Causation in Psychology

John Campbell

• Publisher: Harvard University Press
• Year: 2020
• ISBN:
  9780674249523
• Topic: Philosophy
• Subtopic: Mind & Body in Philosophy

A renowned philosopher argues that singular causation in the mind is not grounded in general patterns of causation, a claim on behalf of human distinctiveness, which has implications for the future of social robots. A blab droid is a robot with a body shaped like a pizza box, a pair of treads, and a smiley face. Guided by an onboard video camera, it roams hotel lobbies and conference centers, asking questions in the voice of a seven-year-old. "Can you help me?" "What is the worst thing you've ever done?" "Who in the world do you love most?" People pour their hearts out in response.This droid prompts the question of what we can hope from social robots. Might they provide humanlike friendship? Philosopher John Campbell doesn't think so. He argues that, while a social robot can remember the details of a person's history better than some spouses can, it cannot empathize with the human mind, because it lacks the faculty for thinking in terms of singular causation. Causation in Psychology makes the case that singular causation is essential and unique to the human species. From the point of view of practical action, knowledge of what generally causes what is often all one needs. But humans are capable of more. We have a capacity to imagine singular causation. Unlike robots and nonhuman animals, we don't have to rely on axioms about pain to know how ongoing suffering is affecting someone's ability to make decisions, for example, and this knowledge is not a derivative of general rules. The capacity to imagine singular causation, Campbell contends, is a core element of human freedom and of the ability to empathize with human thoughts and feelings.

CHAPTER ONE

THE SPACE OF REASONS AND THE SPACE OF CAUSES

1.REASONS VS. CAUSES

Many people have thought there are differences between mental causation and physical causation. The differences have seemed so weighty that many philosophers have thought that mentalistic explanation ought not to be called “causal” at all. One way in the idea is put is by contrasting the “space of reasons,” where we find properly psychological explanations of behavior, with the “space of causes” (Sellars 1956, §36; Rorty 1979, 157).

In characterizing an episode or a state as that of knowing, we are not giving an empirical description of that episode or state; we are placing it in the logical space of reasons, of justifying and being able to justify what one says. (Sellars 1956, §36)

But on the face of it, there is no immediate tension between the existence of normative relations between psychological states and actions and the idea that psychological states can be the causes of actions. That is, we can recognize that there’s a difference between the “space” of normative relations among psychological states and the “space” of causal relations among psychological states and actions without accepting that psychology provides only explanations in the space of reasons.

In fact, recognizing a role for normative considerations in psychological explanation seems to require that we think of psychological explanations as causal. After all, it is not as if the normative relations between the psychological states are thought to hold outside the ken of the subject. The natural thought is that it’s because the subject recognizes the force of the normative relations that the psychological states propel the subject to action. The beliefs show the action to be a good idea, and it’s the subject’s recognition of this that causes there to be a causal relation between the states and the action. If the subject hadn’t recognized that the beliefs made the action a good thing, the beliefs wouldn’t have propelled the subject to action. This way of thinking of things seems to require that the subject was caused to act by the reasons. The normative relations provide the scaffolding within which these causal relations hold.

There’s a particularly lucid version of the idea that psychological explanation is not causal explanation in Dennett (1981). On Dennett’s picture, the relations between psychological states to which we appeal in characterizing someone’s action are indeed normative relations. He talks about “the intentional stance” in which one uses the rational, normative relations between psychological states in characterizing the action (1981, 61). To talk in psychological terms—to adopt “the intentional stance”—is to suppose that the subject is rational. These normative relations, however, are not exploited to find the causes of the action. Rather, the use of psychological terms, the appeal to the normative relations found in the intentional stance, is not an attempt at causal explanation at all. Therefore we have here a very strong distinction between “the space of reasons,” in which we find psychological explanations, and “the space of causes.” What causes the action is stuff in your brain; the normative relations among propositions matter only for the exercise of predicting behavior, not for explaining it causally. The position is made vivid by an example from Somerset Maugham.

There was a merchant in Baghdad who sent his servant to market to buy provisions and in a little while the servant came back, white and trembling, and said, Master, just now when I was in the market-place, I was jostled by a woman in the crowd and when I turned I saw it was Death that jostled me. She looked at me and made a threatening gesture; now, lend me your horse, and I will ride away from this city and avoid my fate. I will go to Samarra, and there Death will not find me. The merchant lent him the horse and the servant mounted it, and he dug his spurs in its flanks and as fast as the horse could gallop he went. Then the merchant went down to the marketplace and he saw me standing in the crowd and he came to me and said, Why did you make a threatening gesture to my servant when you saw him this morning? That was not a threatening gesture, I said, it was only a start of surprise. I was astonished to see him in Baghdad, for I had an appointment with him tonight in Samarra. (Maugham 1933, Act 3, 112)

The key point here is that Death does not know the causes of things. Death does not know that it was the jostling that caused the servant to go to Samarra. Death has an appointment book (we may surmise) and can predict where people will be at particular moments. But the appointment book does not of itself offer any insight into why any one person will be at any place at any time. In Dennett’s picture, being well versed in common-sense psychology is like having Death’s appointment book. You have a way of generating predictions about who will be where when, but you do not as yet have any insight into why they behave as they do.

Dennett summed up his view like this. The “intentional stance”—what we use when we’re talking about people’s minds—requires an assumption of rationality on the part of the target. We assume they’re kind of sensible. For example, “sheltered people tend to be ignorant; if you expose someone to something he comes to know all about it . . . our threshold for accepting abnormal ignorance in the face of exposure is quite high. ‘I didn’t know the gun was loaded,’ said by one who was observed to be present, sighted, and awake during the loading, meets with a variety of utter skepticism that only the most outlandish supporting tale could overwhelm” (Dennett 1981, 61–62). The “intentional stance” is like Death’s appointment book. It allows you to predict what someone is going to do without giving any insight into causes. Being predictable in this way is all it takes to have a mind, because the only use for our mentalistic talk is in giving these kinds of prediction:

Any object—or as I shall say, any system—whose behavior is well predicted by this strategy is in the fullest sense of the word a believer. What it is to be a true believer is to be an intentional system, a system whose behavior is reliably and voluminously predictable via the intentional strategy. (Dennett 1981, 59)

This is a version of the idea that the “space of reasons” in which we find psychological explanation is quite different to the “space of causes.” When we are operating in terms of the space of reasons, we are merely using a predictive abacus. Finding causes would require us not to be talking in mentalistic terms at all but rather to be looking at the biology of the brain.

Dennett does not give much weight to the idea that the normative connections that characterize the space of reasons are known about in a way that is quite different to the way in which we know about the causes of things. But for many philosophers, the trouble with supposing that mentalistic connections are causal connections comes when we reflect on the normative relations that characterize the “space of reasons.” The idea is that these normative relations must themselves be a priori, or, as philosophers used to say, in some broad sense, “logical.” But when a priori or logical connections hold between two states, it’s said, they can’t be “distinct existences” in Hume’s sense (Hume 1748 / 1975, IV / 1). But cause and effect have to be distinct existences. Therefore, the relations between these normatively linked states can’t be causal. Moreover, traditionally there is no predictive point to the exercise. It is usually thought to be just one of the fundamental modes of explanation, when one shows why what someone did is normatively correct in the light of one’s mental states. There may be no particular predictive payoff from the exercise. Indeed, it’s often thought that common-sense psychology is not predictive. Consider, for example, an ordinary conversation with someone you know well. You might be quite unable to predict what they’re going to say next; indeed, that’s one element that makes ordinary conversation worthwhile. But that inability to predict doesn’t of itself mean that it will be incomprehensible to you why the other person says the things she does. In fact, every single thing she says might be perfectly explicable in terms of her mental life, in that you understand why she says every single thing she does and why it was a good thing to say given the rest of her mental life.

This whole line of thought of separating the space of reasons from the space of causes is quite wrong. We can and do give causal mentalistic explanations, and the notion of causation here is exactly the same as the notion of causation that we use in the physical case. And although normative considerations do have a role in psychological explanations, they have a quite different role than is here envisaged.

2.RANDOMIZED CONTROLLED TRIALS OF THE MIND

Suppose you have to assess a number of paintings for their pictorial merit. Without being professional critics or even knowledgeable amateurs of art, many people would be willing to do this. You might have to do it without knowing who the painters are, or you might be told the painters; some of them may be quite famous. Would knowing who the painter was affect your assessment of pictorial merit? Judgments of specifically pictorial merit presumably ought to be independent of knowing the painter. But in a study by Hansen et al (2014), most of their subjects thought that the judgments of other people would be biased by knowing who painted what picture, and they thought their own judgments would be similarly biased. For the most part, each of the subjects thought that they themselves would tend to give more favorable assessments to the pictures known to be by famous painters. The subjects in this experiment were then divided into two groups. In one group, each person was given an array of paintings to assess for pictorial merit but without any identification of the painters. The other group was given the same array of paintings to assess, but some of the paintings were labeled with a famous painter as the author. The subjects who had been told the painters of various pictures did their best to assess the painting on their purely pictorial merit. In fact, they thought they had managed to do this. They thought their assessments were independent of the knowledge. But comparing their assessments to those of the group that had been given no information about the authors of the paintings clearly showed the effect. They were rating more favorably the pictures that they knew to be by famous painters.

This study illustrates a number of points about implicit bias. Most strikingly, it shows that

a.a belief, in this case a belief that a painting is by a famous painter, can be causally impacting your assessment of its pictorial merit, even though

b.you sincerely think that this belief is having no impact on your judgment, and

c.this can be so even though you agree that this kind of belief does in general have a biasing effect and even though you have agreed that you yourself are likely to be subject to this effect.

The whole point about these findings is that they’re established by scientific study. They aren’t manifest to common sense. You might suspect that they’re true or that they aren’t, but it’s a scientific study that determines whether they’re true.

On the face of it, this kind of study demonstrates that the dynamics of the mind can be studied by science in very much the same kind of way as the dynamics of any physical system. We can investigate the causal impact of a belief—in this case, the belief that a painting is by a famous painter—on your other thoughts and feelings. We do this using exactly the same experimental methods science uses to establish the causal impact of physical factors. In particular, we are here using a randomized controlled trial. To give a physical example, suppose you’re trying to determine whether a drug has any effect on an illness. You divide your subjects into two cohorts. You do this “randomly” in the sense that there isn’t any systemic factor differentiating the two groups that might conceivably have any causal bearing on the illness. You give the drug to people in one of your cohorts but not to the people in your other cohort—this second group being your control group. That’s the sense in which the trial is “controlled.” Then you look at whether there is any difference, on average, in the incidence or severity of the disease across the two groups. If there is any systematic difference between the two groups, that can only be an effect of the drug.

The logic here is compelling. Each of us trusts it whenever we take a drug. Of course, in any particular case, there are many problems to discuss about whether the general idea has been correctly implemented. For example, it’s notable that “random” here refers to the outcome of the process of selection rather than to the process of selection itself. Suppose people were sorted into two groups by flipping a fair coin for each of them to determine into which group they’d go. In one sense, the process itself then would be “random”: the factor used to determine which group a subject goes into, the flipping of the coin, presumably has no systematic causal impact on the outcome we are interested in (the disease the drug might protect against, for example). But the point of the thing might still not have been achieved. It might have been that by accident, one group had all the people with high levels of a hormone that confers natural resistance to the disease, and the other group had low levels of that hormone. In that sense, we would not have a successful randomization. We try to minimize the likelihood of this happening by having large group sizes. But it is often possible to wonder, in particular cases, whether group sizes have been sufficiently large and whether sufficient measures have been taken to guarantee that factors relevant to the outcome have not been operating to affect which group a patient gets into. For example, socioeconomic status might be a factor that in one way or another affects the incidence or severity of the disease. It might be that in the case of a disease for which there is no known drug treatment, patients of high socioeconomic status are particularly good at getting themselves into the treatment groups for experimental drug therapies. Still, these are problems in applying the general design. They do not point to any difficulty with the general design itself. This is the key way in which we experimentally demonstrate causality.

The very same design is what we use to demonstrate causality in the mind. That’s what was used in the example of knowledge about painters above. We divided our subjects into two groups, one of which got the information about painters and the other of which did not. To vary the example a little, suppose that we want to find the causal impact of a belief that the applicant for an academic post is a woman on the evaluations made by academic assessors. We divide our assessors into two groups. We give the two groups exactly the same information about the candidate, except that that one is given the information that the candidate is a woman and the other is not. Any systemic difference between the evaluations made by the two groups can be caused only by the difference in whether they believe that the candidate is a woman. That’s the logic of the design. Of course, here as in the drug trial, it is often possible to argue about whether the design has been correctly implemented. You can argue about whether we have managed to randomize the two groups in the sense that there isn’t any prior systemic difference between them, and so on. But the design itself seems compelling. We have here a way of establishing causation in the mind.

Indeed, arguably all of us use this way of establishing causation, at least implicitly, from an early age. Children show a sensitivity to statistics, and a use of statistics in causal reasoning, from astonishingly young ages (for reviews, see Gopnik and Wellman 2012 and Xu and Kushnir 2012). A sensitivity to statistical patterns shows up already at eight months old. For example, Xu and Garcia (2008) set up an experiment that used the fact that infants generally look longer at unexpected events. There were two urns. One visibly contained mostly white balls. The other visibly contained mostly red balls. When the researcher took a sample of mostly red balls from the mostly white urn, infants looked longer at it than they did at a similar sample of mostly red balls taken from the mostly red urn. They implicitly recognized that the probability of a mostly red sample being drawn from the mostly white urn was less than the probability of a mostly red sample being drawn from the mostly red urn. Kushnir, Xu and Wellman (2010) spun this point in an experiment with 20-month olds. The researcher again had two urns, one of which contained mostly toy rubber ducks. The other urn held mostly toy rubber frogs. Similarly to the Xu and Garcia paradigm, there were two conditions. In one, the researcher took a handful of all toy rubber frogs from the urn holding mostly ducks and played enthusiastically with them. In the other condition, the researcher took a handful of toy rubber frogs from the urn containing mostly frogs and played enthusiastically with them. The behavioral cues to emotional preference were the same in the two conditions. But the children in the first condition were more likely to select a frog to give to the researcher than the children in the second group. This indicates that children were using the nonrandom sampling as a guide to preference; the unlikely selection of all frogs from the mostly ducks urn was being taken to reveal a preference for frogs, whereas selecting only frogs from the mostly frogs urn was not interpreted as exhibiting any particular preference for frogs.

Alison Gopnik and her colleagues have argued for some time that children from around age four or five will use experiment and observation to find which characteristics of an object will make a machine’s lights flash or sound a bell, for example (Gopnik et al. 2004). They try objects with the characteristic and objects without the characteristic to see whether it makes a difference to the outcome. They also seem to be capable of using this approach to human psychology. Betty Repacholi and her colleagues set up an experiment in which children were presented with a bowl of cookies and a bowl of broccoli, and an experimenter asked them to give her food from one or other of the bowls. The experimenter enthusiastically mimed distaste for the cookies and a strong liking for broccoli. Nonetheless, children at fourteen months resolutely fed her cookies. They seemed not to understand the possibility that she might prefer broccoli. At eighteen months, however, the children transitioned: they fed the experimenter broccoli. Repacholi hypothesized that children only begin to understand the ver...

Table of contents

• Cover
• Title Page
• Copyright
• Dedication
• Contents
• Introduction: General vs. Singular Causation
• Chapter One: The Space of Reasons and the Space of Causes
• Chapter Two: Singular Causation
• Chapter Three: Social Robots
• Chapter Four: The Mind-Body Problem
• References
• Acknowledgments
• Index