Causation with a Human Face
Normative Theory and Descriptive Psychology
JAMES WOODWARD
Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and certain other countries.
Published in the United States of America by Oxford University Press 198 Madison Avenue, New York, NY 10016, United States of America.
© Oxford University Press 2021
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by license, or under terms agreed with the appropriate reproduction rights organization. Inquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above.
You must not circulate this work in any other form and you must impose this same condition on any acquirer.
CIP data is on file at the Library of Congress ISBN 978–0–19–758541–2
DOI: 10.1093/oso/9780197585412.001.0001
Printed by Integrated Books International, United States of America
For my dear wife Julia and my brave daughter Katherine
1.
2.
3. Methods for Investigating Causal Cognition: Armchair Philosophy, X-Phi, and Empirical Psychology
4. Some Empirical Results Concerning Causal Learning and Representation
5. Invariance
6.
7.
8.
Acknowledgments
I have been interested in the empirical psychology of causal cognition for decades, but the topic really began to come into focus for me when I participated in a series of workshops at the Stanford Center for Advanced Study in the Behavioral Sciences (CASBS) in 2003–4. These were funded by the McDonnell Foundation, with Alison Gopnik as the principal investigator under the title “Causal Learning: Computational Learning Mechanisms and Cognitive Development.” The workshops brought together philosophers, psychologists, statisticians, and others with a common interest in using computational theories of causal learning to illuminate its empirical psychology. From my point of view, they were paradigms of successful interdisciplinary interaction. I am very grateful to the participants in these workshops for all that I have learned from these interactions. In this connection, I particularly want to mention John Campbell, Patricia Cheng, David Danks, Alison Gopnik, Clark Glymour, Chris Hitchcock, Tamara Kushnir, Andrew Meltzoff, Laura Schulz, Josh Tenenbaum, and Henry Wellman and to apologize to the many other participants whose names I have omitted.
In the intervening years I wrote several papers on the relationship between normative/computational theories of causal cognition and empirical psychology, but I did not begin this book until I was a fellow at the CASBS during 2016–17. My fellowship at the Center (along with support from the University of Pittsburgh) provided me with an uninterrupted nine months that allowed me to complete most of an initial draft of this book. I’m very grateful to the Center and to Margaret Levi, its director, for this support. While at the center I interacted very regularly with several social scientists with an interest in causal inference, including Paul Jargowsky, Ross Matsueda, Teppei Yamamoto, and David Yeager. I am very grateful to them for many helpful conversations.
A number of people have commented on portions of the present manuscript. Thomas Blanchard and Clark Glymour read through the entire manuscript and provided a number of very helpful suggestions. Others who have commented or who have provided me with advice and feedback on issues related to those I discuss include Colin Allen, Wesley Buckwalter, Patricia Cheng, Alison Gopnik, Chris Hitchcock, Josh Knobe, Edouard Machery, Michael Rescorla, Michael Waldmann, and Zina Ward. I taught portions of an early version of this manuscript in a graduate seminar at the University of Pittsburgh in 2018 and have benefited from the input of students in that class. I have also greatly benefited
from the comments of an anonymous referee for Oxford University Press. (There was also, inevitably, referee number 2, who did not comment at all on the substance of my manuscript but objected very strongly to its “anti-metaphysical” tone. I’ve tried in revising to be gentler in my treatment of metaphysics.)
The Department of the History and Philosophy of Science at the University of Pittsburgh has provided a very supportive intellectual environment for my work on this and related projects. I am grateful to my colleagues there, as well as Bob Batterman and Mark Wilson in Philosophy for their friendship and encouragement. Having colleagues at Carnegie-Mellon with a particular interest in causal inference and in some cases empirical psychology has also been a tremendous benefit—here I particularly want to mention David Danks, Clark Glymour, Peter Spirtes, Richard Scheines, and Kun Zhang. Finally, although they have not commented directly on this manuscript, I want to mention Frederick Eberhardt and Jiji Zhang, who have certainly influenced many aspects of my views about causation and causal reasoning.
Introduction
I.1. Preamble
The title (as well as the cover) of this book reflects my conviction that in understanding causal cognition (and causation itself) it is crucial to appreciate that causal reasoning is an activity carried out by human beings. More specifically, the ways we think about causation are in part shaped by human interests and goals and human abilities, including various epistemic limitations we face. Understood as a claim about causal cognition or the causal notions we employ, it is hard to see how this could be disputed. However, it is somehow easy to lose sight of the fact that when we talk about causal relations “as they are in the world” we are still employing categories and ways of thinking that are human products, reflecting human concerns.
This conviction—that “causation” and “causal cognition” need to be understood together and with reference to the sorts of creatures that we humans (and perhaps other sorts of subjects)1 are—provides one of the primary motivations for the issues and topics taken up in this book. These issues are various—perhaps they will seem to some readers to be too much so to be appropriately discussed under one cover. I discuss theoretical or normative treatments of causation and causal reasoning (normative in the sense that these are theories about how we ought to reason causally).2 Some of these derive from philosophy, others from disciplines like machine learning and statistics. I discuss how these normative theories interact with descriptive research on the empirical psychology of causal cognition in humans and (to some small extent) in other animals. I try to illustrate how these two enterprises—the theoretical/normative and the empirical—can mutually and beneficially inform one another: normative ideas can guide and structure empirical research, and the results of empirical research can suggest possible normative ideas. I also attempt to connect all of
1 In subsequent chapters I briefly discuss learning and behavior with cause-like features in nonhuman animals as well as causal learning in very young children. So when I refer later to human limitations and goals, the reader may think of this as shorthand for limitations and goals for creatures who are broadly like us in not being omnipotent or omniscient.
2 The idea that accounts of causal reasoning and causation can be understood as “normative” may strike some philosophical readers as odd, but this (and a corresponding contrast with “descriptive” accounts) is standard outside of philosophy. I defend it in what follows. In my view, it better captures what is really at stake when we theorize about causation than alternative approaches.
Causation with a Human Face. James Woodward, Oxford University Press. © Oxford University Press 2021. DOI: 10.1093/oso/9780197585412.003.0001
this with methodological debates in philosophy concerning the role of appeals to “intuitions” or “judgments about cases,” focusing on examples involving judgments about causal relationships. Such appeals figure prominently in many philosophical discussions of causation, but an additional reason for making this connection is that claims that people hold various intuitions or make various judgments are empirical claims, and in this respect arguably not so different in principle from empirical claims about causal cognition resulting from psychological experiments. I advance some suggestions about how such judgments about cases as well as empirical results from psychological experiments can, when appropriate additional background assumptions hold, be relevant to normative theorizing about causation.
My justification for discussing all of these issues together is that however ungainly the result, they belong together but are often not discussed in a connected way.3 Thus one goal is simply to get readers to see that such connections exist and are worth exploring—hopefully this claim will be persuasive even to readers who do not accept many of the details of what I have to say. What I would like to do is to open up discussions of causation and causal reasoning, particularly but not exclusively in philosophy, helping participants in these discussion to appreciate the relevance of considerations drawn from many different sources, both descriptive and normative, that have not always received much philosophical attention. I thus oppose various strategies of dismissal (as I call them) that are sometimes taken to show that many of the issues that I discuss together have little or nothing to do with each other. These include the claims that normative issues about causal reasoning have no bearing on empirical studies of causal cognition and conversely and that the “metaphysics” of causation (understood as having to do with causation as it is in the world) has no connection to the epistemology or methodology of causal inference. I discuss some of these strategies in a general way in this introduction and in more detail in subsequent chapters.
I am of course aware that a number of philosophers will think that empirical facts about how humans and others think causally, as well as ideas from disciplines like statistics and machine learning, could not possibly be relevant to philosophical projects connected to causation. What I hope to show is that the results will be richer and more interesting if one does not proceed on such assumptions of irrelevance.
3 So, to readers who think this is far too much to discuss in one book, I sympathize but also think not discussing these issues together would be a missed opportunity.
I.2. Interventionism
The framework for thinking about causation adopted in this book is broadly interventionist in the sense described in Woodward 2003. As a very rough characterization (there is more detail in Chapter 2), interventionists think of causal claims as claims about the outcomes of hypothetical experiments. Different sorts of causal claims may be distinguished on the basis of the kinds of experiments with which they are associated but a generic version is that C causes E if and only if there is some possible intervention on C (and depending on which causal notion is being characterized, perhaps on other variables) that, if performed, would be associated with a change in E. An intervention may be thought of as an idealized experimental manipulation. However, many of the conclusions I reach—about the importance of normative theorizing and its relation to descriptive claims, about the legitimate role of appeals to intuitions, as well as the descriptions of particular experimental results—seem to me to be relatively independent of a commitment to this framework. I hope that they will interest readers who are not persuaded by interventionism.
I.3. Some Misused Dichotomies (and Other Assumptions)
It is not a very deep or original observation that the reception of philosophical ideas often depends upon presuppositions that are regarded by those who hold them as so obvious and uncontroversial as to require little examination and defense. This is a problem for those, like me, who do not share a number of these assumptions: views of the sort I will defend will not seem prima facie plausible or even intelligible unless readers are prepared to consider the possibility that some of their presuppositions may be mistaken.
In what follows I attempt to identify several of these presuppositions and to briefly explain why I do not find them persuasive. I do not expect to change minds, but I hope that this exercise will help some readers to better understand what I am trying to say. I see these presuppositions as connected to entrenched distinctions or dichotomies. I do not object to the distinctions themselves, which I take to be uncontroversial, but rather to some of the uses to which the distinctions are put and to certain mistaken inferences they are used to motivate. Roughly speaking, the mistaken inferences involve moving from the correct observation that X is different from Y to the conclusion that X and Y are completely unrelated and can (and should) be discussed entirely independently of each other, so that any attempt to relate them involves “conflating” X and Y. The distinctions thus function rhetorically as devices for dismissal of the sort mentioned earlier.
I.3.1. Normative and Descriptive
I take it to be uncontroversial that the descriptive claim that so-and-so is the case does not (aside perhaps from some trivial exceptions) by itself entail the normative claim that so-and-so ought to be the case. Similarly the normative claim does not entail the corresponding descriptive claim. From the fact that people make certain causal inferences or judgments and not others, it does not (without further assumptions) follow that they are correct or normatively justified in doing so. From the claim that people ought to make certain judgments, it does not follow that, as a descriptive matter, they do make such judgments.
However, although there is no interesting entailment from “is” to “ought” or conversely, it does not follow, at least in the case of causal cognition, that we should think of the descriptive and normative as completely disconnected areas of inquiry: one obvious possibility (but not the only one) is that there are additional true assumptions that might be used to establish connections between these two areas. As an illustration, suppose that there is reason to believe that some causal inferences and judgments we make are the result of selection processes (involving either natural selection or learning or a combination of these) that operate so as to preserve true judgments while discarding false or mistaken ones, so that people are, in this respect, successful in many of the causal judgments and inferences they make.4 Then the descriptive fact that people make such judgments and inferences may, in conjunction with the assumption about truth-conducive selective processes just described, be taken to have some bearing on whether those judgments or inferences are correct or normatively justified. (Again, I say more about standards of normative justification in subsequent chapters.) This need not mean that we should conclude just on the basis of the descriptive facts and the assumption about selection that the judgments
4 I say more about success in subsequent chapters. But for those who are worried that invocations of success presuppose the correctness of some particular account of causation, I offer the following: First, we do have some uncontroversial cases of causal knowledge—one mark of success in an account of causation or a causal inference procedure is that it reproduces these, that it is well calibrated in this sense. Second, and relatedly, different accounts of causation often agree on what causal relations are present in particular cases, even if they disagree about the bases for such judgments— again, these cases can be used as standards for success or correctness that don’t presuppose the correctness of any particular theory. Finally, accounts of causal inference and of other features of causal reasoning can sometimes (depending on the account in question) be assessed “internally”—in terms of whether they accomplish the goals they specify. For example, if one thinks (as interventionists do) that causal discovery has to do with identifying relationships that support manipulation, one can ask whether some particular inference procedure reliably accomplishes that. Of course, even if this is the case, a critic might deny that such manipulation-supporting relationships are causal relationships, but the requirement just described can function as an internal check. A useful analogy, developed in more detail subsequently, is the claim that the human visual system enables us to interact “successfully” with our environment. One can have good grounds for thinking that this is correct without presupposing any particular theory of how the visual system works. Similarly for human causal cognition.
and inferences are normatively justified. A weaker possible conclusion is that the descriptive considerations in conjunction with assumptions about selective processes are only suggestive: they suggest the judgments and inferences in question may have a normative justification—that this is a possibility worth investigating. Showing that there is such a normative justification then requires an independent (normative) argument of some kind. It is this weaker conclusion that I will defend in subsequent chapters, but my point here is that this illustrates one possible way in which the descriptive and normative might be connected. Of course one might think that in some particular case or cases, this contention about selective processes operating on judgments and inferences in such a way as to preserve correctness is mistaken, but whether this is so is (I will argue) a broadly empirical matter. If this contention is correct in some cases, one can’t argue that, as a matter of principle, descriptive claims about how people judge can have no bearing on how they ought to judge.
A similar point holds in the other direction. One general assumption that might be used to connect claims about normatively appropriate causal reasoning to empirical claims about how people in fact reason is the assumption that many people are “rational,” in the sense that they make normatively appropriate causal inferences and judgments much of the time. To the extent that this assumption is correct, it will be reasonable to begin with normative models of causal reasoning and then investigate the extent to which, as an empirical matter, people conform to these models. In fact, as we shall see, a methodology of this kind is widely employed in psychological investigations of causal cognition, often with fruitful results. Again, I will not try to defend this methodology here (that is done in Chapter 3); my point is rather that the assumption of rationality is one obvious way in which the results of normative theorizing might be brought to bear on or shown to be relevant to descriptive investigation.
In fact, there are a number of different ways (in addition to what I have just described) in which descriptive and normative claims might be relevant to or brought to bear on one another. In my experience, philosophers tend to frame questions about the relevance of the descriptive to the normative as questions about whether descriptive claims can be “evidence” for normative claims (or vice versa) or whether there are nontrivial entailments between the two. But there are many other possibilities. A normative claim might provide motivation for engaging in a descriptive inquiry (to see whether people conform to the normative claim). The empirical fact that some feature is present in people’s cognition might motivate an investigation into whether there is some normative rationale for the feature. (Plenty of examples of both possibilities will be discussed subsequently.) In neither case do we need to think of the relation between the normative and descriptive as a matter of one providing evidence for the other or as one
entailing the other. Trying to force the descriptive/normative connections into such frameworks will make them seem more mysterious than in fact they are.
I.3.2. Metaphysics and Epistemology
I.3.2.1.
Metaphysics
As a non-practitioner, I approach this subject with considerable trepidation. At a very general level I take it that “metaphysics” has to do, at least in part, with what the world is like, with what is “out there” in nature. In the case of causation, this will have to do with what true causal claims are answerable to in the world, and with the structures in nature that support the application of causal reasoning (what the world has to be like, very broadly speaking, if we are able to engage in successful causal reasoning about it).5 Different conceptions of causation will give different answers to such questions. For example, if you are an interventionist like me, taking true causal claims to describe what would happen if certain interventions were to be performed, then (at a minimum) you will think that there are facts about what would happen if those interventions were to be performed—facts about what would happens if rocks with certain properties were to strike windows, facts about what happens to headaches when aspirin is ingested, and so on. In this sense there are intervention-supporting relationships “in” the world.6 (I will say a bit more about this momentarily.)
If issues of this “what is out there” sort are all that is meant by metaphysics, I am fully on board regarding its importance. Indeed, I think it is impossible to talk coherently about successful causal learning and judgment without some conception of what causal relationships in the world involve, since “success” implies discovering such relationships However, as a naturalist, I take the relevant features and relations that are “out there” to be just those countenanced in science and in other kinds of empirical inquiry—it is ordinary empirical inquiry that tells us which are the intervention-supporting relationships. Moreover, if we ask why these particular intervention-supporting relationships obtain, the explanations will be ordinary scientific explanations appealing to, for example, the chemical structure of aspirin and the way in which this interferes with the
5 For more on this theme of structures in the world that support causal reasoning, see Woodward, forthcoming c and Weinberger et al. forthcoming.
6 In saying this, all that I mean is that claims about such intervention-supporting relationships are truth-apt and that whether they are true or false has to do with how matters stand in the world. In my view, saying this does not require an elaborate “ontology” of such relationships. To employ an analogy that I have used elsewhere: it is a factual matter whether a given roulette wheel has such-andsuch probabilities of producing red or black outcomes when spun by a macroscopic agent. Moreover, there are physical explanations for why roulette wheels behave in this way in terms of the method of arbitrary functions and similar results. Acknowledging this does not require that one subscribe to any particular “ontology” of chance.
enzymes that synthesize prostaglandins. We don’t require in addition some variety of “metaphysical” explanation. This view amounts to what I will call “minimalist” metaphysics or “minimal realism.”7
Although the contrast may not be a sharp or clear one, I would distinguish the minimalism just described from more ambitious or expansive forms of metaphysical theorizing, which (it seems to me) go beyond what science tells us about nature or involve forms of explanation that are something other than scientific.8 These have taken a variety of different forms in recent discussion concerning causation and related topics. Some approaches postulate special entities or relations—powers, relations of necessitation between universals, and so on— to serve as “truth-makers” for causal claims. In other cases criteria for the acceptability of some proposed metaphysical theory are imposed that (in my opinion) have no obvious scientific motivation, as in the demand that the theory be reductive and/or that it only make use of primitive notions (“perfectly natural properties,” etc.) that are metaphysically acceptable.
I said earlier that it is an important fact that we have procedures for learning about and reasoning with causal relations that are sometimes successful and that to characterize such success we need some conception of what is out there in nature that such procedures discover. However, my view is that all the resources that we need to characterize success and to explain how it is possible will be of minimalist sort previously described. This is one reason why I do not think that the topics discussed in this book require more ambitious metaphysics. Some will hold that even if there are facts about what would happen if rocks strike windows or aspirin is ingested, these cannot be part of the “fundamental ontology” of the world. I won’t dispute this since I don’t mean to be making claims about fundamental ontology. What matters for my purposes is that whatever may be the constituents of nature at more fundamental levels (and however “fundamental” is understood), these are somehow arranged in such a way that certain claims about what would happen if certain intervention were to occur turn out to be true and others turn out to be false. Suppose, for example, that fundamentally all that exists is a Humean mosaic of perfectly natural properties at spatiotemporal points of the sort characterized by Lewis (1986). My version of minimal realism assumes that under this supposition it will still turn out to be the case that there are also what-would- happen-if facts of the sort previously mentioned.9 It is
7 If this strikes you as just science, rather than metaphysics, I’m fully on board. I use the phrase “minimal metaphysics” to accommodate the many philosophers who think that any claims about what exists amount to metaphysics.
8 I don’t have a hard-and-fast criterion that distinguishes minimal from more ambitious metaphysics. Perhaps one diagnostic for the latter is whether some distinctively metaphysical (as opposed to “scientific”) form of explanation is invoked, as in Loewer 2012.
9 If this sort of minimal realism seems weak and unambitious, I have intentionally designed it to be so. It does rule out some views that have been entertained by prominent philosophers—for example, the claim that there are no worldly facts about what would happen if such and such were
these that we represent and reason about when we engage in causal inference and judgment. My picture is thus one in which we can legitimately talk about causal claims and associated what-would-happen-if counterfactuals at many different “levels,” including “upper levels” involving social, psychological, and biological variables and relations, as well as facts disclosed by ordinary manipulative activities like throwing a rock at a window. Some of these intervention-supporting relationships will be more local (less invariant—see Chapter 5) than others but all are part of reality.
I.3.2.2. Epistemology
Metaphysics, whether taken in a minimalist or more expansive sense, is obviously a different enterprise from epistemology. I take the latter (and here we can also include methodology) to have to do with what we can know and how we know it and, more generally, with strategies for learning about various aspects of nature. Obviously what the world is like is a different matter from what we know or even can know about it. However, it does not follow from this truism that investigations of the epistemology and metaphysics (including, especially, minimal metaphysics) of causation should be pursued completely independently of each other. Instead, as I see it, these enterprises should mutually constrain one another. Thus we should resist arguments along the lines of “I’m interested in the metaphysics of causation. Your arguments, even if correct, have to do only with the epistemology of causation (or methodologies of causal inference), which is a completely different subject.”
Why should we resist such arguments? To begin with, as emphasized previously, from a normative perspective, one wants claims about how we ought to learn about and judge causal relationships to fit with or track which such relationships hold in nature—a matter of metaphysics, even on a minimalist conception of that subject. Different conceptions of the conditions a relationship should meet to qualify as causal (and different views about whether there are any relations in nature that satisfy these conditions) will fit with different conceptions of the evidence and inferential strategies that are appropriate for discovering such relationships. We thus have the following desideratum on an account of causation: there is a strong prima facie case that it should be understandable, given our conception of what causal relationships involve, how the kinds of evidence and strategies we employ to discover such relationships sometimes actually successfully lead to their discovery. Again consider interventionism about causation: true causal claims describe what the results of hypothetical experiments would be were we to perform them. If this view is correct
the case, with modal claims of this sort rather being “grounded” in, or reflections of, our projective activities.
we should be able to see how normatively appropriate methods for discovering causal relationships provide reliable information about the outcomes of such hypothetical experiments, thus showing that the methods are normatively appropriate because they provide such information. Of course well-designed actual experiments will trivially meet this condition, but so will various inference strategies involving non-experimental data. For example, these may include the use of instrumental variables and regression discontinuity designs10 these techniques work because they provide information about what would happen if certain experiments were to be performed, without performing these experiments.11 It is less clear why such techniques are particularly appropriate for discovering causal relationships on certain other conceptions of causation (or so I claim).
As a concrete illustration of how one would like the epistemological and the (minimal) metaphysical to be connected, consider a randomized experiment assessing the role of a certain pedagogical technique P in improving educational performance E among a group of students. If the experiment is properly performed, it is generally thought that it can provide evidence that P causes E, among these students, where this means something like the average causal effect (see Chapter 5) of P on E is positive. Now consider the common metaphysical suggestion that true causal claims are just those claims that relate events that “instantiate” laws of nature. Putting aside any other misgivings one may have about this suggestion (including, for example, the unclarity of the notion of “instantiation”), one question we should ask is how this metaphysical claim relates to the evidence provided in the experiment. Why are we entitled to think of the results of the experiment as evidence for the instantiation of a law of nature? After all, “P causes E” is not itself a law of nature on any reasonable conception of what a law is, and there is unlikely to be any other explicit invocation of laws in the experiment. Moreover, why does it matter that the experiment involves randomization, given that “randomization” is not a notion that plays any role in our notion of a law of nature? Of course if we are willing to assume that true causal claims must instantiate laws, then we can presumably conclude, when the experiment is properly carried out, that what is discovered is the fact that P and E instantiate some (unknown) law, but this gives us no insight into why it is normatively appropriate to interpret the experiment as discovering this.12 Such a connection between the method employed in the experiment and the interpretation of the causal claim it establishes is one thing that we are looking for when we ask for a connection between the epistemology/methodology and metaphysics of causation.
10 See, for example, Angrist and Pischke 2009.
11 For more on this see Woodward 2015a.
12 In other words, it is consistent to assume that (i) randomized experiments are good ways of discovering causal relationships and that (ii) what is discovered is instantiation of a law. But what one would like is a closer connection between (i) and (ii) than mere consistency.
To enlarge on this theme, consider the many strategies and procedures, experimental and non-experimental, that are commonly taken to be appropriate for learning about casual relations and judging which causal claims are true. Now consider the possibility that these fail to fit at all with the correct metaphysical account (either minimalist or expansive) of causation in the following sense: given the metaphysical account, the procedures and criteria we now employ are all completely inappropriate for discovering causal relationships. Going further, consider the possibility that there are no procedures or criteria for recognizing true causal claims (as described by some correct metaphysical theory) that human beings can actually carry out or implement.
Whatever we may think about these as logical possibilities, the idea that they might be the correct account of the relationship between the epistemology and metaphysics of causation ought to strike us as deeply puzzling, if one holds, as I do, that the notion of causation is a functionally useful one and human causal reasoning is to some extent “rational.” For starters, why would we have developed conceptions of what it is for a causal claim to be true and conceptions of how to tell whether such claims are true that are so completely out of sync in this way? If we have done this, why wouldn’t we notice this problem and take steps to correct it, either by adjusting our understanding of causation or its epistemology/methodology or both? Given such a failure of synchronization, both our methodology and our conception of what it is for causal claims to be true would be completely useless—the former because unreliable, the latter because, given this unreliability, we have no way of telling which causal claims are true.
I don’t mean to suggest that this sort of systematic mismatch can never happen in any area of investigation. I take it to be very plausible, for example, that given a literal reading of what is required for various theological claims to be true, the procedures and arguments for determining which such claims are true that believers standardly employ are completely inadequate for this purpose. But I also take it to be a plausible starting point that causal claims and causal reasoning are not like theology in this respect. When it comes to causation, my working assumption will be that how we find out about causal relations can tell us something about what those relationships involve. Conversely, a satisfactory account of what causal relations involve can help us to understand how our procedures for finding out about them succeed to the extent that they do.
Let me add, by way of elaboration, that the arguments of the preceding paragraphs depend on (or at least will make most sense in the light of) a certain conception of what a defensible epistemology/methodology of causal reasoning involves. There is a tendency among philosophers to be rather abstract and nonspecific when it comes to epistemology/methodology: it may be thought
sufficient, for example, to say that causal claims are established by “our usual inductive practices” or “inference to the best explanation,” or by “making the appropriate adjustments in our web of belief,” without further specifying what these involve. If one operates at this level of generality, it may be hard to see how epistemological/methodological considerations could impose any tight constraints on an account of causation, since the epistemology seems to amount to little more than “We somehow figure out which claims are true.” However, as subsequent chapters and discussions elsewhere illustrate, actual strategies for reliable causal inference are far more specific and discriminating. It is these strategies that can provide useful restrictions on accounts of causation.
I.4. More on Epistemic Limitations and the Metaphysics and Epistemology of Causation
I have been defending the idea that the ways in which we think about causation are shaped by what we can know or find out, among other factors.13 This is likely to prompt the reaction that even if this is true of how we think about causation, causation as it exists in the world is a very different matter. I have already agreed that there is an obvious sense in which this distinction between how we think and what the world is like is uncontroversial. However, we need to be careful about the conclusions we may be tempted to draw from this distinction. When we talk about causation as it exists in the world, we of course are still making use of conceptions and ways of thinking that we humans have developed. There is very strong empirical evidence that these conceptions are influenced or shaped by features of our psychology and situation in the world— including our epistemic limitations, goals, and interests. (This is true, I claim, of both ordinary lay thinking about causation and causal notions deployed in the various sciences.) Denying these empirical facts is not a viable strategy. An alternative possibility is to concede the empirical facts but to claim, as a normative matter, that in talking about causation as it exists in the world (or what causation “is”) we should be aiming at an account of causation that completely abstracts away from such human influences—an account of causal relations as God would describe them or some such. But I see no reason to suppose that we have access to such a godlike conception of causation. Nor, to the extent that we attempt to
13 What I understand by “shaped by” will become clearer subsequently, but very briefly I mean “causally influenced by in something like the way in the way that the holding of a set of ideas can be influenced by one’s material circumstances, interests, and goals.” I do not mean that built into the semantics of causal claims are references to epistemic limitations, goals, and the like, that “causation itself” is in some way interest- or goal-relative or anything similar. Various factors can influence how we think about causation without being part of the semantics of causal claims.
entirely abstract away from our goals and limitations, is it clear what the use (for us) of the resulting account would be—why, normatively speaking, it would be a good thing to have. My contrary assumption is that in talking about causation, we cannot avoid employing categories and modes of thinking that reflect characteristically human concerns.14 Part of understanding causation as it exists in the world involves understanding these categories and modes of thought and what work they do for us. Moreover, a normatively useful notion of causation must connect with these human concerns.
I emphasize again that this does not imply some kind of subjectivism or projectivism about causation. Acknowledging that how we think about causal relations is shaped by human concerns does not mean that there is nothing in the world outside of us that is tracked by such relations or that these relations are “mind dependent” in some problematic way.15 Given a specified way of thinking about causal relations such as interventionism, it is an objective matter whether the world contains any such relations and, if so, which ones. However, it is also a mistake to think that acknowledgment of the objectivity and mind independence of causal relationships requires that we adopt the causation-as-Godwould-understand-it idea.
I.5. The Role of Goals and Interests
Alongside the tendency to look for accounts of causation (and other notions of scientific interest) that abstract entirely from our epistemic and calculational limitations (as well as our procedures for finding out about these), there is a parallel tendency to think that a suitably objective account of such notions should abstract away from any considerations having to do with goals and interests or what we want to do with these notions.16 I view this tendency as misguided. As
14 Some may think that fundamental physics gives us just such a God’s view conception. But physical theorizing is just as much a human activity as any other. The idea that physics describes the universe from God’s perspective is theology, not science.
15 I acknowledge glossing over some complex issues here, not the least of which is what is meant by “mind dependent.” It is clear that the acceptability of some kinds of causal judgments is more influenced than that of others by, for example, social norms. This appears to be true, for example, for “actual cause” judgments involving norm violations—see, for example, Hitchcock and Knobe 2009. In other cases, causal judgments are sensitive to which possibilities are regarded as serious, as opposed to far-fetched, although such assessments in turn are heavily influenced by “objective” facts—see Chapter 6. But even in these cases, there is a more objective core of dependency relations that underlies the judgments in question. Moreover, in many other paradigmatic cases, factors such as norms play little or no role.
16 This is reflected in talk of “perfectly natural properties,” “cutting nature at the joints,” and the like, where at least in philosophy the assumption seems to be that these can be characterized without any reference to the purposes for which one may want to use such descriptions and distinctions—the descriptions and distinctions instead being supplied by nature alone. (If nature could talk, this is how she would describe herself.)
an interventionist, I see causal cognition as shaped by human goals and interests having to do with manipulation, control, and prediction. Again, I claim that this does not by itself introduce an objectionable kind of subjectivity. Connecting causal relationships to relationships that are exploitable for manipulation and control does not imply that whether or not it is true that manipulating C is a way of changing E is somehow produced or created by our goals or interests. My picture is this: our goals lead us to care about certain sorts of relationships— those that support manipulation—but which relationships in nature have these features is determined by what nature is like. The idea that acknowledging a role for goals and features of our epistemic situation in shaping how we think about causation undermines the “objectivity” of that notion seems to me to reflect a confusion about what a reasonable sort of objectivity requires. Again, as I will understand it, “objectivity” for causal relationships holds to the extent that causal claims are the sorts of things that are true or false, and which they are turns on how matters stand in nature and not on such factors as our projective activities. Relatedly, an important mark of the objectivity of causal claims is that there are procedures involving objective information about the world (statistics, results from experiments, etc.) that can sometimes be used to determine whether they are true or false.17
I.6. Philosophical Strategy and an Apology
Next a remark about philosophical strategy (or at least the strategy followed in this book) and standards of assessment. The standard I will adopt is fruitfulness. This standard evaluates in terms of criteria like the following: whether the approach and positions advocated lead to new discoveries or insights, either conceptual or empirical; solve previously unsolved problems; provide connections to work in other disciplines; generate new research problems; and provide resources for dealing with them. When one evaluates in terms of this standard, one assumes that virtually any philosophical program or set of positions will encounter problems and difficulties at some points, but the idea is that rather than focusing entirely on these, evaluation should focus more on the program’s positive achievements and successes, at least as long as these continue. We should thus resist the temptation to think that the successes are not genuine as long as there are also unsolved problems, counterintuitive
17 This is another illustration of the way in which the metaphysics and epistemology of causation are intertwined. Strongly subjective or projectivist views about the “nature” of causation are hard to reconcile with the apparent objectivity of the procedures we have for determining whether causal relationships hold.
implications, and so on. I take this to be the standard often employed in the evaluation of scientific research programs, and I think it is an appropriate one to employ in philosophy as well.
Finally, an apology. I’m aware that there is a fair amount of repetition in what follows. To some extent this is deliberate: I’ve re-iterated certain points in an effort to avoid being misunderstood, at the cost perhaps of trying the reader’s patience.
