Implicit Bias: The limits of control/character: Continued.


This post continues from the last one. I was saying that if we need to trace responsibility back to a suitable prior moment (t-1) at which the agent could have foreseen the consequences of his choices (following Vargas’s 2005 description of control theory), then we need to assess not only that agent’s internal capacities at t-n, but also the agent’s ‘moral ecology,’ and the relationships between the agent and the moral ecology. This follows from the fact that ‘indirect control’ is a matter of whether an agent could have exercised or acquired a capacity using available resources in her local environment. (Indeed, these two things are related: we acquire capacities in part by exercising more basic capacities: by taking piano lessons, I learn how to play piano well, i.e., I acquire piano-playing reasons-responsiveness. This is a result of exercising a more basic capacity – the basic human capacity to master a symbolic system with a combination of tutelage and practice). According to Holroyd, if someone is implicitly biased but could have avoided becoming implicitly biased or expressing an implicit bias, the person may be blameworthy. I am suggesting that to determine a person’s blameworthiness, we must evaluate, not just the individual person at t-n, but the person’s location in the social ecology.

This picture suggest a different kind of objection to Fischer’s notion of control – the dominant model. Fischer explicitly states that to appraise a person’s responsibility status, we must home in on the ‘actual sequence’ of deliberation, ignoring counterfactual circumstances in which the agent would have deliberated differently. That is, control for Fischer is ‘actual-sequence control.’ This strategy is (I believe) meant to undercut incompatibilist and nihilistic objections to control: objections to the effect that no one is responsible for anything because no one is capable of exercising ultimate control (being an unmoved mover) in a determinist universe (G. Strawson 1986), or because control is irrelevant in light of moral luck: whether you’re capable of exercising compatibilist control is just a matter of luck, not agency (Levy 2008). I think this last view places too much importance on the moral ecology and not enough on agency, but we can return to this later. By restricting control to the actual sequence, we cut off counterfactual circumstances in which the agent is metaphysically determined, as well as circumstances in which all agents are equally capable of control – distant possible worlds. But there is a case to be made that even if we include some counterfactual circumstances as morally relevant – and thus, some possible worlds – we do not need to include all counterfactual possibilities. Even if the buck doesn’t stop at actual-sequence control, it might stop in the next nearest possible world, cutting off the kind of slippery-slope objections that lead to nihilism.

I’m going to make this case in a moment, but first consider an existing objection to actual-sequence control, from Levy (2008). Levy argues that, while counterfactual disabling circumstances superficially appear to be irrelevant (consider, for example, Frankfurt-type cases in which the counterfactual device is never activated), it is possible to construct a counterfactual enabling circumstance that seems to make a difference. An enabling circumstance is one in which the agent gains a capacity, in contrast to the standard disabling scenarios, in which the counterfactual device would prompt the agent to go against the demands of morality if it were activated. Here’s one of Levy’s examples:


Jillian is walking along the beach when she notices a child growing. Jillian is a good swimmer, but she is pathologically afraid of deep water. She is so constituted that her phobia would prevent her from rescuing the child were she to attempt to; she would be overcome by feelings of panic. Nevertheless, she is capable of trying to rescue the child, and she knows that she is capable of trying. Indeed, though she knows that she has the phobia, she does not know just how powerful it is; she thinks (wrongly) that she could effect a rescue. Unbeknownst to Jillian, a good-hearted neurosurgeon has implanted her with a chip with which he monitors Jillian’s neural states, and through which he can intervene if he desires to. Should Jillian decide (on her own) to rescue the child, the neurosurgeon will intervene to dampen her fear; she will not panic and will succeed, despite her anxiety, in swimming out to the child and rescuing her. (2008: 170, 2008b: 234).

Levy actually presents this scenario variably as an objection to control theory and an objection to character theory. Suppose that Jillian decides not to rescue the child, in spite of (falsely) believing that she can. In the first place, it seems as if Jillian is responsible (says Levy) because she believes that she can rescue the child, and fails to act on this belief. But if Fischer is right, then Jillian isn’t responsible because she can’t succeed by her own (independent) means. Without the help of the benevolent intervener, success is impossible, and the intervener is external to Jillian’s motivational set. To vindicate the intuition that Jillian is responsible, we need to include the counterfactual intervener; so counterfactual scenarios seem relevant. (At least, this is my understanding). In the second place, if we include the counterfactual scenario as relevant, we have to accept that Jillian’s motivational set, and thus her character, include this scenario as a component part, and so Jillian’s character is “smeared” across time and space (Levy 2008: 179). Hence, locational externalism (i.e., the extended mind hypothesis) is true. These proposals call into question the viability of actual-sequence control and of character as traditionally conceived. I take it that this is supposed to support responsibility nihilism, i.e., the idea that responsibility (in a desert-entailing sense) doesn’t exist, as per Levy’s thesis in ‘Hard Luck.’

I think that that some of Levy’s claims are accurate and others need to be modified. Here’s what I endorse and what I dispute. I agree that counterfactual circumstances matter – at least, some counterfactual circumstances (specifically, those in which the agent could have intervened and succeeded with some kind of help), but I disagree with the ‘intuition’ that Jillian is responsible for an omission in this particular case. The reason is that, if someone, for bizarre reasons, thinks that she can save someone from drowning, but there is good objective reason to think that she can’t, the person doesn’t have a duty to intervene. Arguably, if we are not lifeguards, we don’t have a standing duty to save someone from drowning in any circumstance, because the risk of drowning ourselves in the attempt is too high, even if we are excellent swimmers for ordinary people (not lifeguards). So there’s no reason to think that Jillian ought to act on her belief that she can help. Normal people don’t have a standing obligation to risk their lives to save drowning victims. What we ought to do is alert the lifeguard, call emergency medical services, or look for an indirect means of intervening that doesn’t risk our own safety. So if we don’t have an obligation to risk our lives to save drowning victims, then people with water-related anxiety disorders certainly don’t, even if they have bizarre beliefs about their capacities and their moral duties.

I raise this point in part because this is the response I usually get when I present this case to other people. We could adjust the scenario by stipulating that the child is drowning in a wading pool. Normal adults have a duty to save children from drowning in wading pools, surely. But Jillian is not a psychologically-normal adult, so the worry recurs. Just as it is unreasonable for a normal adult to think that she has a duty to save a drowning victim in open water, it may be unreasonable for someone with a water-induced phobia, which could endanger her safety, to think that she has a duty to save a wading-pool victim. The worry is that the sacrifice required of the person is (objectively) too great to ground the existence of a duty, even if the person thinks that she has a duty for bizarre, subjective reasons.

But now consider a case in which it’s more obvious that a moral duty obtains. I’ll try to construct it to resemble the Jillian scenario, i.e., to include a protagonist who cannot achieve some end using her own (internal) capacities, but could succeed with external support. Suppose that Jack can’t contain his anger towards women, and regularly berates his female employees, family members, servers at restaurants, and so on. Jack (falsely) believes that he can control his misogynistic anger using willpower alone, and decides to exercise his willpower. But this is a false belief – he has much less willpower than he imagines. Unbeknownst to Jack, a local therapist specialises in anger-management problems, and would have been able to help him if he had sought out her help. But Jack never does. So Jack tries to exercise his willpower and fails, and continues to demean women on a daily basis.

If we home in on Jack’s internal capacity for control, we have to excuse Jack, since he lacked the internal resources to suppress his misogynistic urges. But if we consider the counterfactual scenario (in which Jack visits the therapist) as relevant to Jack’s responsibility status, then we can hold him responsible. It seems very reasonable to say that Jack had the capacity, in a very basic sense, to look for resources to control his misogynistic anger. But Jack does not have unassisted actual-sequence control over his anger.

I think that this scenario presents a plausible argument for the idea that, although Jack lacks actual-sequence control over his misogynistic anger, he has counterfactual-sequence control over it, and this counterfactual control is relevant to Jack’s responsibility status. Jack would not be risking his life by seeking out therapy. In fact, he wouldn’t be sacrificing anything of moral value. And if psychiatric counselling is free, as it is in Canada and some of the moral socialist countries, he isn’t even sacrificing anything of prudential value. By not seeking help, he’s not exercising his (basic human) capacities in the way he should. And because he’s not exercising these capacities responsibly, he has a character flaw.

Now, assuming that all of this is plausible, there’s an argument to be made that the proposed counterfactual-control model initiates a slippery slope into responsibility nihilism. Once we allow that counterfactuals are relevant, we have to admit that determinism precludes agency. But that’s only true if we hold all counterfactual possibilities – or at least, very many counterfactual possibilities – to be morally relevant. Yes, in a deterministic universe no one is incompatibilist-responsible for anything. But why go back to the metaphysical basis of reality – the metaphysical underpinnings of human behaviour? Why engage in ‘panicky metaphysics’ at all, when we can stop at the moral ecology? All I’m suggesting in constructing the above example is that some counterfactual possibilities matter – the ones that the agent could have availed himself of relatively easily, without sacrificing anything of moral (and in this case, even prudential) value.

Recall that in my last post I noted that control theorists typically espouse an implicit ‘reasonableness’ constraint in their conception of ‘indirect responsibility’: they hold an agent responsible for omissions for which it is reasonable to complain against the agent. As Levy says when commenting on implicit bias, an agent might “be fully morally responsible for [a] behaviour [resulting from implicit bias], because it was reasonable to expect her to try to change her implicit attitudes prior to t” (2014). This implies that only counterfactual circumstances that were reasonably available to the agent are morally relevant – scenarios in which the sacrifice demanded of the agent is not overly stringent. I have argued that in the Jillian scenario, the moral demand is too high, but in the Jack case, it’s not. This gives us a foundation on which to say that someone can be responsible for failing to exercise counterfactual control if doing so was reasonable. Of course, ‘reasonableness’ is a vague concept, and I won’t precisify it here, but it’s a concept that makes intuitive sense, and one that we regularly rely on without clarification, to constrain the scope of normative concepts. (Consider Scanlon’s account of moral principles as those that no suitably-motivated person could reasonably reject; we get the point without delving into the semantics).

Character theorists of Sher’s stripe similarly hold that reasonableness is critical to responsibility. A person is responsible for an omission just in case a reasonable person with relevantly similar capacities would have done better. So once again, we are to judge the agent’s responsibility status by what it would be reasonable to expect of her, in light of certain counterfactual possibilities – what the agent could have achieved under conditions C.

If this is right, then counterfactual circumstances do matter morally. But not all, or just any, counterfactual circumstances are relevant. Just those that were reasonably available to the person, at relatively low personal cost. This goes back to what I was saying about the moral ecology, and about tracing. Counterfactual scenarios are part of the moral ecology, external to the person’s material brain. So when evaluating a person’s responsibility status, we have to consider the person’s relevant brain states, and the reasonably available counterfactual circumstances supported by the agent’s moral ecology. We have to look at the person’s capacities and the person’s moral ecology and the potential interaction between those two variables, to see if they support the possibility of agency cultivation. And with regard to tracing, we need to trace responsibility to that possibility. We need to assess whether Jack, for example, had the (general) capacity to acquire the (specific) capacity to remediate or suppress his misogynistic anger, given the resources of his moral ecology.

This suggests the following revisions to control theory and character theory. We need to see control as more than actual-sequence control to account for the possibility of indirect responsibility for omissions that were reasonably avoidable. Specifically, we need to include as morally relevant, counterfactual circumstances in which the agent could have interacted with the moral ecology in such a way as to bring about a new capacity, or (more precisely) to leverage an existing basic (undifferentiated) capacity into a more specific (specialised) capacity. And it suggests that we ought to regard character as diffuse or locally extended, i.e., co-constituted with agency-supporting or agency-enhancing social supports. (This is not a revision to Sher’s view, in fact, but it emphasises that aspect of it). And finally, it suggests that when ‘tracing’ back to control, we need to trace beyond the agent’s actual sequence, to reasonably available aspects of the agent’s moral ecology, which would have enhanced the agent’s capacities if the agent had taken the right kind of initiative. And similarly with character theory, we need to trace character to relevant features of the local ecology, to determine if the agent is using those features to the best of her ability. If she is not, she may be culpably indifferent. (In this way, tracing applies to character theory as well, though its reach is more limited – we don’t need to trace as far back).

These considerations strike against any theory that is too narrow in its conception of responsible agency, particular the actual-sequence control model. And it suggests that control theory and character theory are perhaps more similar than they may initially appear, in that control theory admits a greater scope for blame under the category of ‘indirect responsibility,’ properly understood. These considerations build on Levy’s objections to actual-sequence control and character internalism. But I recruit them to show that there is a broader scope for blame than we tend to think, and he does the opposite – he recruits them in support of responsibility nihilism. Our views are technically compatible, though, because he’s refuting a desert-based notion of responsibility, and a relatively harsh form of desert-based responsibility, on which blame (1) is justified by reference to an agent’s actual sequence of deliberation or internal traits, and (2) entails fairly punitive sanctions. I also reject this notion of responsibility, because it combines a metaphysically tenuous conception of agency with dubious assumptions about the kind of thing blame is (punishment) and proportionality (harshly punitive). I think that the same objections can be taken to support a modification of responsibility rather than a rejection of it.

Here’s one substantive alternative to the actual-sequence control model – the ‘limited counterfactual-sequence control model.’ People are responsible for (1) intentional infractions (like explicit bias), (2) failures to properly exercise control (e.g., manifestations of implicit bias that could have been avoided by suitable reflection), and (3) failures to enhance the capacity for control (e.g., failing to search for remediating measures available in the local moral ecology, when it would be reasonable to do this). And here’s a viable version of character theory: people are responsible for character defects just in case those defects could have been remediated with reasonable effort, using the resources of the local ecology. In case it’s not obvious, here’s why the local ecology matters for character theory. If Smith is a misogynist because he lives in 1950s middle America and doesn’t have access to good examples of egalitarian behaviour, while Jones is a misogynist in present-day New York just because he hates women, Jones has worse character than Smith, because he’s not only a misogynist, he’s also indifferent to women’s interests. That is, Jones exhibits a greater degree of indifference than Smith (see my 2015 paper and my 2013 paper for lengthier examples, and see also Fricker 2012). Some theorists assume that tracing doesn’t apply to character theory, but that’s false. We have to trace the causal source of a character defect, to see if the character defect is amplified by indifference to available reasons.

These are two viable versions of control theory and character theory that present plausible alternatives to responsibility nihilism. But there’s a third option that may seem to be a better fit with what I’ve said so far. It’s a consequential approach, along the lines of Vargas’ agency cultivation model (2013). On that view, we’re responsible to the extent that praise or blame is likely to enhance our agency (very crudely put). Here’s how this view works. Suppose that Smith is a misogynist who berates all the women in his life, but Smith is still a moral agent (not a full-blown, unresponsive psychopath or something of that nature; he has some vestige of the capacity to respond to reasons). Blame might function to enable or enhance Smith’s capacity to respect women (and it might function that way for all misogynists – let’s suppose that this is its general effect). So Smith is blameworthy. This approach, because forward-looking, might seem to eliminate the need for tracing, which might seem to be a desideratum, since tracing is hard. But I don’t think it does. First, we need to know if Smith is a misogynist, as opposed to, say, a foreigner who doesn’t know he’s using a misogynistic slur, or a brain-washing victim, or someone whose family is being held hostage on condition that he demean his female acquaintances, etc. I artificially stipulated that Smith is a misogynist above, but in real life, we need to discover things about a person’s circumstances to make correct moral appraisals, so we need to get to know people and inquire into their lives. Second, we might want to consider whether Smith had opportunities to develop a more egalitarian sensibility – control-based considerations. The point is, even on a forward-looking account, we need to know things about an agent’s capacities and environment, and so we need to do some non-negligible amount of tracing. We can’t just guess what someone is like on the basis of one time-slice. People are notorious for jumping to conclusions, but to be responsible in our responsibility attributions, we need to be committed to giving fair consideration to relevant data.

The upshot is that there are convincing arguments against narrow versions of control theory and character theory, but they don’t force us down a slippery slope to responsibility nihilism. There are viable (extended) versions of control theory and character theory that we can adopt; and consequentialism is also an option. But we should not, I think, assume that we can do away with tracing on any of these alternatives. If anything, once we grant that the moral ecology is relevant to responsibility –  more relevant than we might have previously thought – we have to extend the scope of tracing beyond the agent’s material brain. But I think that we implicitly do this anyways (in our ordinary judgments of praise and blame), which is why we consider people from past times and foreign cultures to be less blameworthy for certain infractions. Yet some accounts of responsibility don’t adequately explain this kind of contextual assessment – they don’t sufficiently appreciate the significant of context. Circumstances matter because they co-constitute, enable, and support – or conversely, impair – the capacities that underwrite responsible agency.


Here’s how all of this relates back to implicit bias. As I said in my first post on implicit bias, it’s not at all clear how wide the scope of control has to be for responsibility to obtain. If we think that reasonably-available counterfactual circumstances are morally relevant, we can hold people responsible for exhibiting implicit biases if they failed to use remediating measures that were locally available, provided that this was a reasonable expectation. People with special duties – people on hiring committees, for example – have stronger reasons to use these measures, and thus are more susceptible to blame for relevant omissions. And this is true even if they lack responsiveness to such measures now, provided that they could have acquired suitable patterned sensitivity at some time in the past, by a reasonable effort. Ordinary people can be blamed if we think that it was within their ability to avoid manifesting implicit bias through some reasonable act of will. The case for blame is stronger if we admit counterfactual circumstances into the equation, because then we have grounds for saying that someone is (indirectly) responsible for an omission, just in case a counterfactual enabling circumstance was within reach. This brings the view somewhat closer to modern character theory in its scope for attributing blame.



Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s