Portal de Periódicos da CAPES

Guidance, epistemic filters, and non‐accidental ought‐doing

2019; Wiley; Volume: 29; Issue: 1 Linguagem: Inglês

10.1111/phis.12166

1758-2237

Maria Lasonen‐Aarnio,

Philosophical Ethics and Theory

One of the most familiar ideas in the domain of normative inquiry is the thought that normative principles should be able to guide us—or, more generally, that that the normative is action-guiding. In ethics, guidance has often been appealed to in order to defend subjective, as opposed to objective theories, subjective theories often being explicitly characterized in terms of guidance.1 For instance, classical utilitarianism has been thought by many to fail disastrously on the guidance front.2 Epistemologists have likewise invoked guidance, most often to criticize various externalist norms and theories.3 Almost all parties to these disputes agree that normative guidance requires epistemic access to a domain of facts: in order to be guided by a normative principle, one must have access to whether the conditions specified by the principle apply, and in order to be guided by a normative reason, one must have access to the reason itself.4 And herein lies the problem: guidance ambitions, many have argued, are doomed, for there is no domain of facts that we can invariably access.5 This is the Access Problem for guidance. If this is right, then it looks like there simply are no principles or norms that we are always in a position to be guided by in the desired way, though there might be a different, weaker kind of guidance that even more externalist or objectivist norms can provide. If the normative is action-guiding, then we are forced to draw the grim conclusion that there is no such thing. If normativity is to be salvaged, we must, it seems, settle for less and tone down our guidance ambitions. I won't here reiterate so-called anti-luminosity arguments attempting to show that there are no non-trivial luminous conditions, conditions such that whenever they obtain, we are in a position to know this. (I do, however, revisit the structure of anti-luminosity reasoning below.) My focus will be on a class of views that at first sight seem to fully sidestep such arguments, evading the Access Problem altogether. One might concede that the project of identifying some domain of facts that we purportedly always have access to is doomed to fail. In particular, views that internalize the facts grounding normative truths only yield guidance if we invariably have access to these internal facts. But we are sometimes in no position to know our beliefs, desires, appearances, seemings, motivations, or the quality of our will. For this reason going internal or subjective does not guarantee guidance.6 There is, however, the option of using epistemic access as a criterion on facts that determine normative facts in the first place, thereby requiring that such facts pass an epistemic filter.7 Those who press the Access Problem can no longer complain that we don't always have access to the domain of ought-making facts, for access is now a requirement on being part of that domain! Epistemic filtering is normally implemented in the context of a reasons-centered picture of normativity. It is first assumed that normative facts hold in virtue of facts about normative reasons: if I ought to ϕ (in a relevant sense of 'ought'), that is so in virtue of the fact that the overall balance of my reasons supports ϕ'ing. But there is an epistemic filter on these reasons: a fact can only count as a normative reason for a subject if she has epistemic access to it.8 According to a different implementation of filtering, only normative reasons that one has or possesses determine the relevant normative facts, and there is an epistemic access condition on possession.9 On either view, the potent normative reasons—reasons that enter into determining normative facts—are subject to a knowledge-constraint.10 Knowledge Filtering p is a potent normative reason for s just in case s knows p.14 I will argue that epistemic filtering does not solve the guidance problem, for it does not guarantee the availability of non-accidental ought-doing, a central good that guidance has been invoked by many to deliver. Interestingly, the reasons for this derive from some of the very same structural features of knowledge that the Access Problem rests on. Guidance If one ought to ϕ, then one can ϕ for the normative reasons in virtue of which one ought to ϕ.17 In what follows, my focus will be on views that endorse Guidance together with Knowledge Filtering. According to such views, if one ought to ϕ, then one can ϕ for the potent normative reasons in virtue of which one ought to ϕ, which are propositions one knows, or is at least in a position to know. The discussion below will be focused on a central good that guidance has repeatedly been invoked to deliver, that of non-accidental ought-doing18. It is often assumed that guidance is necessary and sufficient for acting as one ought in a way that is not a mere accident, fluke, coincidence, or matter of good luck: for instance, being guided by a principle is often opposed with mere accidental conformity to it.19 Indeed, a common complaint regarding objectivist theories and principles is that if these theories are true, and we lack access to the relevant domain of objective facts (as we often do), then we can only do what we ought in a haphazard, accidental kind of way. Many authors, however, eschew the possibility of cases in which one ought to do something, but can only act as one ought by accident. Because guidance, properly understood, involves not only acting as one ought, but acting as one ought in a way that is not merely accidental, the availability of non-accidental ought-doing requires that normative guidance always be available to us. It is worth noting that non-accidental ought-doing also plays a pivotal role in discussions of morally worthy action. The one thing that all views of moral worth seem to agree on is that an action that is morally right merely by accident lacks moral worth.20 Kant himself was concerned with cases in which one is motivated in the wrong way, and hence conformity to the moral law in a way that is "contingent and precarious".21 It is not a coincidence, then, that a popular view of moral worth on which morally worthy actions are actions performed for the reasons that make them parallels the responsiveness view of guidance.22 Why is non-accidental ought-doing important? Perhaps no deeper explanation is needed: it is valuable to succeed in a way that is not merely accidental, or a matter of good luck. But further, most parties agree that non-accidental success is closely connected with credit or praise. When one's action has moral worth, for instance, one is praiseworthy for doing the right thing.23 Given that non-accidental conformity to norms is necessary for conforming to them in a way that is praiseworthy or creditworthy, a core motivation for guidance is a need, via the notion of accidentality, to tie together the deontic and hypological.24 The thought is that it should always be possible to act as one ought in a creditworthy way—and hence, in a way that is not accidental. Guidance is supposed to guarantee this, in so far as non-accidental, creditworthy ought-doing just is a matter of ϕ'ing for the potent normative reasons in virtue of which one ought to ϕ. I will assume, then, that an adequacy condition on views of guidance is that being guided in the desired way entails that one acts as one ought in a way that is not merely accidental. I will argue that a responsiveness view of guidance, even when coupled with Knowledge Filtering, does not satisfy this condition: ϕ'ing for the potent normative reasons in virtue of which one ought to ϕ does not suffice for non-accidentally acting as one ought. My argument will draw on a modal condition on non-accidentality: a certain kind of modal robustness or invariance of a success is necessary for its being non-accidental. I will show how the modal condition offers a helpful diagnosis of why, in certain cases that have been central to attacks on more objectivist theories and norms, subjects are in no position to conform to these norms in a non-accidental way. As I see it, the modal condition is rather weak, and compatible with various full accounts of non-accidentality. Here are the bare bones of the modal condition I have in mind. Assume that a subject ϕ's—chooses a particular course of action, performs that action, forms a belief in a proposition p, etc. Assume that her ϕ'ing is successful: she believes, chooses, or does what she ought. For the success to be non-accidental, her way of ϕ'ing must issue in normative success in a sufficiently invariant manner across a range of relevant counterfactual cases in which one believes, chooses, or acts in that way. ϕ'ing in a way that is robustly successful in this sense is necessary for non-accidental success. Think of the relevant cases, very roughly, as ones in which we allow the idiosyncratic features of the actual case to vary in acceptable, somewhat normal ways.25 If one's success depends on the obtaining of some idiosyncratic feature, then it is accidental. The epistemology literature is replete with proposals for how to understand evaluations sensitive to, in particular, ways of forming beliefs: a belief is formed in a good way if it is properly based on sufficiently good, undefeated reasons or evidence, if it is the output of a reliable process, of it is formed by a reliable method. Ultimately, I prefer to think of these ways in terms of the dispositions that manifest as one's ϕ'ing, though the main points I make below are compatible with alternative views. My reasons for adopting the dispositional view have to do with its theoretical utility. Understanding ways in terms of dispositions allows giving a unified account of non-accidental successes—whether ones involving conscious deliberation, automatic actions, doxastic revision, or athletic performances.26 The general proposal, then, is that in order for one's ϕ'ing to be an instance of non-accidental ought-doing, ϕ'ing in that way must issue in normative success in a sufficiently invariant way across a relevant range of counterfactual cases. One's way of ϕ'ing must track what one ought to do across the relevant cases—and hence, it must track normative facts concerning what one ought to do.27 Given a dispositional understanding of ways, non-accidental ought-doing requires manifesting dispositions that don't, across a range of normal cases, manifest as acting in ways that one ought not to act. If I am in no position to manifest dispositions that discriminate between relevant cases in which I ought to ϕ, and those in which I ought not to ϕ, then I am in no position act as I ought in a non-accidental way.28 It is helpful to see how the modal condition on non-accidentality diagnoses why one is sometimes in no position to conform to objectivist norms or principles in a non-accidental way. Consider a well-known case from Jackson (1991: 463) involving Dr Jill, a physician who must choose which of three drugs to prescribe for her patient with a skin condition. Jill knows that drug A is likely to relieve the patient's condition without completely curing it. Further, Jill knows that one of drugs B and C will completely cure the condition, while the other will kill the patient. However, she has no way of telling which is which. Appeal to such examples has become the standard argument against objectivist theories and norms, such as a norm telling one to do what is best. In this case, this norm would require prescribing whichever of B and C cures the condition. But Jill is in no position to prescribe the best drug save in an accidental way! The modal account provides the following diagnosis of why this is so: no way of choosing between B and C is available to Jill that would track what is best across a range of counterfactual cases. Assume that B is in fact the cure, and C the harmful drug. Now consider available ways in which Jill might choose between B and C. She could, for instance, choose randomly, but randomly choosing would lead her to prescribe C, the bad drug, across a range of cases. It is also helpful to see how the modal account diagnoses how the same problem of accidentality arises for subjectivist theories and principles. A somewhat standard remedy to the no-guidance worry for utilitarianism is the following: instead of maximizing utility, we ought to maximize expected utility.29 At first sight, it may look as though such a view is able to solve the problem, for where actual utility is a function of contingencies regarding how the future plays itself out, expected utility is a function of one's credences or degrees of belief. But can we always track facts about our own credences? Those who press the Access Problem will argue that we sometimes lack access to our own doxastic states. Moreover, we sometimes lack access to mathematical facts about expectations. Assume, for instance, that I performed some complicated calculations regarding the expected utilities of various actions, but have forgotten the results. There are now three available courses of action, A, B, and C. I remember that A has a rather high expected utility. And I remember that the expected utility of one of B or C is very high, and that of the other is very low. This example looks like another Jackson-case: my predicament is structurally similar to that of Dr Jill.30 There is no way of choosing the action that in fact maximizes expected utility except in an accidental manner. We have looked at a range of examples in which lack of epistemic access to ought-making facts explains why we are in no position to act as we ought in a non-accidental way. At first sight, it looks like a view endorsing Knowledge Filtering are not susceptible to this challenge. In what follows, I will argue that epistemic filtering does not solve the problem of non-accidental ought-doing, for it makes normative facts depend on facts about our epistemic position, and we are sometimes in no position to track these facts. Knowledge Filtering guarantees that ought-making facts are never beyond one's ken. But one of its immediate consequences is that normative facts now depend on epistemic facts, facts about what one knows, or is in a position to know. Indeed, this is one of the main selling points of views that appeal to epistemic filtering, often defended by considering Jackson-type cases.31 In the original Dr Jill case, it is false that Jill ought to prescribe drug B. But now assume that she learns that drug B is in fact the cure, and C the killer. Given her new epistemic position, she ought to prescribe drug B. The only relevant change is her coming to know which drug cures the skin condition. True, the normative reason Jill acquires is that drug B is the cure, and not that I know that drug B is the cure: we can distinguish between what is in the set of potent normative reasons and criteria for getting there. But there is at least a modal, counterfactual kind of dependence between facts about one's epistemic position and normative facts. This point is worth emphasizing, for one of its consequences, as I will argue, is that sometimes non-accidental ought-doing will require being able to track the relevant facts about one's epistemic position in a modally robust way. And this, given the nature of knowledge, is not something we can always do. Before giving two examples in which this problem arises, let me describe their general structure. First, the normative fact that one ought to ϕ either (a) depends on one's knowing (being in a position to know) a relevant proposition p, or (b) on one's not knowing (not being in a position to know) p. In cases falling under (a), while p is among one's candidate ought-makers, a set of facts including all of one's other ought-makers except for p would determine that one ought not to ϕ. In cases falling under (b), while a proposition p is not among one's candidate ought-makers, a set of facts including all of one's candidate ought-makers and p would determine that one ought not to ϕ. Moreover, the relevant knowledge facts are not invariant across relevant counterfactual cases, and as a result of this, neither are the ought-facts. Finally, one is in no position to track whether or not one knows p across the relevant cases. As a result, one is unable to track the normative facts and hence, cannot act as one ought in a non-accidental way. Assuming that knowledge is subject to a margin-for error principle, knowing can sometimes be precarious in the sense that one could easily have been in no position to know a relevant proposition. Auto Belay Li knows that a child weighing exactly 15 kg or more is heavy enough to be lowered down by one of the auto belay devices. It would be a disaster, however, to let a child lighter than 15 kg climb: if they eventually got up, they would be left dangling, probably screaming, on the wall 16 meters above the ground, and it would be difficult to get them down – not to mention a huge PR disaster for the gym. Li sees a certain high-profile parent involved in local politics walk in with their daughter Ada. There is a lot of pressure to let the child climb, for the whole future of the gym depends on a favour from the parent. In fact, Ada weighs 16,05 kg. Li comes to know, just by looking, that she weighs over 15 kg. I will assume a margin for error principle on knowledge. In particular, given the accuracy of Li's estimates, had Ada been just 100 grams lighter, Li would have been in no position to know that she weighs at least 15 kg. As a result, Li just barely knows that Ada weighs at least 15 kg. In this sense, Li's knowledge is precarious: he could easily not have known. Moreover, given the absence of a sufficiently reliable scale (or parent), he could easily have been in no position to know. I will make the following assumptions about the case: Ada weighs 15 kg or more is an excellent reason to let her climb. Hence, given Knowledge Filtering, as long as Li knows this, he ought to let her climb. Moreover, it is difficult to see why the proposition that Ada weighs at least 15 kg couldn't be his motivating reason for letting her climb. It is, however, essential that this proposition be among his reasons, given that it would be very bad to let a child weighing less than 15 kg climb. Hence, propositions like it is probable that Ada weighs 15 kg or more, or it seems that Ada weighs 15kg or more just wouldn't cut it. (I revisit this below.) As a consequence, had she been 100 grams lighter, the normative facts would have been different due to Li's being in no position to know that the child weighs at least 15 kg. Li's knowledge is precarious. He could very easily have not known, and no way of acting or choosing is available to him that discriminates between his case and the case in which Ada is just 100 grams lighter. Moreover, this case is relevant for evaluating whether Li's normative success is accidental: being in no position to track the normative facts given very slight changes to the child's weight means that Li is in no position to act as he ought in a non-accidental manner.32 Similarly, consider a counterfactual case in which the child is a bit lighter, weighting 15,95 kg. In that case Li ought not to let the child climb. But again, Li is in no position to act as he ought in a way that is not accidental, since he could easily have known, in which case the normative facts would have been different. One might, of course, quibble about the details of the case. But if knowing can be precarious in the way assumed, and if a relevant normative fact can essentially depend on one's having such precarious knowledge, then examples with the structure described are bound to arise. Those who resist anti-luminosity arguments might argue that the kind of precarious knowing I have described does not exist. But at this point it is worth reminding the reader of the dialectic. My focus has been on appeal to epistemic filtering as an answer to the Access Problem for guidance. In a nutshell, the problem is that since there is no domain of luminous facts, guidance ambitions are doomed to fail. Epistemic filtering, it seemed, offered a way to sidestep the problem without contesting anti-luminosity reasoning. I have argued, in effect, that the very structural facts about knowledge that anti-luminosity arguments rely on create an accidentality problem even assuming Knowledge Filtering. My second example does not rely on the assumption that knowledge is subject to a margin for error principle. Notebook trouble Jill rushes to her office, and pulls out notebook #158. She opens the page on which she compiled her main findings on drug C, and written in the bottom of the page is the number 1: C is the killer drug, and therefore, Jill concludes, B is the cure! Alas, though this is true, Jill does not know it. About a week ago Jill had to leave her children alone in her office for 5 minutes. Four-year old Fanny, looking for something to apply her new eraser to, happened to pull out notebook #158 from the bookshelf and started randomly erasing numbers. She soon ended up on the page describing drug C, carefully erasing the '1' Jill had written on the bottom of the page. At this point eight-year old Eartha realized what was happening. Eartha asked what had been written on the page. The number 10, Fanny confidently (but falsely) told her. Not wanting to get into trouble, Eartha began carefully writing '10' on the bottom of the page. Very soon after she began, before she had time to write '0', she heard her mother's footsteps in the hallway, quickly closed the notebook, and put it back in its place. Dr Jill holds a justified, true belief that drug B will cure the skin condition, but she does not know this, for she is in a Gettier case: she could very easily have falsely believed that C is the cure, and B the killer, which she would have done had Eartha had time to complete her task. In fact, she may currently not even be in a position to know which drug is the killer. Hence, the proposition drug B will cure the skin condition is not among Dr Jill's potent normative reasons. Jill may know that it is likely on her evidence that drug B is the cure and that drug C the killer, but assuming the consequences of prescribing the wrong drug to be harmful enough, this does not suffice to make it the case that she ought to prescribe B. Similarly, she might know that it appears to her that drug B is the cure. But a mere appearance is compatible with B being harmful, and does not weight heavily enough in favour of prescribing B. In the absence of the proposition drug B is the cure, Jill's potent normative reasons just don't weight heavily enough in favour of prescribing drug B. (Below I discuss a reply insisting that they do.) The totality of potent normative reasons that Jill has in Notebook trouble is not relevantly different from those in our original Dr Jill case. Since she doesn't know that drug B is the cure, what Dr Jill ought to do is prescribe medicine A (that is the standard verdict in the original case).33 The important question is: can she act as she ought in a way that is not merely accidental? I think not. Relatedly, if Jill ends up prescribing drug A, acting as she ought, her normative success is not creditworthy or praiseworthy. This verdict can be supported by appeal to the modal condition on non-accidentality. Jill cannot discriminate her case from counterfactual cases in which she is not in a Gettier-case, and does know that B is the cure on the basis of consulting her notebook. Indeed, had everything been normal, she would have known. Any way of making a choice that results in prescribing medicine A will lead her astray in cases in which she knows that B is the cure. Compare a subject in a Gettier-case and a subject who forms a true belief on the basis of a random guess. Neither subject knows. However, as several epistemologists have pointed out, only the subject in a Gettier-case is unlucky not to know.34 Jill is unlucky not to acquire knowledge by consulting her notebook. The modal condition on accidentality outlined above explains this, for the way in which she forms her belief results in coming to know across a wide range of relevant, somewhat normal cases. As a result, her failure to know is merely accidental. Before concluding, I will briefly discuss a response to my arguments. One way to try to block my argument is to concede the claims I have made about knowledge, but to contest their implications for normativity in the context of Knowledge Filtering. In particular, one could contest the assumed dependence between facts about what Li and Jill know and facts about what they ought to do. Even had Li not known that Ada weighs at least 15 kg, its seeming to him that way would have sufficed to make it the case that he ought to let her climb. And in Notebook trouble Jill ought to prescribe medicine B after all, for even though she does not know that it is the cure, it seems or appears to her to be the cure, and she knows which number is written in her notebook, and these reasons are sufficiently strong on their own. Instead of debating the details of particular examples, we should ask what must be assumed for such a reply to be available across the board. One would have to subscribe to the general claim that p can't be a weightier reason than an appearance or seeming that p: if p would (if known) be a potent normative reason for s to ϕ, then its seeming to one that p (or the seeming itself) would be a potent normative reason to ϕ with the same weight.35 Moreover, in the overall balance of reasons, its seeming that p as it were screens off whatever weight p itself has: if, for instance, it first merely seems to me that p, and I then come to (be in a position to) know p, acquiring the new potent normative reason that p can make no difference for what I ought to do. Among our potent normative reasons, propositions concerning the world outside our appearances turn out to be normatively inert! I think this is extremely implausible. For instance, if p (obviously) entails q, then p is a maximally strong reason to believe q. But its seeming to me that p is not a maximally strong reason to believe that q, for it is compatible with q being false. One would think that the strength of one's reasons for believing p had something to do with probabilities. But unless one's prior probabilities rule out the possibility of misleading seemings or appearances, the probability of q on p will be higher than the probability of q on its seeming to one that p.36 I deliberately chose the examples so that the consequences of acting if the relevant propositions are false are disastrous, and in such cases it matters whether one's reason is that it seems that p, or that p is the case. But ultimately, even the backup view does not solve the problem. Let me once again remind the reader of the dialectic here. Anti-luminosity arguments, it seemed, simply have no bearing on views that have an epistemic filter on the potent normative reasons. But whether appeal to appearances or seemings, together with the backup view discussed, solves the problem I have raised depends on whether we can run an anti-luminosity argument for conditions such as it seems to me that p. One way to press the problem is to ask what it takes for propositions about one's seemings or appearances to be among one's reasons. If a proposition must be known in order to be among one's reasons, then the proposed view is susceptible to anti-luminosity reasoning: there are cases in which it seems to one that p, but one is in no position to know this. Precarious knowledge cases and Gettier cases arise for propositions about seemings themselves. If, on the other hand, one thinks that it is seeming-states themselves that figure among our potent normative reasons, and do so as long as we undergo them37, one still confronts cases with the structure that give rise to the anti-luminosity argument: if it can seem to me that p, but I cannot discriminate my case from a very similar case in which it doesn't seem to me that p, and if the normative facts depend on whether or not it seems to me that p, then I am in no position to track the normative facts in a way that is not accidental. These remarks point to a more general conclusion that goes beyond knowledge filtering. Consider, for instance, seemings filtering: in order for the fact that p to pass the filter, it must seem to one that p. According to such a view, normative facts come to depend on facts about how things seem to one. But then, in some situations acting as one ought in a way that is not accidental will require being able to track these seemings facts—which, given anti-luminosity reasoning, one is sometimes in no position to do. Normative guidance requires having access to facts in virtue of which normative facts hold. Think of these ought-makers as going into a box, the contents of which determine the relevant normative facts. The Access Problem arises because given a wide range of views about ought-makers, we don't always have epistemic access to them, not even if they belong to an internal domain concerning our beliefs, experiences, or motivations. Epistemic filtering would seem to bypass the problem, for it guarantees epistemic access. The new problem, however, is that epistemic filtering makes normative facts depend on facts about one's epistemic position, which we are sometimes in no position to track. As a result, we sometimes cannot act as we ought in a non-accidental way—and hence, cannot act as we ought in a creditworthy way. The very same structural features of knowledge that created the Access Problem for guidance in the first place seem to make non-accidental and creditable ought-doing unavailable in some cases, even if access to one's potent normative reasons is guarantee by Knowledge Filtering. It is instructive to see why piling on more access conditions won't help. Consider a more stringent epistemic filter: in order for a proposition p to pass the filter, one must know that one knows p. Normative facts now depend on facts about whether one has such iterated knowledge. The problem is that there will be cases in which p is an essential normative reason weighting in favour of ϕ'ing, but one is in no position to track the relevant epistemic facts, facts regarding whether one knows that one knows p. That is because knowledge that one knows p is susceptible to exactly the same phenomena as knowledge that p: one's knowledge that one knows can be precarious—one could easily have merely known—and one can have a Gettier-belief that one knows.38