in Kandel's theory of learning in Aplysia has been driven by the idea, suggested in the passage we quoted from Hubel, that exemplars of neuroscientific theorizing might be taken as an inductive basis from which to infer the radical neuron doctrine. We have argued that while Kandel's account of learning is an important part of neuroscientific theory, it does not support an inference to the radical doctrine because it involves psychological notions.

One might respond to our argument with the suggestion that Kandel's theory is in some sense a special case and that there are other cases of successful neuroscientific theory which might do the job better. There are a variety of senses in which one could take the phrase 'special case.' We consider three and try to show that Kandel's theory is not a special case in any sense that matters to the argument from exemplars.

5.4.1. The problem of a single case. One might argue that, given its structure, the instance of the argument from exemplars supporting the radical neuron doctrine is not refuted by the failure of a single case. The argument requires that there be some exemplars that support the radical neuron doctrine, and the failure of this particular case does not show that there are no others that would do the job. Indeed, one might argue, for the same reason, that the argument would not strictly be refuted by the failure of a number of cases.

These claims are quite true, but they are irrelevant to our critique because they apply to all arguments of the present form and not to ours in particular. We cannot prove that there are, or will be, no cases that support the argument from exemplars. A reasonable agnosticism about the future of science explicitly prohibits us from trying. But the burden is on the supporters of the argument to offer a better candidate than Kandel's theory. In the absence of such a candidate, one ought to accept that the argument fails to support the radical neuron doctrine even though it hasn't been strictly refuted.

5.4.2. The problem of an inadequate case. A second way in which Kandel's theory might be claimed to be a special case is if it were to be shown that the theory were not the best neuroscience could offer. If there were other neuroscientific theories of mental phenomena that answered the requirements of the argument from exemplars, our choice of Kandel as a paradigm case would be unfair to the defender of the radical neuron doctrine.

To this we can only appeal to the consensus in the neuroscientific community which we mentioned at the outset of our discussion. In our opinion, Kandel's theory is widely recognized as the sort of model to which neuroscience aspires, and, to that extent, is not special in the sense of being a poor representation of neuroscientific theory.[46] We know of no other theory that is as successful as Kandel's, but we are, of course, prepared to be convinced otherwise.

5.4.3. The problem of a case with unique features. There is one important sense in which it might be argued that Kandel's theory is a special case. If one could show that the theory had unique features that supported our critique above - i.e. features that supported our claim that psychological theory is necessary to interpret it - then one could argue that our view of Kandel does not defeat the argument from exemplars. That is to say, if it could be shown that future neuroscientific theories won't rely on psychological notions as Kandel's theory does, then the arg