Notions of Cause : Russell ’ s thesis revisited ∗

We discuss Russell’s 1913 essay arguing for the irrelevance of the idea of causation to science and its elimination from metaphysics as a precursor to contemporary philosophical naturalism. We show how Russell’s application raises issues now receiving much attention in debates about the adequacy of such naturalism, in particular, problems related to the relationship between folk and scientific conceptual influences on metaphysics, and to the unification of a scientifically inspired worldview. In showing how to recover an approximation to Russell’s conclusion while explaining scientists’ continuing appeal to causal ideas (without violating naturalism by philosophically correcting scientists) we illustrate a general naturalist strategy for handling problems around the unification of sciences that assume different levels of naı̈veté with respect to folk conceptual frameworks. We do this despite rejecting one of the premises of Russell’s argument, a version of reductionism that was scientifically plausible in 1913 but is not so now. 1 Russell’s Naturalistic Rejection of Causation 2 Psychology, Folk Notions and Intuitions 3 Causes in Science 4 Letting Science Hold Trumps 1 Russell’s Naturalistic Rejection of Causation Russell ([1913])1 characterized what he called the ‘law of causality’ as a harmful ‘relic of a bygone age’, and urged the ‘complete extrusion’ of the word ‘cause’ from the philosophical vocabulary. His reasons include the descriptive claim that practitioners of ‘advanced’ sciences, particularly physicists, do not seek causes, and the normative claim that it is improper for philosophers to ∗ We thank John Collier, Harold Kincaid, James Ladyman, Adriano Palma, David Papineau, and an anonymous referee for this journal for comments and criticism of earlier versions of this article. 1 All references to ‘Russell’ in this article are to ‘the Russell of 1913’ unless otherwise stated. Citations are to ‘On the Notion of Cause’ as reprinted in Mysticism and Logic (1917). The Author (2007). Published by Oxford University Press on behalf of British Society for the Philosophy of Science. All rights reserved. doi:10.1093/bjps/axl027 For Permissions, please email: [email protected] 2 Don Ross and David Spurrett legislate on whether they should. His claim about advanced sciences construes cause-seeking as the quest for, or assumption of the existence of, ‘invariable uniformities of sequence’ ([1913], p. 178) or laws of constant succession. According to Russell, belief in the ‘law of causality’ leads us erroneously to expect and to prize these uniformities. Our aim is to re-evaluate Russell’s thesis in the light of subsequent developments in science and philosophy. Our motivation is not Russell scholarship but interest in philosophical naturalism. By ‘naturalism’ we refer to any thesis according to which metaphysical hypotheses should be derived, as far as possible, from scientific discoveries and arguments, rather than extra-scientific sources like ontological intuitions, ‘common sense’ or ‘first philosophy’.2 Naturalism is, then, an epistemological doctrine that will lead its proponents to favour some ontological views over others. The adequacy of such naturalism has lately become the focus of lively philosophical debate. (See McDowell [1994], Almeder [1998], Rea [2001] for the tip of the iceberg). Though we promote naturalism, this article is not a polemic in its favour. Rather, it is an exploration, focussed by way of the topic of causation, of a perennial naturalist issue: the conflict between what Sellars famously referred to as the manifest and scientific images. The naturalist is acutely aware that there is a difference between the world as it appears to so-called ‘common sense observation’ and the modally structured, perspective-independent world that science aims in the limit to describe, and holds that insofar as there can be any legitimate role for metaphysics in the description of reality, it must develop its concepts and arguments by reference to the scientific account. This interest led us to Russell, whose 1913 essay on causation is a particularly illuminating (partly because of its occurrence in a different philosophical era) expression of this contemporary naturalist problematic. Let us first indicate what makes Russell’s 1913 argument representative of the sort of naturalism we have demarcated. His contention that causation is not a significant concept in metaphysics is partly based on his claim that it is an artefact of an anthropocentric perspective that science supersedes. He also blames philosophers for supposing, against the evidence in science, that there exists a ‘law of causality’ identifying causal relations with ‘invariable uniformities of sequence’. Here Russell anticipates Cartwright’s (e.g., [1980]) contention that laws understood as generalizing or quantifying 2 An anonymous referee wondered whether we fully intend the force of the idea that metaphysical hypotheses should be derived from science as opposed to something weaker such as ‘informed by’ science or the yet weaker ‘compatible with’ science. ‘I don’t see,’ writes the referee, ‘why a naturalistic metaphysician cannot think that philosophical modes of thinking do provide distinctive evidence of their own that bear on metaphysical hypotheses.’ Ladyman and Ross ([forthcoming]) give arguments for doubting that these weaker forms of naturalism are viable. Therefore, it is the strong form that interests us here, and we intend the formulation seriously. Russell’s Thesis Revisited 3 event regularities have numerous counterexamples. Russell agrees with Cartwright that such laws do not describe how bodies actually behave. Instead of concluding from this that the laws of nature are false, he maintains that they are not about event regularities per se, defending this through an account of the features of scientific practice that both generate the counterexamples and explain the inappropriateness of an event-regularity view of laws. First, he argues, the Humean interpretation of laws as regularities depends on a qualitative basis for individuating events that cannot be rendered in objectively quantitative terms; yet a mature science is one that aims at strictly quantitative description in the limit ([1913], p. 183f). Second, the idea that uniformities of sequence might be ‘laws’ makes asymmetry between the past and the future fundamental, whereas maturing sciences aim always at laws that avoid reference to time in this sense (though they often refer to intervals, some of which may be rightly called ‘temporal’) ([1913], p. 185f). Russell concedes that in the ‘infancy’ of a science ([1913], p. 178) the principle of ‘same cause, same effect’ may prove useful for discovery, but maintains that advance in science consists in moving away from such simple formulae, and towards understanding phenomena in terms of mathematically specified relations of functional interdependence in which ‘past’ and ‘future’ are not systematically different,3 which the ‘law of causality’ is too crude to embrace, and which Russell therefore thinks should not be called ‘causal’. In calling Russell’s argumentation here ‘naturalistic’, we do not mean to imply that Russell consistently observed naturalism as his first allegiance. He was first and foremost a Platonist. Though Platonism has sometimes been included among the foils of naturalism, there are in fact versions of Platonism compatible with naturalism (Brown [1999]; Ladyman and Ross [forthcoming], Chapter 4); and Russell’s Platonism was motivated by facts about mathematics and its relationship to science, not by a priori logic. In any case, Russell’s main basis for argument in the essay on causes is description of the general content of science. Such appeals have no force without an accompanying normative thesis to the effect that science should have authority over philosophy. Thus Russell cites James Ward as complaining about physics on the grounds that ‘the business of those who wish to ascertain the ultimate truth about the world . . . should be the discovery of causes, yet physics never even seeks them,’ to which Russell objects by saying that ‘philosophy ought not to assume such legislative functions’ ([1913], p. 171). Later, having found Bergson attributing an event regularity principle of causation to scientists, Russell scolds that ‘philosophers . . . are too apt to take their views on science from each other, not from science’ ([1913], p. 176). Towards the end of his essay, Russell diagnoses commitment to a law of causation as stemming from projections into the metaphysical 3 That is, as Field ([2005]) notes, being ‘deterministic in both directions’. 4 Don Ross and David Spurrett interpretation of science of ‘anthropomorphic superstitions’ based on the practical human predicament with respect to the asymmetry of past and future in our memories and capacities for control. Finally, at one point ([1913], p. 184) he suggests a very radical naturalist response to the conflict between everyday and scientific conceptions: the idea that sound metaphysical insights cannot be stated in natural language, but require expression in a formalism from which anthropocentric distortions have been purged. Histories of naturalism in philosophy typically describe Hume as the original radical naturalist, and then, perhaps after a nod at Peirce, leap straight to Quine and the subsequent rise of naturalized epistemology inspired by cognitive science. This pattern provides part of the interest in finding such forthrightly naturalistic argument coming from Russell in 1913. Further, Russell’s argument, like that for the version of naturalism most often debated by philosophers of science today, derives its conclusion mainly from putative facts about what scientific theory implies there to be, rather than from taking human psychological dispositions to construe experience in particular ways as evidence that the construals in question have some content that prima facie merits preserving in a more sophisticated account. (Certain kinds of reflections on typical human psychology are relevant to any version of naturalism, including Russell’s. We return to this in the next section.) We think we can learn important things about naturalism in the philosophy of science by examining its antecedent expression in the quite different context of 1913. Let us say something about the context in question. Condensing breathlessly, what mainly preoccupied Western philosophy of science between Hume and Quine were two bodies of ideas: Kantianism and positivism. Indeed, as Friedman ([1999a]) has persuasively argued, positivism was initially a variant of Kantianism. Quinean naturalism, anticipated and directly inspired by the later Carnap (Creath [1991]), emerged from the gradual morphing of Kantian positivism into Humean empiricism during the 1930s and 40s. Kantians, including Kantian positivists, privilege science over philosophy in one sense: they agree that science authoritatively informs us about empirical matters. However, they reject naturalism as we understand it, in supposing that philosophy retains a distinctive, extra-empirical, task of elucidating the forms of understanding under which the empirical world is ‘grasped’. Positivistic Kantians differ from non-positivists in denying a transcendental basis for this elucidation, reducing the forms of understanding to conventions. Thus the nonnaturalistic commitment of Kantianism is its belief that philosophers of science have a distinctive task—either transcendental inquiry or logical analysis—that can and should be pursued independently from empirical science (Friedman [1999b]). Russell, writing before positivism took hold, represents one last expression of a version of naturalism descending directly from Locke and not passing through Kant. It is an interesting question, one we will not directly Russell’s Thesis Revisited 5 pursue here, whether some contemporary naturalists might be rediscovering this abandoned thread. We have identified the naturalist as someone who holds that metaphysical hypotheses should be derived from science. Naturalists sharing this commitment may yet differ sharply on what science in fact tells us, and how, if at all, metaphysical conclusions can be derived from it. Some with good claim to be called naturalists (at least with respect to the philosophy of science) such as van Fraassen ([1980]) and Fine ([1986]) oppose metaphysics entirely. van Fraassen’s constructive empiricism retains the positivist’s suspicion of metaphysics, seeing no role for philosophy as a unifier of science and expressing this through denial that we have any sound reason for believing that there is an objective modal structure to the world (van Fraassen [2002]): there are simply observable4 events, and models by which we simplify and summarize parts of the observational record that are salient to us and relatively stable in our experience. This position suggests the retirement of the philosopher. Within the ranks of those naturalists who think there is work for philosophers, we note a significant division between those who think that science suggests a unified account of reality and those who think it does not. Cartwright’s (e.g., [1999]) view that the world is ‘dappled’ is the leading contemporary instance of the latter mood of naturalism. In the present article we do not argue directly against the view that the world is dappled, instead following a line of argument internal to the concerns of ‘unificationist’ naturalists. Claims about the way in which the world might be unified necessarily amplify our empirical data. Unificationist naturalists hold that this amplification can be legitimated in the way that ampliative inferences are justified within scientific disciplines, using only data endorsed by science. (Disagreements over the acceptability of these ampliative inferences are, of course, part of what is at issue with the anti-unificationists.) What would make some facts about the world—the unifying ones—more general than those they unify would be that they are projectible from data drawn from domains of greater generality. So, perhaps, only fundamental physical facts, true everywhere in the universe at least since soon after the Big Bang, can unify branches of non-fundamental physics that describe special aspects of our galaxy. In this sense—the sense of their being highly general facts to which less general would-be facts are ‘answerable’—the world for the unificationist must be modally structured, even if the existence of the universe itself might be ultimately contingent.5 This is the position that Ladyman ([1998], [2000]; French and Ladyman 4 See Ladyman ([2000]) for argument that van Fraassen, as a modal anti-realist, is inconsistent in concerning himself with whether science is explanatorily adequate to observable (as opposed to observed) events. 5 That is to say: a physicist (for example) might, in working up a theory, treat some relations as holding necessarily given that there is any universe, or any universe containing matter, or any 6 Don Ross and David Spurrett [2003]; Ladyman and Ross [forthcoming], Chapters 2 and 3) has baptised and defended as ‘ontic structural realism’. The naturalist metaphysician differs from the special scientist in combining sets of data that individual sciences leave separate. Needless to say, a scientifically justified hypothesis about the way in which the world is unified is fully as fallible as any other generalization inducted from data.6 The Russell of 1913 is a unificationist naturalist in this sense. He denies ([1913], p. 185f) that scientists make inferences from the hypothesis of the ‘uniformity of nature,’ because such a procedure would treat this hypothesis as at least implicitly a priori. However, he then goes on directly to argue ([1913], p. 186f) that science advances generalizations about ‘relatively isolated systems’ only for practical purposes: its guiding aim is to achieve generalizations of maximum scope. Thus ‘it should be observed that isolated systems are only important as providing a possibility of discovering scientific laws; they have no theoretical importance in the finished structure of a science’ ([1913], p. 187). Despite anticipating Cartwright’s criticism of the event regularity interpretation of laws, then, Russell is not led from this toward the view that the world is dappled. As sciences widen the scope of the generalizations they produce, he says, they abandon simple regularities among ‘causes’ and ‘effects’ which they used as ladders to discovery in their immature phases, and identify progressively ‘greater differentiation of antecedent and consequent’ and a ‘continually wider circle of antecedents recognized as relevant’ ([1913], p. 178). Where more recent work in naturalistic metaphysics has been preoccupied with questions about unification, this has usually been in the specific context of trying to understand the relationship between causation and explanation (Friedman [1974]; Kitcher [1981], [1989]). Whether or not all science reduces to physics (in which case responsibility for providing unity would fall to a single discipline) the naturalistic metaphysician concerned with unity addresses questions about whether, or in what ways, the findings of the various sciences ‘hang together’. In this she supposes that the distinctions amongst the sciences are ontologically significant—that is, reflect real distinctions in the structure of the world, rather than just epistemic or institutional conveniences for us. In recent debates, a flagship instance of which is Kim ([1998]; see also Walter and Heckmann ([2003])) a crucial aspect of this problem arises from the multiplicity of causal claims, and types of causal claims, made by the various sciences. Where there are multiple (for example mental and neural) causal universe in which the value of Planck’s constant is what it is, or etc., while allowing that the relevant condition on which the necessity in question is premised is an accident. 6 Naturalists claim only that science pursues a unified account of the world as its aim; they do not assert that we can know, at least in advance of the completion of science, that the world in fact has a fully general unifying structure. Russell’s Thesis Revisited 7 claims about what happens in some part of the world, it seems to many philosophers that some of them must be false, unless we accept reduction or overdetermination. This is known in recent analytic metaphysics as ‘the causal exclusion argument’ (CEA). Naturalistic anti-reductionists must, if they deem it valid, regard the conclusion of the CEA as a reductio against the conjunction of its premises. For example, Kincaid ([1997]) takes as his central problem the challenge to unification of the sciences raised by the failure of reductionism, and by the fact that (he argues) different sciences display a proliferating range of explanatory strategies distinguished by reference to special, parochial causal patterns—patterns which indeed provide much of the basis for the locations of the boundaries between the scopes of the sciences in the first place. Because of the priority Kincaid attaches to the empirical evidence against reductionism, he regards the CEA as a reductio against the premise that where there are multiple (for example, both mental and neural) causal claims about what happens in some part of the world, all but one of these must be false unless we accept reduction or overdetermination. In later work, Kincaid ([2004]) makes realism about causes the basis for an argument that special sciences (including social sciences) discover laws of nature, despite the fact that the laws in question must usually be stated with ceteris paribus clauses. Note the point of agreement here: we find a reductionist champion of the manifest image (Kim) agreeing with an anti-reductionist naturalist (Kincaid) that the scientific worldview obtains such unity as can be had by reference to causation. In the context of this current dialectic, it is interesting that Russell combines the commitments to naturalism and unification shared by us and Kincaid with the latter’s reductionistic physicalism, and to the mereological atomism embraced by the CEA’s champions. In our view, atomism and naturalism were compatible when Russell wrote his essay on causes but were about to be rendered incompatible by the progress of fundamental physics. 1913 was the very year in which the old quantum theory was consolidated with Bohr’s quantization of Rutherford’s theory of atomic orbits. The old quantum theory, in turn, made the late nineteenth-century impasses that had inspired Mill’s emergentism about chemistry (and, by implication, the life and behavioural sciences) look much less likely to remain pressing problems. Specifically, the key breakthrough of old quantum theory, Einstein’s introduction of photons, made it seem likely that wave-particle duality with respect to light was about to be dissolved in a way favourable to atomism. That the duality would instead be generalized to all matter by de Broglie—let alone that quantum entanglement would be discovered—was something Russell could hardly anticipate. Russell kept a close eye on the frontier of science, however. By the time of his ABC of Relativity a few years later (Russell [1925]), he knew that de Broglie’s work spelled trouble for metaphysical atomism. As Ladyman 8 Don Ross and David Spurrett and Ross ([forthcoming]) demonstrate, many analytic, metaphysicians writing now still do not seem to know that entanglement kills classical atomism in fundamental physics stone dead, and the relevance to actual science of at least Kim’s version of the CEA crucially depends on this ignorance (see also Glymour [1999], Ross and Spurrett [2004]). These considerations might suggest at first glance that Russell’s 1913 essay is unlikely to be of anything but purely historical interest. Furthermore, the fact that the one point of agreement we just identified between the naturalist Kincaid and his (perhaps unwittingly) anti-naturalist opponent Kim, that causation is ontologically fundamental, constitutes prima facie grounds for supposing that Russell is simply incorrect about the irrelevance of causes to science and metaphysics. Those who now look to causation to provide the unifying structure of the world have in mind a notion of cause as in some sense the ‘cement of the universe’ (see Hume [1978], p. 662 for the leading source of this metaphor) or what we elsewhere (Ross and Spurrett [2004]) call the ‘glue’ holding all objective relations in place.7 The causation as glue tradition contains a substantial body of work, developing what are collectively called ‘causal process theories’. Such glue would be or would provide the necessity that Hume could not find in the impressions and their regular relations. Causal process theories, that is, can be understood as attempts to answer Hume’s epistemological challenge to say how anyone could know, by any amount of observation, which links between processes are causal and which are not. They purport to show that we can at least observe something that is precisely diagnostic of what we have traditionally meant by ‘causation’. Contra Russell’s complaints about the law of causality, causality, or the glue, is not itself an event regularity; it is supposed to be that in virtue of which there are projectible regularities in the first place.8 Salmon’s ([1984]) causal process theory followed Reichenbach ([1957]) in describing real processes in terms of the transmission of marks.9 According to this view a genuine process is one that can be modified (or ‘marked’) at some stage, and observed to carry the same modification at subsequent stages. Some processes, that is, transmit information about their antecedent stages, while others do not; those which do are genuine processes, the others pseudoprocesses. A crucial scientific motivation for Salmon’s project of distinguishing 7 An anonymous referee observed that cement and glue might be regarded as inapt metaphors to the extent that causation is supposed to be dynamic, while the metaphors suggest stasis. This is grist to our mill, insofar as not all naturalistically respectable lines of thinking about ‘causation’ are primarily dynamic. See the final two sections of the present article. 8 To suppose that unification might consist in the identification of some privileged ‘glue’ is, then, to suppose that there is a master relation (in this case causation) that is implicated in all projectible generalizations. 9 Process theories differ in what they take real processes to consist in. For Dowe (e.g., [1992]) they are possessors of conserved quantities, and for Collier ([1999]) they are transfers of information. Russell’s Thesis Revisited 9 causal from pseudo-processes, as with Reichenbach, is the alleged relativistic requirement not to count a faster than light process as causal (Salmon [1984], p. 141; Reichenbach [1957], pp. 147–9). A commonly used illustration of the problem concerns a spot of light moving on some surface because the light source is rotating. For large enough distances between spotlight and surface, the spot cast by the rotating light will move across the surface at a superluminal velocity (Salmon [1984], p. 143). Even so, it would not be a causal process for Salmon, because it cannot be marked. A filter that turned the beam red, placed just before one part of the surface, would make the light that struck that bit of surface red (diagnosing the process between the filter and surface as causal) but not change the spot elsewhere, showing that the successive stages of the spot were not causally connected. The metaphysic suggested by process views is effectively one in which the universe is a graph of real processes, where the edges are uninterrupted processes, and the vertices the interactions between them. Thus process views, if correct, would make putatively causal claims by scientists subject to a critical test. Those that pick out real processes could be causal, those that do not, cannot. Then, when scientists find causes in their domains of enquiry, whether they put the matter this way or not, they will find ways of picking out real information-carrying relations. That is to say, if macroeconomics, for example, has its own notion of ‘macroeconomic causation,’ this notion will turn out to correspond to some information-carrying relations really instantiated on the general graph of real processes. Russell’s 1913 claims about the irrelevance of cause to science and the tradition of causal process theories are, then, in conflict. It is bound to seem at first glance that unless Russell is completely wrong, the process theory tradition must be a mistake. There are at least two central reasons why most contemporary philosophers are apt to regard Russell as wrong. First, Russell, on the other side to us of decades of reflection and debate over forms of reductionism and the autonomy of the special sciences, takes for granted a now quaint-seeming view (among philosophers of science though not, unfortunately, analytic metaphysicians—see Ladyman and Ross [forthcoming], Chapter 1) of the capacity and right of physics, rather than science more generally, to populate our ontology. As we have explained, attention to the proliferation of non-reducing patterns of causal explanation in special sciences largely defines the contemporary naturalistic (unificationist) metaphysician’s mission. For this, process theories are one possible piece of useful ordinance. Second, Russell relies on a Humean conception of causation in pronouncing for eliminativism. Many, perhaps most, contemporary philosophers of science would agree with him that scientists do not seek causes in that sense, but would then go on to say that that is not the sense in which the idea of causation is either scientifically or 10 Don Ross and David Spurrett metaphysically interesting. For these reasons taken together, Russell’s thesis might be written off as of historical interest only. Yet Russell’s thesis is not dead. Some contemporary philosophers of physics continue to endorse and argue for it. Redhead ([1990]) explicitly acknowledges it when developing his own argument that talk of causes has no place in physics. He considers an explanation of the fall of a body using Galileo’s law, and asks what in the explanation might be regarded as the cause. As he notes, closely following Russell’s logic (Russell [1913], p. 183f) the position at an earlier time ‘can hardly be cited as the cause’, that the acceleration is merely ‘defined by the kinematic relationship’ expressed in the law, that the law itself cannot be regarded as a cause because it is not an event, and finally that citing the force as a cause is ‘a very anthropocentric notion’ (Redhead [1990], p. 146). Developing his charge of anthropocentrism, Redhead maintains that ‘[to] most physicists the old-fashioned idea of cause arises from the idea of our interfering in the natural course of events, pushing and pulling objects to make them move and so on. In modern physics there are just regularities of one sort or another’ ([1990], p. 147). Batterman ([2002], p. 127) makes similar remarks. Norton ([2003], pp. 3–4) endorses Russell’s thesis for Russell’s reasons, and says ‘mature sciences . . . are adequate to account for their realms without need of supplement by causal notions and principles. The latter belong to earlier efforts to understand our natural world or to simplified reformulations of our mature theories, intended to trade precision for intelligibility’. Glymour ([1999], p. 463) also denies that there are causal laws in physics. The thought that, somehow, both the process tradition and Russell of 1913 could be correct raises perplexing possibilities. Could process theories unify the special sciences but not unify them with physics? That would be surprising and confusing, not least because one of the best known and most carefully articulated criticisms of Salmon’s causal process theory is, roughly, that the opposite is the case, i.e. that it works better for physics than for the special sciences (Kitcher [1989], see also Ross and Spurrett [2004]). Were the thought that process theories worked for everything except physics instead taken as grounds for doubt about process theories, then perhaps the rebound view might be that Kim’s causal exclusion problem and the attractions of Russell’s thesis in physics mutually support one another: causal overdetermination is evidence for reductionism—in particular, for doubt that parochial causal patterns in special sciences are metaphysically significant—and then, once we go reductive, causation, absent in the reduction base, drops out altogether. Two overdetermining causes are not one cause too many, but two. We agree with Russell—and with most current philosophy of science—that scientists do not generally seek ‘invariable uniformities of sequence’. But it does not follow from this that they seek nothing worth calling causes. We argue that scientists do seek causes, in the everyday sense we describe, and that they find Russell’s Thesis Revisited 11 them. This might look like a straightforward confutation of Russell’s thesis. We go on to argue that it is not, since Russell’s argument is most interestingly read as attacking the idea that there is a ‘master’ idea of causation, independent of anthropocentric bias, that should be expected to feature in the metaphysical unification of the scientific worldview. This might be a basis for dissolution of Kim’s problem, but also seems to imply rejection of the point of process theories. However, we will skirt that conclusion too. Instead, we will defend the following consilience between Russell (and philosophers of physics who agree with him) and Salmon: the naturalist metaphysical project of seeking universal glue is well motivated; process theories are on the right track in this search; but ‘causation’ is a semantically unfortunate name for the glue to which they lead our attention. Part of our defence of the conclusion that it is semantically unfortunate that the cement of the universe gets called ‘causation’ involves identifying what we call conflicting centres of semantic gravity for ‘causation’.10 Roughly, there is a folk centre, a philosophical centre, and a centre that has a useful place in special sciences. It is the existence of this third centre that stops us from following Russell all the way to outright eliminativism about causation. 2 Psychology, Folk Notions and Intuitions To decide whether there are causes, or whether anyone seeks them, one needs to decide (i) what to count as a cause, and (ii) how to tell whether such things are real, or are sought. Russell partly settles (i) by reference to psychology. He also, as noted, regards psychological considerations as irrelevant to philosophy,11 maintaining that the content of some cause-talk is anthropomorphic projection, failing to correspond to what is discovered by science. Russell does not claim, though, that the ‘law of causality’ itself arises from psychological considerations—he blames philosophers for the law. Philosophers might reply that in taking ‘invariable succession’ as causation they were following a long tradition, claiming to find the law of causation through psychological observation. This is particularly true of the history of the account Russell criticizes, namely, the Humean view that causes are event regularities or ‘constant conjunctions’. Both Hume and his successors, including Kant ([1933], p. 218) and Mill ([1974], pp. 326–7), derive the philosophical significance of 10 In previous versions of this article we spoke sometimes of different concepts or notions of causation. We agree with those, including Crane ([1995]), an anonymous referee, and David Papineau (personal correspondence), who argued that this was sometimes confusing and unhelpful. 11 Russell ([1913], p. 174) objects to one of the definitions of ‘cause’ in Baldwin’s ([1901]) Dictionary that it is psychological, focusing on ‘thought or perception’ of a process where what is required is a definition of ‘the process itself’. 12 Don Ross and David Spurrett constant conjunctions from purported observation of their importance to both everyday and scientific cognition. Naturalists, trying to shun appeals to intuitions and conceptual analysis, are particularly inclined to look for evidence in scientific psychology. Some of what we know from the study of ‘causal cognition’ or ‘causal learning’ undermines the idea that our causal cognition is Humean. Contrary to early behaviourist dogma, associations are not all equally learnable. Rats find it much easier to learn (and show appropriate avoidance behaviour when cued for) an association between eating a certain food and nausea, and a loud flash and an electric shock, than they do with either complementary pairing (Garcia and Koelling [1966]). We might say that natural selection has projected a regularity in the world of rats, so that any given rat does not experience some types of phenomena as ‘entirely loose and separate’ (Hume [1975], p. 74). The projected regularities here, referred to by Garcia and Koelling as ‘genetically encoded hypothes[es]’ are not, furthermore, ‘invariable uniformities of sequence’. Learning in the nausea case can take place even when the delay between stimulus and response is considerable (over an hour) and avoidance or aversion can also be learned following a single exposure. The work of developmental psychologists already tells us a lot about the causal expectations of, for example, children around 3 months old, who are surprised when apparently cohesive objects seem spontaneously to fragment, to pass through one another, or not to exhibit a single trajectory through space and time (Spelke et al. [1995]), or when apparently unsupported objects fail to fall (Baillargeon et al. [1995]). Some headway is being made with the difficult question of what features of an object’s behaviour and/or structure lead children to regard it as capable of self-motion (Gelman et al. [1995]). Here too some events are not experienced as ‘entirely loose and separate.’ It is also clear that we do not spontaneously think in terms of constant conjunctions, but rather in terms of networks of influence similar to directed causal graphs, or Bayes nets. Young children (between the ages of 3 and 5) can, inter alia, reason to unobserved causes, plan appropriate novel interventions to prevent a process they had only observed in operation (but never seen prevented), and make inferences about the ‘direction’ of causal dependence in cases of simultaneous change (see, e.g., Gopnik et al. [2004]). The evidence is inconsistent with a simple conditioning or associationist view where ‘invariable uniformities of sequence’ are what is learned. The relevance of this to a naturalist should be clear: if ‘our’ concept of causation is an image of the kind of causal cognition that we actually engage in, then any feat of conceptual analysis starting from a model of ‘Humean’ constant conjunctions, or an appeal to alleged intuitions about such constant conjunctions, seems unlikely to be relevant to us, to science, or to metaphysics. To the extent that we have culturally universal intuitions about causation, Russell’s Thesis Revisited 13 this is a fact about our ethology and cognitive dispositions, rather than a fact about the general structure of the world. For naturalists, intuitions are not evidence for their own content, in either science or metaphysics. Empirical evidence in the specific case of intuitions about causation provides a particular exemplification of this general scepticism. Intuitive (in the sense of immediate, unforced) responses to causal questions are culturally varied. In particular it appears that adults in different cultures reason differently about causes. Chinese and American newspaper reports of very similar local multiple murders differed strikingly in the extent to which they referred to dispositional properties of the perpetrator (more common in America) and environmental ones (more common in China). Further, Chinese and American subjects in an experimental task involving counterfactual judgements about the murders had differing views about what sorts of changes (dispositional or situational) would have been likely to prevent the murder (Morris et al. [1995]). Nor are our causal intuitions a reliable basis for basic scientific inference. It seems reasonably clear that people are relatively poor at judgements about conditional dependence, as measured in the Wason selection task (see Cosmides [1989]); probability, where many will, for example, assert that the conjunction of two possibilities is more likely than one of the conjuncts alone (Tversky and Kahneman [1983]), and in other areas, compared to communities of scientists. Our response in such cases is of course not to take our scientific practice as thereby impugned. Just as the scientific study of vision enables us to explain some visual illusions, so the study of causal cognition might be expected to show us that we were prone to ‘causal illusions’ (Gopnik et al. [2004]). The proper task of philosophy and science in cases of causal illusion is to explain the answers people give to some causal questions, rather than to try to make the answers turn out to be correct. The upshot of this section so far is twofold. First, human causal cognition is not Humean. One temptation to take regularity as definitive of causation is thereby removed. Second, our intuitions about causation are neither detailed enough nor stable enough for the entire folk concept as exemplified by this or that culture to be used a basis for constructing the concept as it is appealed to by special sciences. To the limited extent that the folk concept of causation has a common core across cultural elaborations, its content is as follows. First, it construes causal relations as centred on some agent of change (animate or otherwise), thereby distinguishing agents and patients. Second, it postulates various transformative principles (often conceived as ‘forces’) proceeding out from an agent to the recipient of causal influence. Kim’s ([2005]) endorsement of a ‘generative’ account of causation may illustrate the attractions of this feature of the core folk centre of gravity, as we suspect does Armstrong’s ([2004]) contention that the philosophical theory of causation begins with our experience of ‘biff’ in acting in the world. Third, the core folk notion of 14 Don Ross and David Spurrett causation incorporates assumptions about time-asymmetry: causal influences flow from the past into the future. These three features seem empirically to exhaust the core folk concept.12 The artificial intelligence researcher Patrick Hayes ([1979]) spent years trying to work out deductive models of the everyday concept of causation as the basis for designing a simulation of a person. These models began with the three features just identified, then added knowledge people have of the mechanics of their own bodies to produce programs for distinctive manipulations of solid objects and liquids. Folk psychology would be conjoined with the core folk causal model to produce programs for causally influencing other agents. It is open to a naturalist to allow a folk notion to fix what is to count as a cause for purposes of inquiry into anthropically constructed or centred domains and practices. However, the naturalist cannot allow anything but science to say whether there are such things as causes. With a watchful eye open for illicit appeals to folk judgements, then, let us turn to consider science.