Dupré, J. (1995) "The Solution to the Problem of the Freedom of the Will" Philosophical Perspectives 10 385-402 

The Solution to the Problem of the Freedom of the Will

John Dupré
Department of Philosophy
Birbeck College, University of London
Malet Street, London, WC1E 7HX
(0171) 631 6383/6549  Fax:  0171 631 6564
School of English and American Studies
University of Exeter, Exeter, EX4 4QH
(01392) 264267/264257  Fax:  01392 264361 


It has notoriously been supposed that the doctrine of determinism conflicts with the belief in human freedom. Yet it is not readily apparent how indeterminism, the denial of determinism, makes human freedom any less problematic. It has sometimes been suggested that the arrival of quantum mechanics should immediately have solved the problem of free will and determinism. It was proposed, perhaps more often by scientists than by philosophers, that the brain would need only to be fitted with a device for amplifying indeterministic quantum phenomena for the bogey of determinism to be defeated. Acts of free will could then be those that were initiated by such indeterministic nudges. Recently there has been some inclination to revive such a story as part of the fallout from the trend for chaos theory. Chaotic systems in the brain, being indefinitely sensitive to the precise details of initial conditions, seem to provide fine candidates for the hypothetical amplifiers of quantum events.

But this whole idea is hopeless. I need only recall that the interest in establishing free will is not the conviction that humans are random action generators, but a concern that human autonomy is inconsistent with the possibility of fully explaining human actions in terms that have no apparent connection with the wishes and beliefs of the human agent.1 Standard compatibilist claims that human autonomy and mechanistic causal explanation are not mutually exclusive may or may not be defensible. But the attempt to reconcile human autonomy with the complete randomness of human actions is surely hopeless.2 At first sight it appears that, despite the initial worries about determinism, indeterminism makes the conception of freedom of the will even less tenable.3

Despite the untenability of the ideas just mentioned, my aim in this paper will be to show that the solution to the problem of the freedom of the will does lie, nevertheless, with the truth of indeterminism. To see how this is so, it is necessary first to distinguish two very different grades of indeterminism. The indeterminism entailed by the common understanding of quantum mechanics, while it denies that the causal upshot of a situation is a determinate function of any fact about that situation, still insists that there is a complete causal truth about every situation. It is just that this truth is in the form, not of a unique outcome, but of a range of outcomes with specific probabilities attached to their occurrence. Thus situations are still conceived as evolving according to laws, just laws of a somewhat different kind. I shall refer to both determinism, and this brand of moderate indeterminism, as versions of the thesis of causal completeness. Even if determinism is false, causal completeness requires that there be some quantitatively precise law governing the development of every situation. If we maintain the doctrine of causal completeness, then the only retreat from physical determination of our actions is in the direction of more or less unreliability, hardly a desirable philosophical goal. However, the indeterminism that I wish to advocate is something quite different, the denial of causal completeness. I shall maintain that few, if any, situations have a complete causal truth to be told about them. Causal regularity is a much rarer feature of the world than is generally supposed. And the real solution to the problem of freedom of the will, I shall argue, is to recognize that humans, far from being putative exceptions to an otherwise seamless web of causal connection, are in fact dense concentrations of causal power in a world where this is in short supply.

The solution to the problem of human autonomy that I propose, then, is a complete reversal of traditional non-compatibilist approaches. Such solutions have assumed that the non-human world consists of a network of causal connections, the links in which instantiate lawlike, exceptionless generalizations, but tries to show that humans, somehow, lie outside, or partially outside this web4. By contrast, I am suggesting that causal order is everywhere partial and incomplete. But humans, by virtue of their enormously complex but highly ordered internal structure, provide oases of order and predictability. Thus the significance of recognizing indeterminism is not at all to show that human actions are unreliable or random. It is rather to show that the causal structure that impinges on a human being, whether externally from macroscopic causal interaction, or internally, from constitutive microstructural processes, is not such as to threaten the natural intuition that humans are, sometimes, causally efficacious in the world around them.

This picture immediately accords with some obvious empirical facts: among the most apparently orderly features of the external world, such as straight roads and vertically stable edifices, not to mention complex machines, are products of human action; and among the most predictable entities in the world, as Hume, to a rather different purpose, argued, are people. Plans can be coordinated among many people, and complex human institutions can function, because human behavior is to a substantial degree reliable. All of this is quite unproblematic if we see humans as sources of causal order rather than either as exceptions to a universal external order or as insignificant components of some all-encompassing cosmic order. Thus a radical rejection of the traditional mechanistic assumption of causal completeness does indeed defuse the traditional problem of free will.

I shall expand on these claims at various points in this paper. Prior to that, however, the main task of the paper will be to render its presuppositions plausible. In the next part of the paper I want to argue that determinism, specifically microphysical determinism, really is a problem for an adequate account of human autonomy. Thus I reject the post-Humean compatibilism that holds no amount of determinism to provide any difficulty for freedom of the will. In the third section, I shall argue that we have, fortunately, no reason to believe in determinism--or even causal completeness, whether microphysical or any other kind. The paper will conclude with some further discussion of how I conceive the rejection of causal completeness to provide a way out of the traditional problem of free will.

Microphysical Determinism and the Causal Inefficacy of Everything Else

Suppose that there is some set of microscopic entities undecomposable into any smaller constituents, and of which all larger entities are composed. Assume that all putative entities that might appear not to be composed of anything (numbers, abstract objects, universals, etc.) are either wholly dependent for their existence and behavior on objects made of these microscopic entities, or non-existent. Though these suppositions could certainly be questioned, I believe that they would be widely accepted among the many philosophers who think of themselves as physicalists. Now suppose that we also have a fully deterministic account of the behavior of these microscopic entities. Although heroic attempts have sometimes been made to deny it, it seems to follow inevitably from this set of assumptions that the behavior of everything is fully determined by the laws at the microlevel. This seems to follow immediately from the assumption that objects at higher levels are composed entirely and exhaustively of the microscopic objects. For, given the assumption of determinism, it is true of every individual microscopic object that its behavior is fully determined by the laws governing microscopic objects. And surely if the behavior of every constituent of a thing is determined, so is the behavior of that thing.

This point can be made more graphic by thinking of a constituent of a human being, say an electron in my finger. I might be inclined to explain the movement of that electron by saying, for example, that I was reaching for a glass of water, and my hand brought the electron along with it. But clearly this explanation is going to have to be consistent, at the very least, with the explanation in terms of the microphysical laws acting on the electron. If we now consider the same condition applying to all the various electrons and suchlike in my arm, it would appear that only cosmic coincidence or some kind of dependence of the higher level on the processes at the lower level could insure this overall compatibility. The bold conclude at this point that either the higher level phenomena are reducible, in the sense of derivable, from the lower level phenomena, or they cannot really exist at all (elminativists). The more cautious fall back on claims of supervenience, though as far as I can tell this is merely reductionism with a modest reticence about the capacity of humans to carry it out. At any rate, none of these positions allows any genuine autonomy to the higher structural level5.

The importance of emphasizing the concept of causal completeness rather than merely determinism is that nothing is significantly altered in the preceding argument by moving from a deterministic to an indeterministic but complete system of laws at the microlevel. Given my intention to drink from the glass of water in front of me, the probability that the electron referred to in the preceding paragraph will move in a certain direction is very high. Again there must be some parallel explanation at the microlevel that also attributes a similar high probability to such a move. And again, when we aggregate all the particles that compose my arm, some explanation is required of the apparently extraordinary coincidence between the phenomena at the two levels.

My general point is just that causal completeness at the microlevel appears to entail reductionism, at the very least in the sense of the supervenience of everything else on the microphysical. And even supervenience, I claim, is sufficient to deny any real causal autonomy to higher structural levels. The alternative picture I would like to advocate denies causal completeness at any level. Objects at many, probably all, levels of the structural hierarchy have causal powers. One of the reasons why these causal powers are never displayed in universal laws (deterministic or probabilistic) is that objects at other levels often interfere with the characteristic exercise of these powers6. I take it that the example of the electron in my hand is best seen as such a case.7 If that is right, then the behavior of microlevel objects is very frequently consequential on processes at higher structural levels. As a simple example in the opposite direction, a person's plans can be seriously impeded by a dose of radiation. Elsewhere I have also advocated ontological pluralism at particular structural levels against the essentialism that tries to insist on a uniquely privileged position for one set of kinds. My present claim is that the same ontological tolerance should be accorded between structural levels. As objects are united into integrated wholes they acquire new causal properties (perhaps that is exactly what it is for a whole to be--more or less--integrated). I see no reason why these higher level wholes should not have causal properties just as real as those of the lower level wholes out of which they are constructed. But of course many philosophers have seen many such reasons, all grounded, I suggest, in the conceptual nexus that links determinism (or at least causal completeness) and reductionism. I believe that both of these doctrines are inherently implausible, but because the former is more widely, or at least explicitly, believed, and because it is more closely and traditionally associated with the question of human freedom, in this paper I shall focus exclusively on causal completeness. To this, the central task of this paper, I shall now turn.8

Causal Incompleteness

The thesis of this section of the paper is that there is no plausible ground for the belief in determinism. I shall address most of the argument to the doctrine of determinism, but I intend that everything I say will apply equally to indeterministic versions of causal completeness unless I explicitly differentiate the two cases. Later in the paper I shall say something about how I conceive of causal reality in the absence of the assumption of causal completeness. The basic strategy of my argument will be as follows. Presumably determinism is a very strong metaphysical assumption. To claim that everything that happened had to happen, given the totality of prior conditions, is to impose an enormously strong--indeed the strongest possible--restriction on the possible evolution of the universe. And even the claim that the state of the universe at any time fully determines a set of objective probabilities for its subsequent state is a strong assumption. My point is then that such strong assumptions require persuasive reasons if they are to have any plausibility. I do not take seriously the idea that determinism might be established by means of a transcendental argument of some kind, simply because, as I shall explain below, an indeterministic, causally incomplete, world seems to me entirely possible. Thus my question will be whether there is any basis in our empirical interaction with the world for supposing that it is causally complete. My answer will be in the negative.

There are two main kinds of experience that might be held to legitimate a belief in determinism. These are our familiarity with scientific laws and our everyday causal experience. An important special case of the latter is our experience of highly organized systems, especially machines and organisms. I shall deal with these topics in turn, but reserving the special cases of machines and organisms to a separate section. First, then, do the results of scientific investigation lend support to the idea that the world is deterministic, or at any rate causally complete? Here I must first dispose of an important red herring. It is often claimed that science must assume determinism as a methodological imperative. The idea is that it would be sheer defeatism, when confronted with a phenomenon anomalous in the light of current belief, to assume that this was simply a phenomenon outside the causal nexus. We naturally and correctly attempt to broaden our understanding of the range of phenomena in question so as to remove the appearance of anomaly. But that, it is claimed, is to assume that the anomalous phenomenon is in fact part of a uniform and complete causal nexus. Thus it might be suggested (and this is the non sequitur) that science must assume determinism; and then perhaps, that the successes of science provide evidence that the presuppositions of science, in particular determinism, must be true. But of course to say that science aims to explain phenomena does not entail that all phenomena can be fully explained. And to say that science has had some explanatory successes hardly implies that everything that happens can be fully explained as part of an underlying universal regularity.

So do the actual results of scientific research provide more direct evidence for determinism? The most compelling such results, for the reasons spelled out in the preceding section, would be those that provided evidence for casual completeness at the microlevel. But clearly there is no such evidence. Although certain very specialized phenomena in extremely carefully controlled conditions do exhibit some impressive regularities, this is the entire extent of such evidence. (As should become apparent later on, the fact that these regularities are produced in extremely elaborate machines--machines painstakingly designed for the very purpose of producing these regularities--is of great significance.) Evidence for causal completeness would require that increasingly complex systems of physical particles could be shown to be amenable to causal explanation in terms of the laws said to govern individual particles, evidence, that is to say, for general reductionism. I cannot here go into the general difficulties that confront the project of reductionism. But I do not need to do so. No one has claimed to be able to explain the behavior even of very small collections of particles in terms of the behavior of individual particles; the reduction even of relatively simple parts of chemistry to physics is now looked on with considerable skepticism; and even physics itself is acknowledged to consist of laws the relations between which are obscure, though at least the unification of physics is still looked upon by some physicists as an attainable goal. At any rate, the view that every physical particle has its behavior fully determined by microphysical laws must derive any plausibility it has from some source other than the development of microphysics.9

It appears then that microphysical determinism must be motivated, somewhat paradoxically in view of the connections between determinism and reductionism, by experience at the macroscopic level. But before turning to our everyday experience of causal regularities we might consider the possibility that microphysical determinism could be motivated by our knowledge of macrophysical laws. The obvious candidates, since they remain the most widely admired paradigm of scientific knowledge, would be the laws of Newtonian mechanics. But here we encounter exactly the same difficulty that we saw at the microlevel. Whereas scientists have been able to subsume very simple systems such as the solar system under impressively reliable regularities, the ability to apply Newtonian laws to more complex systems has proved severely limited. The notorious failure to solve the three-, let alone N-, body problem marks this failure. Thus we have no empirical evidence for the general truth of Newtonian mechanics as applied to complex systems of bodies unless we are prepared to countenance inductions grounded on one kind of case (very simple systems) to all cases, most of which are very different from those empirically studied. Moreover, to reiterate a point emphasized by Nancy Cartwright (1983), we know that laws such as those of Newtonian mechanics are true only under a very stringent ceteris paribus condition, a condition we know to be generally false. Thus, far from knowing that these laws are universally true, we know that they are generally false. The assumption that the laws of Newtonian mechanics are, in some sense, carrying on regardless under the overlay of increasingly many interfering and counteracting forces is sheer speculation. Thus this can hardly be a good empirical ground for the alleged universality of microphysical laws.10

The other common idea, mentioned above, is that determinism is evident from our everyday experience of causality. This assumption can be seen in classical regularity theories of causality from David Hume to J.S.Mill and J.L.Mackie11. Hume appeared to take determinism outside the human sphere to be so obvious as not to need much discussion. He was more concerned to show, with well-known examples such as the sure and swift appropriation of a purse of gold abandoned at Charing Cross, that humans were subject to regularities just as immutable as those governing the natural world. Mill was a good deal more sensitive to the complexities of regularities of the latter kind, realizing that the regularities of common experience could easily enough be defeated by either the absence of necessary background or auxiliary conditions, or by the presence of interfering conditions. Thus a lighted match thrown onto a pile of dry straw will always start a fire--unless, that is, there is no oxygen, or a fire extinguisher is simultaneously directed at the straw, etc. While thus acknowledging the complexity of everyday causal regularities, Mill appears to have thought that with sufficient care to include all the relevant auxiliary conditions and exclude all possible blocking conditions, a truly universal regularity could be discovered. This idea reached its most sophisticated expression with Mackie's analysis of an everyday cause as an insufficient but non-redundant part of an unnecessary but sufficient condition, or an "inus" condition. The sufficient condition in this analysis is the cause with all the auxiliary conditions and the negation of possible interfering conditions. The non-necessity of such conditions points to Mackie's additional recognition that there might be many such complex sufficient conditions of which none, therefore, would be necessary (a bolt of lightning might equally well have ignited the pile of straw).

Many objections can be raised against this picture, at least if it is assumed that it intends one to take seriously the universality of the implied laws rather than merely to illuminate the relations between miscellaneous items of causal lore. One may well doubt, to begin with, whether there is any definite limit beyond human imagination to the number of conditions that we might need to add to produce a fully universal generalization. More seriously, the more conditions are added, the further these putative regularities recede from any possibility of empirical support or refutation. Indeed the reason we are forced to move from simple regularities (e.g. lighted matches cause fires in flammable materials) to increasingly complex and qualified regularities is simply because we recognize the general falsity of the simpler ones. But as we move to such ever more complex regularities, first, the amount of evidence even bearing on the truth of the regularity will rapidly decline; and second, in keeping with the process that brought us the complex regularity in the first place, were we to find an exception to the complex regularity we would presumably respond by looking for a further interfering condition rather than by rejection of the entire regularity. This suggests that the Mill/Mackie program might better be seen as embodying a methodological rather than a metaphysical conception of determinism.

A second kind of objection casts doubt on the empirical basis of everyday causal determinism from a rather different perspective. Many everyday phenomena give no superficial appearance of being deterministic or even nearly deterministic. Consider, for example, a tossed coin. Now it is often asserted that this is a fundamentally deterministic phenomenon, and the only reason we are unable to predict the outcome is that we have an insufficiently precise knowledge of the initial conditions. It is much less clear why this is asserted. Presumably it must be because the kinds of laws involved in such a process (mainly Newtonian) are assumed to be deterministic. But I have already considered the weakness of that line of thought. The present case, since it is one in which we cannot in fact make any such predictions, provides further support for the argument against basing determinism on macroscopic scientific laws. At any rate, the thesis that everyday causal experience, suitably refined in the style of Mill and Mackie, provides grounds for the belief in determinism, simply ignores the fact that a great deal of our experience, whether of gambling devices such as tossed coins and roulette wheels, or just of seemingly quite erratic natural phenomena such as falling leaves or swirling smoke, provides no such grounds.

The final argument I shall mention is perhaps the most telling. It is that if there is causal indeterminism anywhere, it will surely be (almost) everywhere. Suppose, as is sometimes rather bizarrely suggested, that the only locus of indeterminism is in quantum mechanics. But surely--and here phenomena such as hypothetical quantum amplifiers in the brain have genuine significance--it must be impossible to insulate the indeterminacy of quantum events so fully from consequences at the macroscopic level. Consider again, for instance, the tossed coin, and suppose that its trajectory deterministically produces--ceteris paribus--its final outcome. Suppose the coin is at a point at which it is about to land heads. And suppose finally that a collision with a fast-moving air molecule is sufficient to reverse this outcome and produce a toss of tails. If the situation is sufficiently delicately balanced this must surely be possible. Assuming that the molecular trajectory is a sufficiently microscopic event to be subject to some degree of quantum indeterminacy, then we can easily see that the claim to determinacy of the coin-tossing event cannot be sustained. We cannot treat this as merely another interfering factor, because whether or not it has any effect on the final outcome cannot be determined by any amount of knowledge of the initial conditions.

It is a further advantage of this example that a coin toss is the kind of event that might imaginably have massively ramifying consequences. Perhaps the last degenerate scion of some aristocratic line is wagering his fortune on this coin toss. The outcome will dramatically affect the lives of his dependents, servants, creditors, etc. and their fortunes will have an increasing cascade of consequences. (This sort of thing will be familiar to readers of Victorian novels.) The general point that this argument is intended to illustrate is that indeterminism anywhere, by virtue of the variety of causal chains that might be initiated by an indeterministic event, is liable to infect putatively deterministic phenomena anywhere. It is significant that this applies equally within and across levels of structural complexity.

One final point will conclude this section. The last argument discussed is an argument against determinism, but not necessarily against causal completeness. In the case of the coin toss, provided only there is no correlation between interfering molecular events and outcomes, we should expect that these would be equally likely to change heads to tails and vice versa. So even if these interfering events occurred in accordance with no law even of a statistical nature, they might not render incomplete the supposed law that coins of a certain kind come up heads 50% of the time. On the other hand the preceding arguments, based ultimately on the lack of empirical support for determinism, seem if anything even more pressing against an indeterministic version of causal completeness. For any investigation of a range of phenomena will provide statistical facts. That, for some x, x% of events of type A are followed by an event of type B, is a matter of logic. But for this very reason, even if we have excellent grounds for believing that As really do have a tendency to produce Bs, it is difficult to see why we should be led to believe that there is any x such that it is a law that x% of As produce (or are followed by) Bs. The most plausible basis for such a belief, I suppose, would be microphysical reductionism, a topic about which I have said as much as I have space for here. We might better ask, What would it mean for there to be a law of this kind, as opposed to there merely being a tendency of As to produce Bs, and a statistical correlation of a certain strength between As and subsequent Bs? Ignoring for the present purposes a range of widely explored subtleties concerning spurious and genuine correlations, joint effects of a common cause, and so on, which would be required for a detailed answer to this question, the simple answer which is sufficient for my present purposes is just that a precise causal law should license us to expect that the proportions measured in a suitably large numbers of trials should be (approximately) repeated in the future. It seems to me, on the contrary that in practice such an expectation would often be foolhardy.

In real life, the degree to which we treat statistical experience as a guide to future expectations will vary from almost zero to almost unity. No doubt many explanations could be given of the reasonableness of such a perspective, some consistent with causal completeness. The explanation which seems to me most consistent with both investigative practice and the experience of causal regularity, however, has nothing to do with laws or statistical uniformities at all. Correlations reveal, I believe, (subject to well-known qualifications) the causal powers of certain objects or events to produce particualr effects. Whether we expect the production of such effects to occur with a fairly constant frequency depends whether we think that the frequency of other relevant causal factors is likely to remain reasonably stable. But without some apparently quite arbitrary way of privileging a particular constellation of background conditions, there is no such thing as the quantitatively precise, constant and timeless tendency of As to produce Bs ceteris paribus. Other things can be a particular way, and they can be more or less reliably that way. But except in the very simplest cases, as in Newtonian mechanics where we imagine there being only two bodies in the universe, and everything else is supposed not equal but absent, I do not know what everything else being equal even means. Thus once we have fully appreciated the complexity of the causal nexus, the thesis of indeterministic causal completeness is seen to be not only devoid of empirical support, but even to be of dubious intelligibility.

Machines and Organisms

As I have tried to show in the preceding section, I do not think that direct reflection on our (extremely limited) knowledge of universal regularities lends much support to the idea of a universe with a complete causal structure. However it may well be that deterministic intuitions derive more from reflection on complex and highly organized structures, especially machines and biological organisms. Since the overall metaphysical vision out of which the whole problem of free will arose is aptly referred to as mechanism, it is certainly appropriate to consider the artefacts that have somehow come to provide a model for the universe; the consideration of organisms, notoriously liable to be treated as a kind of naturally occurring machine, will bring us back to the topic with which this paper began, the causal status of humans.

It is easy enough to see why machines should have some tendency to inspire deterministic intuitions. Machines, good ones anyhow, are extremely predictable. I am confident that the text I type into my computer is exactly what will eventually come out of my printer when I connect them up in the right way. (Though not so confident that I do not occasionally make a hard copy; and some people, I am told, even make back-ups of their computer files on disks.) But a little further reflection makes it very puzzling that something like this, rightly admired as one of the great triumphs of modern technology, should be taken as a model for the universe in general. If the sort of regularity that is characteristic of a good computer or car were typical of the universe it would, one might imagine, be fairly easy to make, or perhaps even just find, such things. But it is not at all easy, which is why such technological achievements are admired. If the universe is a machine, it is far from obviously so.

Perhaps a more sympathetic interpretation of the tendency for machines to inspire determinism is the idea that only if determinism were true would it be possible to make reliable machines. And since we can make reliable machines, determinism is proven to be true. Underlying what seems to me a great exaggeration in the first premise there is, nevertheless, a very interesting question: what degree of order must exist in the world for the kinds of reliable machines we possess to be possible? The beginning of a more temperate answer to this question than the immediate appeal to determinism is the observation that no machines are completely reliable, and some are very unreliable. The point of this observation is not to insist--though strictly speaking it is no doubt true--that there is some possibility, however remote, that when I type the word "type" on my computer a four-letter obscenity will instead appear on the screen; or that when the spark ignites in the combustion chamber of my car the gasoline inside it will spontaneously liquefy. Rather I want to focus on the question, what is it that makes machines more or less reliable. And of course the answer is not, at any rate, that reliable machines have access to more universal laws.

Consider, then, what is by modern standards a fairly simple machine, an internal combustion engine. If we ask how such a machine operates we may be content with a very simple story: a mixture of air and gasoline is exploded in a cylinder, pushing a piston down the cylinder; the cylinder is connected to a shaft which is rotated by the moving piston. A number of similar cylinders are connected to this shaft, and a sequence of explosions keeps the shaft rotating continuously. It seems to me that this is, roughly speaking, a correct answer to the question how an internal combustion engine works. But if, on the basis of this explanation, someone lined up some coffee cans partially filled with gasoline on the kitchen floor, stuck toilet plungers in the cans and tied the ends of the plungers to a broomstick, and then posted lighted matches through little holes in the side of the coffee cans, they would certainly not have built an internal combustion engine (though I suppose the broomstick might jump about a bit).

I suggest that it is useful to think of how a machine works in two stages. First there is the question what makes it even possible for the machine to do what it is supposed to do. A slightly more elaborate version of the answer sketched in the previous paragraph might be an answer to this question for an internal combustion engine. Having got that far, however, most of the details of the internal combustion engine concern the more or less ingenious auxiliary devices that make sure it really does do what it is intended to do rather than one of the may other things it has an initial capacity to do. So, for instance, the cylinder must be strong enough to avoid simply disintegrating when the gasoline explodes; the crankshaft must be extremely strong and rigid if it is to reliably convert the linear momentum of the cylinders to rotational motion; piston-rings prevent the energy of the explosion from being dissipated between the piston and the cylinder; oil must be provided to prevent the cylinders getting so hot as to seize in the cylinder, or for that matter melt; some way must be found to dissipate excess heat from the running engine; and so on. Even a Trabant has the capacity to run, and sometimes does so. The difference between this and a well designed car is that the behavior of the parts of the latter is so tightly constrained that it can do nothing but what it is designed to do--though eventually, of course, even the best designed machine will break free of its constraints. My point so far is just that this kind of constraint is not something characteristic of nature generally, but something that engineers devote enormous efforts to attempting, never with total success, to achieve.

Of course, this account of the reliability of machines does assume the reliability of various causal relations. Gasoline and air mixtures invariably explode when sparked; heat will flow from a hot engine to cooling water circulating over it; and many others. It is interesting that many such regularities can be seen as reflecting the overall upshot of very large numbers of similar though indeterministic processes at the microlevel, which suggests the hypothesis that it is just those macrolevel processes that can be roughly reduced in this way that reveal this near determinism. But I do not want to insist on this here. While machines could presumably not work without exploiting extremely reliable regularities such as those just mentioned, the regularities that characterize the machines themselves, as with many other macroscopic causal regularities are only more or less reliable. Reflection on how good machines are engineered, far from making us think of mechanism as generally characteristic of the world, should make us realize how difficult it is to turn even little bits of the world into bits of mechanism.

Turning now to organisms, it is a familiar idea, especially following Descartes, that organisms just are machines. Natural theology until the late nineteenth century considered organisms quite explicitly as the products of a divine mechanic. No doubt there are aspects of organisms for which this analogy is illuminating. Indeed the complex but highly stereotyped performances of many insects in, for example, constructing and provisioning burrows for egg-laying have many of the characteristics of a well-designed machine. To the extent that the analogy is appropriate the same remarks that I made about machines will apply to the relevance of organisms to the prevalence of causal regularity. Looking, however, at the other end of the organic scale, and most especially at humans, the parallel with machines has serious limitations.12

The fact that when, for example, I intend to walk down the garden path, my legs move in just the right way to maintain my balance and propel me forward is, I suppose, something that could be explained in a manner strongly analogous to the performance of a machine, though perhaps more complex than any machine we have yet managed to construct. I suppose that the physiology and cell-chemistry of muscle tissue explains how the physical movements are obtained, and a variety of sensory and neural mechanisms bring it about that the motion is steady and in the right direction, and that a vertical posture is maintained. Although this seems significantly analogous to the account I offered of the working of the internal combustion engine, we should now note that an internal combustion engine is in reality not a machine but a part of a machine. If we think now not just of an engine but of an entire car, an important class of features has yet to be mentioned. I am thinking of such things as the ignition key, the steering wheel, and the brake pedal, those devices by which the machine is made to act in a way conducive to the ends of its human operator. A reliable car, as opposed to a reliable engine, the latter of course being a necessary but insufficient component of the former, is one in which there is a reliable correlation between inputs to these controls and the behavior of the whole machine. Thus machines are not sources of causal autonomy; they are, at most, instruments for furthering the causal autonomy of their users. The superficial, and I think also deep, disanalogy between humans and machines is that humans have no controls.

It may rightly be objected at this point that insects with simple stereotyped behaviors have no controls either, yet I have claimed that they are closely analogous to machines. There are two possible responses. First, a stereotypical performance might simply be produced in response to nothing at all. More typically and interestingly, a kind of behavior might be triggered by some sensory input, the sense organs thus serving as devices for producing behavior appropriate to the external circumstances. This is primarily what I have in mind in talking of the stereotypic and machine-like behavior of certain insects: a certain stimulus triggers a sequence of behavior. There is, of course, a tradition of psychological investigation of humans that applies just this model to humans. Though in it crudest behaviorist versions it has been almost wholly rejected, the idea that sensory inputs, mediated by "information-processing mechanisms," somehow elicit the appropriate "emission" of behavior is still widely, perhaps generally, pursued. This is a mechanistic model, though one in which the complexity of the machine is such that we as yet have no idea what it is designed to do in the innumerable situations it encounters.13 Against this model, I propose that we should recognize that we were not designed at all, and consequently there is nothing we were designed to do in any situation.

Between two views that I have rejected, that we are random action generators and that we are machines, can be found the view that makes sense of human autonomy. Many parts of humans have just the characteristics of machines that I have emphasized in the preceding discussion, namely complex constraints that insure the predictable exercise of somecapacity of an organ or physiological system. But humans are fundamentally different from machines in that they have no controls. Self-control, in the sense of the absence of external controls, is of course nothing but the autonomy, or free will, that was the original topic of this paper. I have not attempted to refute the idea that sense organs might sometimes function as controls, in the sense that the input to sense organs might determine, via a complex intermediate causal chain, the behavior of the whole organism. This is presumably roughly true of simple organisms. But it does not appear to be true of ourselves, except perhaps in purely reflexive actions, such as ducking to avoid a flying object. The reason we are so liable to think of ourselves in this machine-like way is because we are tempted by determinism. If the world is deterministic then my behavior is causally necessary given the stimuli that impinge on me; and presumably the most important stimuli are sensory ones. The point of all the complex machine-like parts of me would then have to be just to make sure that the causally elicited behavior was appropriate to the circumstances disclosed by my sense organs. But the rejection of causal completeness allows a more natural view of things. My complexity of structure gives me a vast array of causal powers, a range of powers that would be inconceivable without that intricate machine-like internal structure. But the exercise of those powers, though obviously influenced by the circumstances I perceive myself to be in, ultimately depends on an autonomous decision-making process. Once we see causal order as something special rather than something universal, there is no obstacle to seeing the human will as an autonomous source of such order.

Some Concluding Remarks

My reference at the end of the last section to the human will leaves, needless to say, some questions unresolved. However, it is not my goal here to offer a detailed account of the will. Rather, my more modest aim is to show that, contrary to a notorious tradition of philosophical controversy, a reasonable metaphysics of causality presents no special difficulties for the idea of human autonomy, and requires neither ghostly nor random nudges of the physical causal order. In order to give a little more positive philosophical substance to the view I shall, in this final section, very briefly relate what I have said to some very famous views on the subject, those of Hume and of Kant.

Although it may well remain the dominant view of the subject, Hume's attempt to reconcile human autonomy with a classically deterministic structure of causal relations seems to me unconvincing14. On the other hand Hume was surely right that exercises of human freedom were much better understood as instances of causality than of its complete absence. His problem, in my view, was his commitment to a universalistic regularity theory of causality. Given such an account of causality, any departure from determinism is a failure of causality itself.15 The solution I have advocated requires espousing an ontology of causal powers of the kind that Hume so famously attacked.16

Kant also appears to have thought that a deterministic causal structure was compatible with human autonomy, though the metaphysical excesses to which he was led in effecting this reconciliation have convinced almost no one. However his conception of human autonomy might provide--though here I end on a more speculative note--a vital and final piece in the picture that I wish to present. My point at the end of the last section was that human decision could be a real source of causal order in the world. However, this claim may seem shallow without some further account of the origins of this order. In particular if one traces human decisions ultimately to contingent human desires, desires which presumably can themselves be traced either to our biological heritage or our upbringing, human autonomy seems at best a focus rather than a source of order. And just such a conception of human decision-making has been cultivated for a century by economists and more recently by exponents of so-called rational choice theories in a variety of disciplines. Some more interesting account of the ultimate springs of human behavior than the economist's standard refrain, "tastes are exogenous," is needed if we are to provide a deeper and more interesting conception of human autonomy.

Without going into what I take to be the deficiencies of the economistic vision of human behavior, it is clear that something more than an unexplained appeal to particular tastes or preferences is required to give interest to the account of human autonomy suggested in this paper. It seems to me that Kant's account of human action suggests a promising direction in which to go. Kant, notoriously, distinguished sharply between action motivated by desire and action motivated by principle. Having dispensed with the deterministic framework, it is possible to emphasize this distinction without either the extreme moralistic distinction in favor of action motivated by duty, or the metaphysically murky appeals to the noumenal world, which have combined to cast deep suspicion on Kant's conception.

In a world where order is a local and incomplete phenomenon, the importance of principle as a source of human action is easily stated: it explains how ideas, the creative acts of the human mind, can change the world. Unlike Kant, I do not make a fundamental distinction here between moral and more mundane principles. I conceive of the principle, "Follow the architect's blueprints in determining where to build the wall," as as genuine a source of autonomous action as "Do whatever is necessary to end hunger." But despite this moderation of the Kantian position, the enormous importance of moral principles in this context should not be downplayed. The most fundamental reason why we should care about human autonomy is that it holds out the hope that human action might produce a better world. And what that requires is action grounded in moral principles. This is something I believe we are free to choose; and making this choice, I claim, can make a difference.

I conclude with one further speculative and perhaps paradoxical suggestion. Principles, I take it, are essentially linguistic phenomena. And language is essentially social. Thus the condition for genuinely free individual action is the embedding of the individual in society. This will not seem surprising to those who take seriously the fundamental biological fact that Homo sapiens is a social animal. It may, however, be an unwelcome suggestion for the tradition that connects human freedom with the profoundly individualistic social philosophy and metaphysics dominant in contemporary English-speaking culture. That, however, is not my concern.17


1. No doubt the belief in such indeterministic events was often also connected with the inchoate hope that these might be sufficiently loose and microscopic that even an immaterial soul might have a chance of subtly influencing them. Though this make the idea less absurd from the point of view of understanding human autonomy, it intorduces new absurdities that I cannot attempt to address here.

2. This point was clearly stated by C. D. Broad (1952).

3. And presumably for this reason the possibility of indeterminism does not figure largely in recent discussions of the problem of free will. In a recent anthology on the topic (Fischer 1986), I could find only one extended discussion of the topic, in the paper "Asymmetrical Freedom" by Susan Wolf. However even this discussion concerns only the failure of determinism at the psychological level, and remains agnostic about the relation of this to underlying physical determinism. Because of this difference in focus, I shall not try to relate the present discussion in any detail to recent philosophical work on the topic.

4. A classic statement of such position is that of William James (1884/1956).

5. This is at the basis of Kim's well-known arguments against non-reductive physicalism (Kim1993, especially essays 14 and 17) . Kim shows that such a position requires "downward causation" the causal influence of macroscopic on microscopic entities. I accept the argument but, as will be clear below and as I have explained elsewhere (1993), I see no problem with downward causation.

6. As of course may objects at the same level. This kind of objection to universal regularities has been emphasized by Cartwright (1983).

7. It will be objected--and may already have been objected--that the electron will be pushed by the microscopic object or objects immediately behind it and will push those in front of it, and thus all the particles are moving in response to microlevel forces. I do not mean to deny this: certainly it would be absurd to suppose that my intention independently acted on each particle in my arm. The real issue is whether all these arm-particles are moving as part of a much wider set of microphysical events (photons bouncing of the glass, hitting my retina, stimulating my brain, etc.) on which my intention to drink the water is ultimately a mere epiphenomenon, or whether, rather, the fundamental explanation for all those particles pushing one another in a certain direction is that I am thirsty and see a glass of water I plan to drink. Evidently I prefer the latter view.

8. The various theses referred to briefly in this paragraph are defended in detail in my book, The Disorder of Things: Metaphysical Foundations of the Disunity of Science (1993). Part 1 of that work addresses essentialism, part 2 provides a detailed critique of physicalist reductionism, and part 3 provides a more extended version of the arguments against causal completeness developed in the following section of the present paper.

9. It is of course true that microphysical laws purport to apply to indefinitely complex systems, in the sense that they determine how the formalism should, in principle, be applied to such systems. But in practice they certainly cannot be so applied. And one need hardly be a radical skeptic about induction to resist extrapolation from a very narrow and limited set of data to every phenomenon whatever that could in principle be subsumed under the purported regularity.

10. See Suppes (1994) for a more detailed argument complementary to the present one.

11. See principally Hume (1748), Mill (1875), and Mackie (1974).

12. I focus here only on what I take to be the extremes of the animal scale. I assume that higher mammals, birds, and perhaps higher molluscs, are more like humans than they are like the most machine-like of insects. But I shall make no attempt to here to draw any more specific lines between different kinds of organisms.

13. Sociobiologists, evolutionary psychologists, and other extreme enthusiasts for explanations in terms of natural selection believe that we have been designed to survive and reproduce. The crudity of attempts to explain human behavior in any detail on the basis of this thesis has been well documented--if inadequately appreciated--and will not detain us here.

14. I cannot begin to discuss the enormous literature on this question. Strawson's (1974) classic paper perhaps brings out as clearly as possible the consequences of taking physical determinism fully seriously.

15. Recently there has been a prominent movement to provide regularity theories of indeterministic causality (e.g. Eells 1991). I think there are deep internal problems with such a position (see Dupré and Cartwright 1988; Dupré 1993, ch.9); and in defence of my commitment in the present paper to address causal completeness more generally than in its deterministic version, I think that such a position is fatally lacking in empirical backing.

16. A detailed response to the Humean view of causality would be beyond the scope of this paper. Some recent advocates of causal powers include Harré and Madden (1975), Cartwright (1991), and Dupré (1993, ch.9).

17. I am indebted to Regenia Gagnier, John Perry, and Debra Satz for comments on an ealrier version of this paper.


Cartwright, Nancy. 1983. How the Laws of Physics Lie. Oxford: Oxford University Press.

Cartwright, Nancy. 1990. Nature's Capacities and their Measurements. Oxford: Oxford University Press.

Dupré, John. 1993. The Disorder of Things: Metaphysical Foundations of the Disunity of Science. Cambridge: Harvard University Press.

Dupré, John and Nancy Cartwright. 1998. "Probability and Causality: Why Hume and Indeterminism Don't Mix." Nous 22: 521-536.

Eells, Ellery. 1991. Probabilistic Causality. Cambridge: Cambridge University Press.

Fischer, John Martin (ed.) 1986. Moral Responsibility.

Harré, Rom and E. H. Madden. 1975. Causal Powers. Oxford: Blackwell.

Hume, David. 1748. An Enquiry Concerning Human Understanding. Reprint, Indianapolis: Hackett, 1977.

James, William. 1884/1956. "The Dilemma of Determinism." In The Will to Believe. New York: Dover.

Kim, Jaegwon. 1993. Supervenience and Mind. Cambridge: Cambridge University Press.

Mackie, J. L. 1974. The Cement of the Universe. Oxford: Oxford University Press.

Mill, J. S. 1875. System of Logic. 8th. Edition. London: Longmans.

Strawson, P. F. 1974. "Freedom and Resentment." In Freedom and Resentment. London: Methuen.

Suppes, Patrick. 1993. "The Transcendental Character of Determinism." Midwest Studies in Philosophy 18: 242-257.