An ongoing debate in philosophy of mind concerns the status of our everyday, “folk psychological” explanations of human actions – explanations that advert to the agents' intentions or goals. It is part of this folk picture that in cases where people do act for some particular reason, they know what that reason is. But the work surveyed by Newell & Shanks (N&S) suggests that this assumption is false. Indeed, some of it suggests not only that we may be frequently wrong about what our reasons are, but also that we may be wrong about having reasons at all. To be a reason for performing an action, a mental state must have semantic content, and that content must bear a rational relation to the agent's conception of the action to be performed. Thus, a judgment that stocking sample D is superior to the other samples would rationalize choosing sample D. But if the factor that in fact determines this choice is the spatial position of sample D, there is no rationalizing. Even if I were aware of a rightward bias, the thought that sample D is the rightmost sample would not give me a reason for choosing sample D. If such a factor were in play at all, it would likely be as a mere physical cause: The perceived position of the sample would immediately determine the choice, without any cognitive mediation at all. Unsurprisingly, then, reductivist and eliminativist critics of folk psychology point to work like that of Nisbett and Wilson to argue that our ordinary attributions of intentions are not only frequently false, but also explanatorily otiose (Churchland Reference Churchland1988).
Some philosophical defenders of folk psychology dismiss such arguments as irrelevant to the probity of the belief-desire framework: Explanation within such a framework, they say, is different from causal explanation, and can be justified independently of any causal account of the production of action (Blackburn Reference Blackburn1986; McDowell Reference McDowell, de Caro and Macarthur2004). I am, however, staunchly naturalistic in my approach to the mind, and so am committed to the continuity of philosophical and empirical work. I therefore recognize the in-principle relevance of the work in question and the seriousness of the challenge it poses. I am very glad, therefore, for the critique N&S offer. I would like to highlight some aspects of their critique that I find particularly germane to the philosophical debate, and then complement the critical points made by the authors with considerations that raise additional questions about the work in question.
N&S point out that Nisbett and Wilson's landmark “stocking study” (Nisbett & Wilson Reference Nisbett and Wilson1977), alluded to earlier, fails to meet at least one of their criteria of adequacy for assessments of awareness, namely, relevance. Although spatial position was correlated with subjects' choices, Nisbett and Wilson illegitimately presume that spatial position per se was causally relevant to their subjects' choices. But this inference neglects the possibility that subjects were running a left-to-right sequential evaluation of the stockings and operating with the rule “if the next one is as good as the previous one, go with the next one.” Indeed, Nisbett and Wilson themselves report evidence that suggests that subjects were doing something like this. I find this point especially significant for philosophy, because it underlines the importance of taking seriously cognitive states and processes as independent variables in the production of behavior, variables that must be studied and controlled for. This is as against the strict behaviorist model (which seems to have more vitality in philosophy than in psychology), which only considers publicly available factors – observable stimuli and behavioral responses – and the reductionist/eliminativist model, which says that it is otiose to posit states at any level of abstraction above the neurophysiological level.
But it's one thing to say that the proximal causes of the subjects' choices were cognitive and another to say that they were introspectible. If N&S are correct about the cognitive procedure the subjects were utilizing, why did the subjects not report that? Why did they insist that their choices were based on the superior quality of the stocking they chose? I have a hypothesis: The set-up of this experiment is a virtual invitation to confabulation. Since there is no good basis for preferring any one sample to any other, subjects will, typically, not be able to cite any such basis. Hence any reason proffered by the subject is going to be wrong. But what does a subject's behavior in this sort of circumstance tell us about the accuracy of introspection in cases in which the subject does have a reason for acting as he or she does?
What I am suggesting is that Nisbett and Wilson were investigating introspective awareness under degraded conditions. In general, it cannot be assumed that the way we solve problems in normal conditions is the same as the way we solve them in degraded conditions. (Consider the very different visual processes activated in daylight and in low light.) Inferences about the unreliability of a certain cognitive process in degraded conditions should not be taken as evidence that the same process is unreliable in normal circumstances. (If we assessed color vision by looking at its operation in low light, we'd conclude that we are terrible at judging colors.) It could well be, therefore, that introspection is highly reliable when our choices and actions are the result of reasons – that is, when there are reasons there to be introspected – but that we have to employ other methods of explaining our own behavior – perhaps, as Nisbett and Wilson suggest, theoretical inference – in cases where introspection finds nothing there. Of course, it would be very difficult to design an experiment to test the accuracy of introspection in what I'm assuming are the circumstances optimal for its operation. We would have to have circumstances in which the agent has a reason, and we know what it is. And it's hard to see how those conditions could be operationalized; it's much easier to set things up so that the agent has to be wrong. But of course, scientists should not be looking under the corner lamppost for watches dropped in the middle of the street.
I called the stocking comparison set-up a case of “degraded conditions.” The degradation here is the absence of any reason in the agent's mind for introspection to detect. Other kinds of suboptimality include hard cases – cases where there are or might be rational bases for decision, but these bases do not readily determine the best course – and marginal cases – cases where there are non-rational factors, such as emotional responses, that feed into the agent's decision. Asking for an agent's reasons in any of these circumstances is likely to provoke a state of mind similar to those that are called cases of dumbfounding in the literature on the psychology of moral judgment – cases in which subjects report strong moral judgments for which they offer no compelling moral justification. Accordingly, I would make a similar criticism of work that attempts to draw inferences about our ordinary moral reasoning from the responses subjects make in such cases: It is methodologically unsound to draw conclusions about our ordinary moral decision making from post hoc rationalizations of judgments about hard or marginal cases.
An ongoing debate in philosophy of mind concerns the status of our everyday, “folk psychological” explanations of human actions – explanations that advert to the agents' intentions or goals. It is part of this folk picture that in cases where people do act for some particular reason, they know what that reason is. But the work surveyed by Newell & Shanks (N&S) suggests that this assumption is false. Indeed, some of it suggests not only that we may be frequently wrong about what our reasons are, but also that we may be wrong about having reasons at all. To be a reason for performing an action, a mental state must have semantic content, and that content must bear a rational relation to the agent's conception of the action to be performed. Thus, a judgment that stocking sample D is superior to the other samples would rationalize choosing sample D. But if the factor that in fact determines this choice is the spatial position of sample D, there is no rationalizing. Even if I were aware of a rightward bias, the thought that sample D is the rightmost sample would not give me a reason for choosing sample D. If such a factor were in play at all, it would likely be as a mere physical cause: The perceived position of the sample would immediately determine the choice, without any cognitive mediation at all. Unsurprisingly, then, reductivist and eliminativist critics of folk psychology point to work like that of Nisbett and Wilson to argue that our ordinary attributions of intentions are not only frequently false, but also explanatorily otiose (Churchland Reference Churchland1988).
Some philosophical defenders of folk psychology dismiss such arguments as irrelevant to the probity of the belief-desire framework: Explanation within such a framework, they say, is different from causal explanation, and can be justified independently of any causal account of the production of action (Blackburn Reference Blackburn1986; McDowell Reference McDowell, de Caro and Macarthur2004). I am, however, staunchly naturalistic in my approach to the mind, and so am committed to the continuity of philosophical and empirical work. I therefore recognize the in-principle relevance of the work in question and the seriousness of the challenge it poses. I am very glad, therefore, for the critique N&S offer. I would like to highlight some aspects of their critique that I find particularly germane to the philosophical debate, and then complement the critical points made by the authors with considerations that raise additional questions about the work in question.
N&S point out that Nisbett and Wilson's landmark “stocking study” (Nisbett & Wilson Reference Nisbett and Wilson1977), alluded to earlier, fails to meet at least one of their criteria of adequacy for assessments of awareness, namely, relevance. Although spatial position was correlated with subjects' choices, Nisbett and Wilson illegitimately presume that spatial position per se was causally relevant to their subjects' choices. But this inference neglects the possibility that subjects were running a left-to-right sequential evaluation of the stockings and operating with the rule “if the next one is as good as the previous one, go with the next one.” Indeed, Nisbett and Wilson themselves report evidence that suggests that subjects were doing something like this. I find this point especially significant for philosophy, because it underlines the importance of taking seriously cognitive states and processes as independent variables in the production of behavior, variables that must be studied and controlled for. This is as against the strict behaviorist model (which seems to have more vitality in philosophy than in psychology), which only considers publicly available factors – observable stimuli and behavioral responses – and the reductionist/eliminativist model, which says that it is otiose to posit states at any level of abstraction above the neurophysiological level.
But it's one thing to say that the proximal causes of the subjects' choices were cognitive and another to say that they were introspectible. If N&S are correct about the cognitive procedure the subjects were utilizing, why did the subjects not report that? Why did they insist that their choices were based on the superior quality of the stocking they chose? I have a hypothesis: The set-up of this experiment is a virtual invitation to confabulation. Since there is no good basis for preferring any one sample to any other, subjects will, typically, not be able to cite any such basis. Hence any reason proffered by the subject is going to be wrong. But what does a subject's behavior in this sort of circumstance tell us about the accuracy of introspection in cases in which the subject does have a reason for acting as he or she does?
What I am suggesting is that Nisbett and Wilson were investigating introspective awareness under degraded conditions. In general, it cannot be assumed that the way we solve problems in normal conditions is the same as the way we solve them in degraded conditions. (Consider the very different visual processes activated in daylight and in low light.) Inferences about the unreliability of a certain cognitive process in degraded conditions should not be taken as evidence that the same process is unreliable in normal circumstances. (If we assessed color vision by looking at its operation in low light, we'd conclude that we are terrible at judging colors.) It could well be, therefore, that introspection is highly reliable when our choices and actions are the result of reasons – that is, when there are reasons there to be introspected – but that we have to employ other methods of explaining our own behavior – perhaps, as Nisbett and Wilson suggest, theoretical inference – in cases where introspection finds nothing there. Of course, it would be very difficult to design an experiment to test the accuracy of introspection in what I'm assuming are the circumstances optimal for its operation. We would have to have circumstances in which the agent has a reason, and we know what it is. And it's hard to see how those conditions could be operationalized; it's much easier to set things up so that the agent has to be wrong. But of course, scientists should not be looking under the corner lamppost for watches dropped in the middle of the street.
I called the stocking comparison set-up a case of “degraded conditions.” The degradation here is the absence of any reason in the agent's mind for introspection to detect. Other kinds of suboptimality include hard cases – cases where there are or might be rational bases for decision, but these bases do not readily determine the best course – and marginal cases – cases where there are non-rational factors, such as emotional responses, that feed into the agent's decision. Asking for an agent's reasons in any of these circumstances is likely to provoke a state of mind similar to those that are called cases of dumbfounding in the literature on the psychology of moral judgment – cases in which subjects report strong moral judgments for which they offer no compelling moral justification. Accordingly, I would make a similar criticism of work that attempts to draw inferences about our ordinary moral reasoning from the responses subjects make in such cases: It is methodologically unsound to draw conclusions about our ordinary moral decision making from post hoc rationalizations of judgments about hard or marginal cases.