Hilary Greaves (Oxford): Cluelessness

p r o c e e d i n g s o f t h e a r i s t o t e l i a n s o c i e t y1 3 7 t h s e s s i o n

i s s u e n o . 3v o l u m e c x v i2 0 1 5 / 2 0 1 6

c l u e l e s s n e s s

h i l a r y g r e a v e s o x f o r d u n i v e r s i t y

m o n d a y, 2 3 m a y 2 0 1 6

1 7 . 3 0 - 1 9 . 1 5

t h e w o b u r n s u i t es e n a t e h o u s eu n i v e r s i t y o f l o n d o nm a l e t s t r e e tl o n d o n w c 1 e 7 h uu n i t e d k i n g d o m

This event is catered, free of charge &open to the general public

c o n t a c [email protected]

© 2016 the aristotelian society

b i o g r a p h y

Hilary Greaves is Associate Professor in Philosophy at the University of Oxford. Her research interests include foundational issues in consequentialism (‘global’ and ‘two-level’ forms of consequentialism), aggregation, moral psychology and selective debunking arguments, population ethics, the interface between ethics and economics, the analogies between ethics and epistemology, and formal epistemology. She currently directs a three-year project on population ethics, funded by The Leverhulme Trust.

e d i t o r i a l n o t e

The following paper is a draft version that can only be cited or quoted with the author’s permission. The final paper will be published in Proceedings of the Aristotelian Society, Issue No. 3, Volume CXVI (2016). Please visit the Society’s website for subscription information: www.aristoteliansociety.org.uk.

4

c l u e l e s s n e s s

h i l a ry g r e av e s

All plausible moral theories agree that considerations of the conse-quences of one’s actions form an important part of the story. But, several authors have worried, we are so clueless about consequences - both at the time of action, and even at any later time of evaluation of that action - that in practice we are not able to take account of any considerations of consequences, either for the purpose of deliberation or for that of evaluation.

A standard response is that this worry can be circumvented by appealing to the subjective expected value of the consequences of one’s actions, rather than to their actual value. Replying to this standard response, Lenman has objected that this succeeds in defusing the cluelessness worry only given a Principle of Indifference, while (he insists) the Prin-ciple of Indifference is indefensible. Against Lenman, I argue that *for the kinds of cases of would-be cluelessness that have been discussed to date* (and, in particular, the ones Lenman has been discussing), an ap-propriately restricted Principle of Indifference is perfectly acceptable, so that an appeal to subjective expected value does indeed dispose of the cluelessness concern.

There is, however, a second class of cases that have quite a different structure, and for which indifference-based reasoning does seem deeply inappropriate, in such a way that cluelessness genuinely does threaten. After highlighting this class of cases, I turn to the question of what is the correct theoretical description of the predicament of a clueless agent, in terms of epistemic and practical rationality. This is a question that has so far received little attention. Finding an account that does adequate justice to our felt sense of cluelessness – if that sense is any more than an artefact of our irrationality – turns out to be remarkably difficult.

i. cluelessness about objective betterness ASSUME DETERMINISM.1 Then, for any given (sufficiently precisely de-scribed) act A, there is a fact of the matter about which possible world would be realised – what the future course of history would be – if I per-formed A. Some acts would lead to better consequences (that is, better future histories) than others. Given a pair of alternative actions A1, A2, let us say that

(OB: Criterion of objective c-betterness) A1 is objectively c-better than A2 iff the consequences of A1 are better than those of A2.

1 Relaxing this assumption would complicate some parts of the discussion, but not in ways that are ultimately relevant to the issues in this paper.

draft paper

5

hilary greaves cluelessness

It is obvious that we can never be absolutely certain, for any given pair of acts A1, A2, of whether or not A1 is objectively c-better than A2. This in itself would be neither problematic nor surprising: there is very little in life, if anything, of which we can be absolutely certain. Some have argued, however, for the following further claim:

(CW: Cluelessness Worry) We can never have even the faintest idea, for any given pair of acts (A1, A2), whether or not A1 is objectively c-better than A2.

This ‘cluelessness worry’ has at least some more claim to be troubling (I return in section 2 to the question of whether it is troubling; and in section 6 to the question of what exactly “can’t have the faintest idea” means). This is most obvious in the case of consequentialism. For if (CW) is cor-rect, and if in addition (as consequentialism holds) the moral status of an action is determined entirely by how it compares to alternative actions in terms of the goodness of its consequences, it seems to follow with particu-lar clarity that we can never have even the faintest idea what the moral status of any given action is. But any plausible moral theory will agree that considerations of consequence-goodness are at least morally relevant – that they should be taken serious account of both in moral decision-making and in moral evaluation, as at least one important factor. And this too seems impossible in practice if (CW) is correct. 2

The argument for (CW) stems from the observation that the relevant consequences include all consequences of the actions in question, through-out all time. In attempting actually to take consequences into account in practice, we usually focus on those effects – let us call them ‘foreseeable’ effects – that we take ourselves to be able to foresee with a reasonable degree of confidence.3 And while we are arguably correct in thinking that we are justified in being reasonably confident in our predictions of these effects, any choice of one act A1 over another A2 inevitably has countless additional consequences that our calculation takes no account of. (A but-terfly flapping its wings in Texas may cause a hurricane in Bangladesh; so

2 Here I concur with Kagan (Normative Ethics, p.63) and Smart (An outline of a system of utilitarian ethics, p.34), both of whom initially raise the issue of cluelessness in the context of consequentialism, but then note that in fact the problem affects a much wider class of moral theories. In contrast, some others (e.g. Lenman (‘Consequentialism and cluelessness’)) appear to regard the problem as peculiar to consequentialism, although they do not give reasons for this.

3 More precisely: effects that we have some particular reason to regard as more likely to follow on some courses of action than on other courses of action. That is, the ‘foresee-able’ effects need not exclude e.g. possible effects that are extremely unlikely either way, but whose probabilities are affected in definite ways by our choice of action. Nor need they exclude effects that are remote from our action in time and/or space, or that follow only via convoluted and lengthy causal chains.

6

hilary greaves cluelessness draft paper

too may my telling a white lie, refraining from telling that lie, moving or not moving my hand; a hurricane will certainly affect which other but-terflies flap their wings or which other agents move their hands in which ways; and so the effects will ripple down the millennia.) Any conclusion, on the basis of the calculations that we have carried out, that one act is indeed objectively better another is justified only insofar as we are justified in assuming

(NR (Non-reversal)) The net effect of taking into account all of these ad-ditional effects would not reverse the judgment that we reach based on the foreseeable effects alone.

But is (NR) true?

Here are two bad arguments for (NR). First, one might think that while there are indeed non-zero effects, traceable to even the most trivial of one’s actions, stretching down through the millennia, still the magni-tude of any individual such effect typically decays with time. Further – letting ΔV≔ V(A1)-V(A2) be the amount by which the goodness of the consequences of A1 exceeds that of A2 – one might think that this decay is sufficiently fast that by far the largest contribution to ΔV comes from the direct effects. This line of thought is suggested (though not strongly advocated4) by Moore:

“As we proceed further and further from the time at which alternative ac-tions are open to us, the events of which either action would be part cause become increasingly dependent on those other circumstances, which are the same, whichever action we adopt. The effects of any individual ac-tion seem, after a sufficient space of time, to be found only in trifling modifications spread over a very wide area, whereas its immediate effects consist in some prominent modification of a comparatively narrow area. Since, however, most of the things which have any great importance for good or evil are things of this prominent kind, there may be a probability that after a certain time all the effects of any particular action become so nearly indifferent, that any difference between their value and that of the effects of another action, is very unlikely to outweigh an obvious differ-ence in the value of the immediate effects.” (Principia Ethica, section 93)

Similarly Smart (who apparently does advocate it):

“[W]e do not normally in practice need to consider very remote con-sequences, as these in the end rapidly approximate to zero like the fur-thermost ripples on a pond after a stone has been dropped into it.” (‘An outline of a system of utilitarian ethics’, p.33)

4 Unlike Smart, Moore for the most part confines himself to asserting the necessity of defending (NR), by the means suggested or otherwise, “if any of our judgments of right and wrong are to pretend to probability” (Moore, ibid., section 93).

draft paper

7


This, though, is not plausible. To see this most vividly, note that even our most trivial actions are very likely to have unforeseen identity-affect-ing effects (although the same points could be made without appeal to identities-variation).5 Suppose, for example, that I pause on my way home from work, in order to help an old lady across the road. As a result, both she and I are in any given place – any given position on the pavement for the remainder of our respective journeys home, for instance – at different times, at least for the remainder of that day. As a result, we advance or delay the journeys of countless others, if only by a few seconds, relative to the situation in which I had not helped her across the road; both we and they affect which further parties enjoy chance meetings with whom; and so forth. At least some of these others were destined to conceive a child on the day in question, and if so, even our trivial influences on their day will affect, if not whether they conceive, then at least which particular child they conceive (since a delay in sexual intercourse of even a few seconds is overwhelmingly likely to affect which particular sperm fertilises the egg).6 But once my trivial decision has affected that, it equally counts as causally responsible for everything the child in question does during his/her life (i.e., for the differences between the actions/effects of this child vs. those that the alternative, in fact unconceived, child would have performed/had) – and of all the causal consequences of all those things, stretching down as they do through the millennia. These are clearly not negligible: many or most of the things that one child does and that the alternative child would not have done (or vice versa) amount to greater differences in goodness than the intrinsic value of one old lady’s receiving help across the road on one occasion. And nor is it at all likely that the number of identities my action affects in generation r will decrease as r increases; on the contrary, it will increase.

Might one resurrect (NR) by arguing that although there are, for any choice of a given action A1 over an alternative A2, countless effects of significant size stretching arbitrarily far into the future, that nonetheless these unforeseeable effects are highly likely to cancel one another out, and to do so to an arbitrarily high degree of precision (and hence to have a combined effect that is much smaller than the foreseeable effect) as the time horizon stretches to infinity? Unfortunately not; the theory of ran-dom walks tells us that while some degree of cancelling-out in such situa-tions is all but certain, the combined effect of a large number n of proba-bilistically independent steps does tend to grow with n, and in particular that it is highly unlikely to end up anywhere sufficiently close to zero.7 It

5 This point has been made forcefully by Lenman (‘Consequentialism and cluelessness’).

6 This point has been argued for, in a different context, by Parfit (Reasons and Persons).

7 More precisely: A one-dimensional simple symmetric random walk is a series {Sn}n=1,2,…,

8


is extremely unlikely, for instance, that the difference in net value between everything this child does in his/her life on the one hand and everything the alternative child would have done in his/her life on the other will just happen to be smaller than the intrinsic value of one old lady’s receiving help across the road on one occasion, even if we pretend that each of a child’s actions is probabilistically independent of each of the same child’s other actions; and increasing the number of children involved will only exacerbate the problem.

These arguments against possible defences of (NR) are equally argu-ments against (NR) itself. We are forced to conclude that (NR) is false, and hence (CW) true.

ii. cluelessness about subjective c-betterness The truth of (CW) would be troubling, however, only if it followed that there was no way for considerations of consequences to guide either deci-sions or evaluations; and (OB) is not the only possible route for that to happen. In fact, consequentialists in particular have long recognised both the availability and the indispensability of a second such possible route, viz. the appeal to a relation of subjective c-betterness among actions:

(SB: Criterion of subjective c-betterness) Act A1 is subjectively c-better than A2 iff the expected value of the consequences of A1 is higher than the expected value of the consequences of A2 (where both expectation values are taken with respect to the agent’s credences at the time of decision8).

At least at first sight, the criterion (SC) is useful in practice for pur-poses of decision and evaluation, the considerations discussed in section

1 notwithstanding. For consider any possible but unforeseeable future ef-

where (1) for each n, Sn = ∑j=1n Zj, and (2) the Zj are independent random variables, each

of which takes the value +1 or -1 with equal probability. It can be shown that in such a series, the expectation value of the magnitude |Sn| is proportional to the square root of n. Thus, in particular, in the limit n→∞, this expectation value E[|Sn|] tends to infinity, rather than to zero. Returning to our case of interest: If (simplifying) we assume that the ‘effects’ of each possible action can be parcelled into discrete components, each of which additively contributes an amount an amount of fixed magnitude but variable sign (thus: either ΔV or –ΔV) to the overall goodness of the world in which it occurs, and that the signs of successive effects are probabilistically independent, then this theorem applies to the case of interest in the way suggested in the main text.

8 Or perhaps: the probabilities that are supported by the evidence that the agent pos-sesses at the time of decision, i.e. the relevant ‘evidential probabilities’ (REFS). For the majority of our discussion, the distinction between subjective credences and evidential probabilities will be of little import. Consideration of evidential probability would affect some of the discussion of section 6, but only in minor ways.

draft paper

9


fect9 E1↦E2 that might, via the sorts of mechanisms we considered there, result from my decision to perform act A1 rather than A2. For sure, it is possible that: if I did A1 then E1 would result and if I did A2 then E2 will result (in symbols: A1 →E1 & A2 →E2). Still, there is no particular reason to think that the correlations between my possible actions and these un-foreseeable effects will be that way round, rather than the opposite (A1 →E2 & A2 →E1). It seems plausible, in that case, that given any credence function that it is rationally permissible for me to have at the time of deci-sion, my credence in the second correlation hypothesis is exactly equal to my credence in the first correlation hypothesis. But if this is true for all un-foreseeable possible effect-changes E1↦E2, then the contribution of those unforeseeable effects to the difference in the expected values of A1 and A2 is precisely zero, and we have the following result:

(EVF) The expected c-value of an action is determined entirely via its foreseeable consequences.

And if (EVF) is correct, then there can be no analogue of the cluelessness worry for subjective c-betterness.

iii. lenman’s objection: against the principle of indifference Lenman (‘Consequentialism and cluelessness’) objects to the reasoning in section 2 on the following grounds: this reasoning presupposes the Prin-ciple of Indifference, but that principle is false.

To state the principle of indifference, we require a notion of evidential symmetry between mutually exclusive propositions. This notion is sup-posed to capture the idea of our having no more evidence in favour of one proposition than the other, or no more reason to believe one proposition than the other.10 In particular, we suppose that two propositions are evi-dentially symmetric when we have no evidence that bears on the question of which of the two is true (say, on the assumption that one or the other is true). The Principle of Indifference can then be stated as follows:

(POI: Principle of Indifference) Let Q1,…,Qn be any mutually exclusive propositions that are evidentially symmetric for S, and let Q be their dis-

9 With slight abuse of terminology, where the context prevents any confusion from re-sulting, I use ‘effect’ both in the absolute sense (E1 would be among the effects of choos-ing A1) and in the comparative sense (the transition E1↦E2 would be an effect of choosing A1 over A2). It is, of course, the comparative sense that us ultimately important for the purposes of c-betterness, and hence of ethics.

10 I follow White (‘Evidential symmetry and mushy credence’) in employing the ter-minology ‘evidential symmetry’ in an attempt to be neutral between these alternative formulations.

10


junction. Let C be any credence function that is rationally permissible for S. Then for all i,j, C(Qi|Q)=C(Qj|Q).

At first sight, this principle looks eminently reasonable. It also seems to tell the right story in at least some cases. For example, suppose you know that I am about to flip a coin, and you know nothing else relevant to the question of whether it will land Heads or Tails. (In particular, you have no information about whether or not the coin is fair, or about the mechanism by which I will flip it.) Plausibly, you are rationally required to have credence ½ that my coin will land Heads; any other credence seems unacceptably arbitrary.

Lenman’s claim that the reasoning in section 2 presupposes some form of Principle of Indifference is quite correct. For that reasoning relies cru-cially on the claim that one’s credence that if the agent did A1 then E1 would result and if the agent did A2 then E2 would result is rationally re-quired to be equal to one’s credence in the opposite act-effect correlation (i.e. that if the agent did A1 then E2 would result and if the agent did A2 then E1 would result). But the only reason given for thinking this is that we have no more reason to believe that the former correlation obtains than the latter, or vice versa. The required claim follows only if we assume something like POI for the present case. Otherwise we have no resources with which to criticise an agent who arbitrarily has credence (say) 0.9 that A1 (resp. A2) would lead to ‘unforeseeable’ effect E1 (resp. E2), while acknowledging that she has no reason or basis in evidence for favouring this correlation over the other.

As is well known, however, an unrestricted POI (such as the one stated above) is inconsistent, at least unless the relation of evidential symmetry holds between far fewer proposition-pairs than we would naively have assumed11. The difficulty is the ‘problem of multiple partitions’. It arises from the fact that for any partition {Q1, …, Qn} of Q, there are many other partitions of Q, none of which we are currently able to single out as privi-leged; and POI generally gives mutually inconsistent results when applied to distinct partitions.

This problem arises, in particular, when one partition is a ‘selective fine-graining’ of another: that is, the second partition involves further fine-

11 White (ibid.) argues persuasively that the culprit in these paradoxes may indeed be a too-liberal interpretation of ‘evidential symmetry’, rather than POI itself. This is an important point for the general discussion of POI. Since such ‘shifting of the bump in the carpet’ would not in the end fundamentally change the state of the debate for current purposes, however, here I set it aside for simplicity of exposition, and assume that POI itself is shown to be at fault by the ‘problem of multiple partitions’ discussed in the main text.

draft paper

11


grainings of some elements of the original partition but not others. Sup-pose, for instance, you know only that I am about to draw a book from my shelf, and that each book on my shelf has a single-coloured cover. Then POI seems to suggest that you are rationally required to have credence ½ that it will be red (Q1=red, Q2 = not-red; and you have no evidence bear-ing on whether or not the book is red), but also that you are rationally required to have credence 1/n that it will be red, where n is the ‘number of possible colours’ (Qi = ith colour; and you have no evidence bearing on what colour the book is).)

This problem would be of merely theoretical interest if it was intui-tively clear, in any given example, which partitions were ‘natural’ and which ‘gerrymandered’. For in that case, we could restrict the Principle of Indifference to ‘natural’ partitions, and even without a precise criterion for naturalness, we would know when to apply vs not to apply POI in practice. And that is arguably an adequate response to the book-colour example: at least on reflection, it is clear that neither the partition {book is red, book is not red} nor the partition {book is red, book is blue, book is yellow…} is especially natural12, so perhaps this is just a case in which POI clearly falls silent.

Unfortunately, however, there are at least some wide classes of cases in which it is not even intuitively clear which partitions should be regarded as privileged for the purposes of POI. In particular, this often happens in scenarios involving credences about some continuous quantity, as con-tinuous quantities are apt to have multiple natural parametrisations that are non-linearly related to one another. Suppose, for instance, I tell you that I am making a square frame and that it is no more than 2ft wide, but this the only relevant information you have. (In particular, you have no information relevant to the question of what my purpose in making such a frame might be, or the question of what materials I have available.) What rational requirements are there on your credences regarding the size of this frame? One natural parametrisation of square-frame sizes is by side length: so, perhaps, by POI you should have equal credences in equal side-length intervals, and in particular you are rationally required to have credence ½ that the side length is between 0 and 1ft. However, another, apparently equally natural, parametrisation of square-frame sizes is by area: so, perhaps, by POI you should have equal credences in equal area-intervals, and in particular you are rationally required to have credence ¼

12 In particular: although it may initially be tempting to say that the partition into ‘all possible colours’ {red, blue, green, …} is natural and an appropriate candidate for apply-ing POI, this is clearly implausible on reflection: there is nothing especially natural or un-natural, for instance, either about a partition that identifies turquoise as a colour distinct from blue and green, or about one that declines to do so.

12


that the area is between 0 and 1 square feet (since the area could ‘equally easily’ be between 0 and 1 ft

2, between 1 and 2 ft2, between 2 and 3 ft2, or between 3 and 4 ft2). And these requirements are of course mutually inconsistent.

In the light of this problem, a widespread consensus (REFS) concludes that, despite any initial plausibility that it might naively seem to have had, the Principle of Indifference must simply be abandoned: that is, that there is no true constraint on rational credences even remotely like POI. (To find genuine rationality constraints on credences, these theorists often hold, we need to leave the domain of the a priori altogether: constraints, on this view, might well be found once information on frequencies and/or empiri-cally obtained knowledge of mechanisms and state spaces is available, but not in any purely a priori manner.) Thus, in rejecting POI, Lenman is far from alone.

Lenman does not say very much about precisely what form of ‘clue-lessness’ would result if his argument were accepted, or precisely why and for whom said cluelessness would be problematic. I will return to these issues in section 6. First, I will say why I think Lenman’s treatment of this ‘old’ problem of cluelessness is too pessimistic (section 4), and then present a type of case that (however) I do think raises a genuine threat of cluelessness (section 5).

iv. defence of (evf) against lenman’s objection The problem with a wholesale rejection of POI, and hence with Lenman’s argument, is that such wholesale rejection proves too much (its popular-ity notwithstanding). Quite independently of the issue of cluelessness, it seems clear that there are at least some cases in which something very like POI gives the right account. We gave one, everyday, example above (in which you know only that I am about to flip a coin, and that the two sides of the coin are labelled Heads and Tails).

Further: we noted above that POI-critics often recommend seeking knowledge of physical mechanisms in order to ground constraints on cre-dences; there might thus be a hope that many of the cases in which we are tempted to appeal to POI can instead be dealt with in this alterna-tive manner. It is clear, though, that (at a minimum) not all of the cases in which rational requirements seem intuitively compelling are ones that are even in principle amenable to this treatment. For some of the cases in which ‘indifference reasoning’ seems compelling concern self-locating propositions, and hence not matters to which considerations of objec-tive chance or physical mechanisms apply in any relevant sense. Consider,

draft paper

13


for example, a simplified version of the Sleeping Beauty problem. In this simplified version, there is no coin: you are (with certainty) to be awak-ened once on Monday and once on Tuesday, with qualitatively identical experiences each time, and with a selective-amnesiac drug administered on Monday night to ensure that your memories of Monday don’t break the conditions of qualitative identity. You find yourself awake and having the experiences in question, knowing that it is either Monday or Tuesday: what should be your credence that it is Monday? Clearly, this credence should be ½; but this cannot be because of any intelligible sense in which the objective chance that ‘now’=Monday is ½, for there is no such intel-ligible sense. So it seems that we must accept some form of POI at least for these self-locating cases; and once we have made that concession for the self-locating cases, a blanket refusal to countenance any form of POI in non-self-locating cases is unmotivated.

Nor is it only ‘everyday’ reasoning that seems committed to some form of POI. Statistical reasoning across the special sciences aims to deliver, as output, recommendations for levels of confidence that one is advised to put in specified hypotheses in the light of the evidence. But it can do so only given an assumption of ‘uninformative priors’; and an ‘uninforma-tive’ prior is precisely one that obeys (a given application of) POI. 13 As has long been recognised, a completely subjective Bayesianism, recognising no rationality constraints on priors aside from mere probabilistic coherence, cannot condemn as irrational any posterior credences whatsoever, regard-less of how intuitively overwhelming the evidence in favour of some given proposition might become.

According to this (I think correct) perspective, the fallout from this dis-cussion is as follows. For sure, the Problem of Multiple Partitions shows, as we conceded above, that the conjunction of an overly general Principle of Indifference with an insufficiently critical application of the notion of evidential symmetry leads to paradox. Suppose we concede for the sake of argument that the culprit is the fully general Principle of Indifference (i.e.,

13 Here, I am taking a stand on matters of deep controversy in the foundations of sta-tistics. ‘Frequentists’ do not agree that science is, or that it should be, in the business of recommending credences for hypotheses (as opposed to: recommending supposedly non-epistemic acts of ‘acceptance’ and ‘rejection’ of hypotheses), precisely because they are troubled by the element of subjectivity that seems involved in the adoption of any particular prior. Against the frequentists, I concur with the Bayesians, who note that until and unless one’s experiments result in e.g. some particular degree of belief (or range of degrees of belief) e.g. that the climate sensitivity is less than 3, or that one’s patient has disease X, those experiments will be useless for action-guidance. Yet action-guidance is one of the central aims of science even according to the most ascetic. This is not the place to revisit this debate in detail; for an accessible overview of the controversy between Bayesians and frequentists, see e.g. Sober, Evidence and evolution, ch. 1.

14


rather than the notion of evidential symmetry). Then this fully general Principle of Indifference is false. But since just about any principle, true or otherwise, is a special case of some natural generalisation that is false, this establishes little. It shows only that the true principles in this vicinity have to be restrictions of the Principle of Indifference, rather than that original principle itself.

We must further concede that we do not (yet?) know how to formu-late the appropriate restrictions, at least for many or most of the cases of interest.14 (We don’t know, for example, precisely what is the true restric-tion of the Principle of Indifference that will explain why one is ratio-nally required to have credence ½ that it’s Monday in the above modified Sleeping Beauty example.) This is an unfortunate situation, in theoretical terms. But (the examples show) it would be absurd to conclude from this that there are no true restrictions of the POI applying to the cases in ques-tion. The situation is one of impoverished understanding – we know what (some of) the rational requirements are in these cases, without having a full understanding of the principles generating those rational requirements – rather than one of rationality imposing no constraints. Situations of impoverished understanding should not in themselves surprise us: no-one thinks that the business of epistemology has been completed, whether or not it is completable.

There are, then, some ‘good cases’: cases in which some form of indif-ference reasoning generates rational constraints on credences, and we are in a position to recognise these cases as such and to know what the con-straints are, notwithstanding the fact that we do not (yet?) know precisely what form of indifference reasoning it is that does the generating.

It is equally clear – intuitively – that the case in hand is just such a ‘good case’. While there are countless possible causal stories about how helping an old lady across the road might lead to (for instance) the exis-tence of an additional murderous dictator in the 22nd century, any such story will have a precise counterpart, precisely as plausible as the original, according to which refraining from helping the old lady turns out to have the consequence in question; and it is intuitively clear that one ought to have equal credences in such precise-counterpart possible stories.15 That is:

14 For one attempt to formulate such a restriction, see Castell, ‘A consistent restriction of the principle of indifference’.

15 Note in particular that , unlike the case of assigning credences to elements of a parti-tion A={A1, A2, A3…}, there is no inconsistency in assigning equal credences both (i) to each of the counterfactuals A1 →B, A2 →B, A →B… and (ii) to each of the counterfactu-als A’1 →B, A’2 →B, A’3 →B…, even where the partition A’ is a selective coarse- or fine-graining of A. (In the cases of interest, in which the question of whether or not B is true in some intuitive sense ‘has nothing predictable to do with’ the question of which of the

draft paper

15


it is intuitively clear that because one has no more or less reason to expect one particular correlation between actions and possible-but-completely-unpredictable indirect effects rather than the reverse correlation, one is rationally required to have credences such that these completely unpre-dictable indirect effects make zero difference to the (differences between) the expected values of actions. And the failure (and paradoxical nature) of a completely general Principle of Indifference provides no grounds for doubting this intuitive verdict.

v. the new problem of cluelessness There are, however, cases that threaten cluelessness in a structurally very different way, and that fall outside the scope of any even remotely plau-sible form of POI. I will refer to the existence of these cases, and the prob-lem that they arguably pose for anyone who seeks to guide their actions even partially by considerations of goodness of consequences, as the ‘New Problem of Cluelessness’.

The cases in question have the following structure: For some pair of actions of interest A1, A2,

1. (NC1) We have some reasons to think that the unforeseeable consequences of A1 would be substantially better than those of A2;2. (NC2) We have some reasons to think that the unforeseeable consequences of A2 would be substantially better than those of A1;3. (NC3) It is unclear how to weigh up these reasons against one another.

The most vivid16 examples of this phenomenon occur in the context of ‘Effective Altruism’ (as outlined by e.g. MacAskill (Doing good better, 2015), Singer (The most good you can do, 2015). In this context, the

Ai is true, we will simply have C(A1 →B)=C(A2 →B)=C(A3 →B)=…=C(A’1 →B)=C(A’2 →B)=C(A’3 →B)=…=C(B), so that POI is satisfied both for the set {Ai →B}i=1,2,… and for the set {A’i →B}i=1,2,….) Therefore, the case of interest is in the relevant sense a ‘discrete’ case – a case in which we can reasonably hope to have a good intuitive handle on which sets of propositions to apply POI to – even if the space of possible actions is naturally continuous, so that (hard experience teaches) we have no reliable intuitive handle on which of the partitions A, A’ is itself more natural. (One will still end up committed to inconsistency if, for a given partition A, one insists on applying POI in the way suggested with respect to an arbitrary proposition B; but, unlike the case of the square factory, I am not aware of any ways of coming up with problematically natural routes to inconsistency along these lines.)

16 ‘EA’ is not, of course, the only source of examples with this structure. In fact, cases with the structure given in (NC1)-(NC3) are ubiquitous. I return to this point, and its significance, in section 7.

16


agent is considering devoting a significant portion of her resources, in terms of time and/or money, with the express purpose of causing as much good as possible for a fixed amount of input resource. Since the actions in question here involve at least moderate and optional sacrifice on the part of the agent, and since in addition the whole point of the actions under consideration would be to maximise good, any cluelessness about which actions have that property feels (intuitively) particularly galling – hence the special vividness.

Here is just one such example. Effective altruists place a lot of weight on the recommendations of independent charity evaluators, whose aim is to rank charities, as far as possible, in terms of overall cost-effectiveness: ‘amount of good done per dollar donated’. One charity that consistently comes out top in these rankings, at the time of writing, is the Against Ma-laria Foundation (AMF), a charity that distributes free insecticide-treated bednets in malarial regions. To justify this verdict, the charity evaluators clearly need (inter alia) estimates of the consequences of distributing bed-nets, per extra net distributed (and hence per dollar donated). Equally clearly, however, these charity evaluators, just like everyone else, cannot possibly include estimates of all the consequences of distributing bednets, from now until the end of time. In practice, their calculations are restricted to what are intuitively the ‘direct’ consequences of bednet-distribution: es-timates of the number and severity of cases of malaria that are averted by bednet-distribution (for which there is reasonably robust empirical data). In fact, GiveWell’s calculation focusses exclusively at the effectiveness of bednet-distribution in averting deaths from malaria of children under the age of 5, and (using data on remaining life expectancy at age 5 together with standard estimates of the ‘value of a healthy life year’) concludes that those benefits alone suffice for ranking AMF’s cost-effectiveness above that of most other charities.17 It is only if our condition (NR) holds when these effects alone are treated as the ‘foreseeable ones’ that the charity evaluators’ calculations can have the intended significance.

Averting the death of a child under 5, however, has knock-on effects that have not been included in this calculation. What the calculation counts is the estimated value to the child of getting to live for an additional (say) 60 years. But this life-extension also has systematic effects on others, which latter (1) have not been counted, (2) in aggregate may well be far

17 A similar, more complete but more complex calculation would also include the impact of bednet distribution on (a) malarial deaths in persons over the age of 5, (b) non-fatal cases of malaria, and (c) mosquito-borne illnesses other than malaria. Including these additional benefits would strengthen the case for donating to AMF as far as that case goes, but this consideration is orthogonal to the cluelessness concern that is the focus of the present paper.

draft paper

17


larger than the effect of prolonging the child’s life on the child himself, and (3) are of unknown net valence. The most obvious such effects proceed via considerations of population size.18 In the first instance, averting a child death directly increases the size of the population, for the following (say) 60 years, by one. Secondly, averting child deaths has longer-run effects on population size: both because the children in question will (statistically) themselves go on to have children, and because a reduction in the child mortality rate has systematic, although difficult to estimate, effects on the near-future fertility rate. Assuming for the sake of argument that the net effect of averting child deaths is to increase population size, the arguments concerning whether this is a positive, neutral or a negative thing are com-plex. But, callous as it may sound, the hypothesis that (overpopulation is a sufficiently real and serious problem that) the knock-on effects of averting child deaths are negative and larger in magnitude than the direct (positive) effects cannot be entirely discounted. Nor (on the other hand) can we be confident that this hypothesis is true. And, in contrast to the ‘old problem of cluelessness’, this is not for the bare reason that it is possible both that the hypothesis in question is true, and that it is false; rather, it is because there are complex and reasonable arguments on both sides, and it is radi-cally unclear how these arguments should in the end be weighed against one another.

To get a Principle of Indifference to be of any help here, we would have to regard conditions (i)-(iii) above – conditions under which there are competing reasons of quite different characters, and no obviously canoni-cal way of weighing those reasons against one another – as conditions of “evidential symmetry” for the purposes of POI. To be sure, at the level of description in the previous sentence, the evidential situation is ‘symmetric’ between the two propositions in question. However, in this case – un-like the ‘old problem cases’ – this appearance of symmetry disappears as soon as we probe to a deeper level. There is an obvious and natural sym-metry between the thoughts that (i) it’s possible that moving my hand to the left might disturb air molecules in a way that sets of a chain reaction leading to an additional hurricane in Bangladesh, which in turn renders

18 A different sort of concern that (however) would equally be grist to the ‘new clue-lessness’ mill has been pressed by Emily Clough (http://bostonreview.net/world/emily-clough-effective-altruism-ngos): that some effective-altruist-funded interventions might have large and negative longer-run consequences via their political effects. In particular, Clough worries that direct funding of front-line health services by outsiders might dimin-ish the tendency of governments of low-income countries to provide high-quality health-care services themselves (and of the citizens of the countries in question to demand such things from their governments).Yet more relevant (but complex) concerns relate to possible systematic effects of child mortality changes on the risk of premature human extinction; I lack the space to explore these here.

18


many people homeless, which in turn sparks a political uprising, which in turn leads to widespread and beneficial democratic reforms… and (ii) it’s possible that refraining from moving my hand to the left has all those effects. But there is no such natural symmetry between, for instance, the arguments for the claim that the world is overpopulated and those for the claim that it’s underpopulated, or between the arguments for and against the claim that the direct health benefits of effective altruists’ interventions in the end outweigh any disadvantages that accrue via diminished political activity on the part of citizens in recipient countries, or between the argu-ments for the claim that increasing the general wealth level increases exis-tential risk and those for the claim that wealth decreases existential risk. Furthermore, a principle mandating equal credences whenever arguments on some proposition are inconclusive seems completely implausible. Even if (as I have urged) there are some cases in which equal credences are man-dated whenever the agent has no relevant evidence, it is not at all plau-sible that equal credences are mandated on the grounds that the agent has conflicted and overall inconclusive evidence. It is surely at least sometimes reasonable, for instance, to react to such conflicted evidence by somewhat favouring P over not-P, on the basis of one’s admittedly tentative assess-ment of the available conflicting reasons.

Relatedly: unlike the ‘old problem of cluelessness’, which strikes many people as sophistical from the start (at least once a notion of subjective betterness is admitted), this ‘new problem of cluelessness’ feels real and important – at least to many of us, in some circumstances. Many who would otherwise be drawn to Effective Altruism nonetheless refrain from donating any significant portion of their earnings, not because of any posi-tive belief that refraining from donating will have better consequences19, but from a sense that they would require more confidence that their dona-tions really would be doing some significant amount of good – less clue-lessness – before they are willing to take the bold-feeling step of donating a significant proportion of their income. And, among those who do do-nate, many donate significantly less than they would if they had no such cluelessness-based worries; they commit partially to the EA ethos and in consequence ‘hedge their bets’, donating some significant amount (in case doing this really does do a lot of good), but far less than they might (in case their sacrifices are all just wasted, or, worse, actually harmful in the long run). And even those ‘hard effective altruists’ who have somehow overcome these worries for practical purposes generally still feel their pull.

19 Some people do refrain from donating for this other reason – “we shouldn’t fund child mortality reduction, because there are too many people anyway”. Those people are (or take themselves to be) in a simpler epistemic situation, and are not my focus here.

draft paper

19


vi. the nature of cluelessness I have argued that although the cases normally focussed on in the clueless-ness literature (‘old problem’ cases) generate no genuine threat of clue-lessness, nonetheless there does exist a different class of cases (the ‘new problem’ cases) that do generate such a genuine threat. In light of this, it is worth having a more precise account of exactly what our predicament is – both in epistemic and in practical terms – if and when we do face a situation of cluelessness.20 Why, in other words, is the situation not simply that each of us settles, by whatever means, on her own credence function for the relevant possibilities, and subjective c-betterness is a matter of ex-pected value with respect to whatever those credences happen to be? This is the task of the present section.

Here, I do not have a unique suggestion to defend. Instead, I will can-vass various possibilities, differing from one another in their accounts of what type of credal state is permissible in the cases in question, and/or of what the principles are linking those credal states to action. There is a size-able literature already on the choice among these rival accounts (REFS), focussing on issues quite aside from that of cluelessness. But the present discussion might throw one more relevant factor into the mix: the choice between these accounts, that is, might perhaps be made partially on the basis of which seems best to capture the phenomenology of cluelessness, insofar as that phenomenology is indeed real.

The first issue is: Which credal states are epistemically permissible for an agent facing a situation of ‘cluelessness’? There are two salient pos-sibilities:

- The agent is rationally required to have some particular (precise) credence function. But many credence functions are rationally permissible21.

- The agent is rationally required to have imprecise credences: that is, to be in a credal state that is represented by a many-membered set of probability functions, rather than a single probability func-tion.

These two possibilities lead to different possibilities for the nature of clue-lessness. In the first case, something recognisable as cluelessness might arise either (1) in connection with the arbitrariness that is necessarily in-

20 Those unconvinced by the arguments of section 4 can take this discussion to apply equally to the ‘old problem’; there has been surprisingly little said about the precise na-ture of cluelessness in the ‘old problem’ literature.

21 That is, the ‘uniqueness thesis’ (REFS) fails in cases of cluelessness.

20


volved in picking one rationally permissible credence function over anoth-er, or (2) due to ignorance of one’s own credences. In the second case, the nature of cluelessness will depend on what the correct principle is linking imprecise credences with the moral status of actions. I will consider how things work out given (3) a ‘liberal’, (4) a ‘conservative’ and (5) a ‘super-valuational’ approach.

I emphasised above (section 1) that the basic issues of cluelessness af-flict all plausible moral theories equally, not only consequentialist ones. In this section, however, several of the details that I seek to explore will depend on the precise relationship between ‘subjective c-betterness’ on the one hand, and the moral status of actions on the other. For this section only, therefore, I will specialise my discussion to the simplest candidate theory of that relationship, viz. maximising consequentialism. We are then working with the following criterion for moral rightness:

(SC: Subjective-consequentialist criterion of moral rightness) In circum-stances C, an agent is morally required to perform an act whose expected utility, with respect to the agent’s credence function, is at least as high as that of every other act that is available in C.

(1) Cluelessness as awareness of uncomfortable arbitrariness. In the first case, the agent may feel ‘clueless’ in the sense that while she is ratio-nally required to have some precise credence function, theory gives her no guidance about which function (within the permitted class) to have. Rationality therefore requires her to behave (epistemically) with some ar-bitrariness; and the imperfectly rational agent may find herself stuck (at least for a time) in a state that is somehow indeterminate between the various permissible credence functions, aware (perhaps) that she is ratio-nally required to settle on some particular such function, but experiencing paralysis over the question of which to settle on. The phenomenology of cluelessness, perhaps, corresponds to the agent’s painful awareness of the arbitrariness that would be involved in picking any one permissible credence function over the others, and the resulting epistemic and (hence) decision paralysis that an imperfectly rational agent may experience in this situation.

The precise source of the cluelessness here is a matter of some delicacy, for arbitrariness per se would be neither new nor problematic. It is no great news that sometimes two or more options are rationally permitted, in such a way that the choice between them can be resolved only with arbitrariness. According to expected utility theory even with uniquely settled precise credences and utilities, in particular, this happens in any situation in which two or more acts tie for first place in terms of expected utility. And while Buridan’s Ass succumbs to decision paralysis even in

draft paper

21


these cases, he is unusual: most of us are happy in this type of case with arbitrarily picking one joint-best action over the others.

There is, though, a difference between the case of Buridan’s Ass and that of the agent picking among rationally permissible credence functions. For the source of arbitrariness in the case of Buridan’s Ass is the fact that (expected utility theory tells us) the options facing the ass are exactly equally good. But theory does not tell us any such thing in the ‘clueless-ness’ cases that are our current focus. Rather, in general it tells us that in virtue of our (arbitrarily) picking this rather than that credence function, act A1 counts as subjectively better than act A2, so that we are (perhaps) morally required to perform A1, while if we had resolved the arbitrariness in a different way, the subjective-c-betterness ordering of acts and thus (given consequentialism) the moral requirements would have been differ-ent. Unlike the case of Buridan’s Ass, this therefore seems to be a case in which matters of importance hinge on our arbitrary decision, yet fail to do so in a way that removes the arbitrariness. This situation can perhaps more appropriately lead to an unsettling sense of cluelessness.

(2) Cluelessness via imperfect knowledge of one’s own credences.22 Suppose again that the agent is rationally required to have a precise cre-dence function, but that there is no particular credence function that she is rationally required to have. Even if she in fact conforms to this require-ment – so, there is some particular credence function that she does have – it need not follow that the agent knows (or has near-extremal credences, etc.) what her own credences are. And her failure to have any such self-knowledge is particularly plausible when, as here, rationality does not tell her precisely which credences she ought to have, so that even a high level of confidence that she conforms to the requirements of rationality would not tell her much about what her credences happen to be. The agent might then be clueless in the sense of not knowing (or having any near-extremal credences, etc.) which act-pairs satisfy the subjective c-betterness criterion (SB), and hence what the moral status of any given act is. Even the subjec-tive sense of c-betterness (and, in the consequentialist case, the subjective-consequentialist criterion of moral rightness) would then fail to be action-guiding, in one clear sense.

22 This general phenomenon of ‘luminosity failure’ can of course arise in other cases too – in cases in which there is a unique rationally permitted credence function, and in cases in which the agent is rationally required to have imprecise credences. (The agent might still be uncertain either about which credence function or imprecise credence-state is rationally required or about which she in fact has, and hence about what is morally required from her even in a subjective sense.) But it seems particularly likely to occur where Uniqueness conspicuously fails, since then, even if the agent knows (i) what the rational requirements on credal states are and (ii) that her own credal state meets those requirements, still she can fail to (be in a position to) know what her credal state is.

22


In the second case (i.e., that of imprecise credences), matters are consid-erably more complex. This is because while there is a canonical subjective-consequentialist criterion of moral rightness for cases of precise credence – expected value theory – it is non-obvious how this standard criterion is best extended to cover the case in which the agent’s “representor” (i.e. the set of probability functions that represents her credal state) has cardinal-ity greater than one. There are three particularly natural possibilities: 23,24

Liberal maximising-consequentialist theory of moral requirement under imprecise credences: Act A is morally permitted iff A maximises expected value with respect to some credence function in the agent’s representor.

Conservative maximising-consequentialist theory of moral requirement under imprecise credences: Act A is morally permitted iff A maximises expected value w.r.t. all credence functions in the agent’s representor.

Supervaluational maximising-consequentialist theory of moral require-ment under imprecise credences: Act A is determinately morally permit-ted iff A maximises expected value w.r.t. all credence functions in the agent’s representor. Act A is determinately morally forbidden iff A maxi-mises expected value w.r.t. no credence function in the agent’s represen-tor. Otherwise it is indeterminate whether or not A is morally permitted.

(3) Maximising-consequentialist cluelessness via a Liberal decision theory for imprecise credences. On the Liberal decision theory, the predic-ament of cluelessness again looks much like the (innocuous) predicament of Buridan’s Ass – more so than in case (1) above. For here, there are many rationally permitted options, nothing further to be said about which one to pick, and the agent simply has to pick one. And, while the theory does not positively say (as in the case of Buridan’s Ass) that the options in ques-tion are all equally as good, neither does it say here (as it did in the case of Uniqueness failure) that which acts are subjectively better than which oth-ers, and hence what the moral requirements are, depends on our arbitrary decision. This therefore seems to be a case in which the phenomenon of

23 Analogues of these theories are more often discussed in the context of the theory of rationality, in which the focus is on expected utility rather than expected value, ‘utility’ being simply a representation of ‘whatever the agent wants’. Here I employ the “sub-jective consequentialist’s“ standard trick of formulating precise structural analogues of the various moves that appear in the discussion of rationality, simply by substituting value for utility, and moral permission/requirement for rational permission/requirement, throughout the discussion. Moral theories other than maximising consequentialism will tend also to have versions corresponding to the liberal, conservative and supervalua-tional ideas in the context of imprecise credences, but the details (including both the plausibility of the corresponding principle linking imprecise credences to action, and the type of cluelessness that follows) can depend quite sensitively on the nature of the moral theory in question.

24 (ADD REFS for others that I won’t discuss: Maximin, Maximax, etc.)

draft paper

23


Cluelessness would be relatively shallow.

(4) On the Conservative theory, the clueless agent faces a moral di-lemma: if the probability functions in her representor mutually disagree about the moral permissibility of each available option, then all available options are impermissible.

It is perhaps worth noting, however, that the resulting ‘moral dilem-mas’ would be significantly different in character from those of a more familiar character. Moral dilemmas are normally thought to arise (if at all) in cases in which the agent faces a set of jointly exhaustive options, each option being in some significant way abhorrent. (This class includes ‘lesser evil’ cases; a typical example is Williams’ case of Jim and the In-dians (‘A critique of utilitarianism’).) Given a deontological theory25 that issues absolute prohibitions, for example, it could easily happen that every available option violates at least one of the theory’s prohibitions, and is therefore wrong according to the deontological theory. On this approach and given this kind of dilemma, it is easy to understand how moral dilem-mas could give rise, if not to cluelessness, then at least to deep discomfort: the agent is forced to make a choice, but (if morally conscientious) has a strong moral aversion to some particular feature of every available op-tion. In the present case, in contrast, there need not be anything abhor-rent about any of the options: it need not be, for instance, that any of the options involves killing, or letting disaster occur, or cruelty, or any such thing. It is only if the agent has a strong aversion to moral wrongness per se that there will necessarily be any such sense of abhorrence attending moral dilemmas of the kind under consideration.

And in any case – more pertinently for the purposes of our present discussion – there is no reason to think that moral dilemmas of either type should lead specifically to a sense of cluelessness, rather than to some other form of discomfort. Even those deontologists who are fans of mor-al dilemma in lesser-evil cases generally acknowledge that in some sense the appropriate thing to do is to choose the lesser evil, rather than to be paralysed by the observation that all available acts are wrong. And in the absence of any notable imbalance among the options in terms of good-ness, amount of ‘evil’ and so forth, a perfectly acceptable way to respond to a situation in which all options are impermissible and that is all that can be said is simply to pick arbitrarily. For this too would be a case in

25 Moral dilemmas are normally thought not to occur on a consequentialist approach; indeed, depending on their intuitions as to the plausibility of moral dilemmas in general, many theorists take this to be either a significant advantage or a significant disadvantage of consequentialism. It is therefore worth noting well that given imprecise credences and a Conservative decision theory for them, this link between moral dilemmas and conse-quentialism would fail.

24


which theory positively tells us that there is nothing to choose, morally, among the options in question. The case of imprecise credences together with a Conservative maximising-consequentialist theory of moral right-ness, therefore, like that of the Liberal theory, also seems to furnish only a very shallow sense of cluelessness, little deeper (if at all) than that facing Buridan’s Ass.

(5) On the Supervaluational theory, for any given option in a situation of cluelessness, it is indeterminate whether or not that option is morally permitted. This verdict of theory seems to offer more promise for captur-ing the intuitive sense of cluelessness: the agent seeks to choose his actions in response to the permissibility facts, but his actions must be determinate, while those permissibility facts remain stubbornly indeterminate. Nor is it obvious that we can say here, as we did in the case of the Liberal and Conservative decision principles: “But theory tells us that all actions have the same moral status (viz., here, that of indeterminate permissibility), so we are free just to choose among them arbitrarily”. For the relevant candidates for moral status arguably do not include ‘indeterminate per-missibility’: rather, they include only the first-order evaluations permitted, required, forbidden. And in these first-order terms, the Supervaluationist theory does not tell us that the available actions all have the same moral status: depending on the details of the case, it either (1) tells us either that it’s indeterminate whether they do or not, or (2) tells us that (it’s determi-nately true that) the available options do not have the same moral status, but that it’s indeterminate which particular options are required/permit-ted/forbidden.

In sum: Imprecise-credence responses to conflicting reasons, together with either a Liberal or a Conservative decision theory for imprecise cre-dences, seem to deliver at most a relatively shallow sense of cluelessness, at least for agents who know that they are approximately rational. Those who feel that the problem of cluelessness is a real and deep one may there-fore feel that either a theory according to which many credence func-tions are rationally permitted, or a theory according to which imprecise credences are mandated together with a corresponding ‘Supervaluational’ theory linking those imprecise credences to action, does the most justice to their experienced phenomenology. Those decision theories do not force any deep phenomenon of cluelessness, but they make more room for one.

vii. mundane cluelessness Effective Altruism makes the new problem of cluelessness particularly viv-id, and is therefore a good context (for us) to focus on in considering ‘new cluelessness’. Clearly, though, insofar as what is distinctive of those cases is

draft paper

25


(as I have suggested) the satisfaction of conditions (NC1)-(NC3) in section 5, the basic phenomenon is far from specific to that context: cases with the structure in question also occur in myriad other decision contexts, at both large and small scales. For example: (1) A government’s decision-making predicament for any large-scale policy decision – for instance, concerning whether or not to go to war, or whether or not to raise taxes to finance additional spending on education or healthcare. (2) An individual’s deci-sion as to which degree course to sign up for, which job to accept, whether or not to have children, how much to spend on clothes, whether or not to give up caffeine.26 In these cases, no less than the effective-altruist ex-amples discussed above, (a) there are good consequence-based reasons/arguments for favouring each of two alternative actions and also (b) there is no obviously canonical way of weighing up those reasons or arguments against one another.

It follows that insofar as the source of cluelessness is the satisfaction of conditions (NC1)-(NC3), one should feel clueless in these everyday cases no less than in the effective-altruist cases.

To what extent do we feel clueless in everyday cases? Some people, for sure, suffer from decision paralysis every time the arguments for and against rival possible actions are inconclusive. But this sort of ubiquitous decision paralysis is a pathology, bordering on a mental illness, rather than the norm. Most of us simply learn to live with the need to resolve decisions with some arbitrariness and with incomplete guidance from data and theory, and, while we may feel more secure in non-arbitrary cases, neither are we especially bothered by the need for arbitrariness, or “sen-sible judgment”, when that need does arise. This suggests that either an excessive deference to the sense of arbitrariness in the effective-altruist cases is also a pathology, or conditions (NC1)-(NC3) do not after all strike to the root of the phenomenology of cluelessness in effective-altruist cases.

viii. conclusions Let A1, A2 be available actions, and let V(A1), V(A2) the overall goodness of the worlds that would ensue if I performed acts A1, A2 respectively. The ‘old problem of cluelessness’ was supposed to arise merely from the

26 Of course, in private decision-making in particular, the ‘theory of the good’ that the decision-maker seeks to employ for the evaluation of consequences is unlikely to be an impartial one. But considerations of cluelessness look structurally just the same in the domain of rationality as they do in the theory of morality: if, for instance, the private individual cares only about his own family, then the relevant value function for that con-text is one concerning well-being of his family alone, but the remainder of the discussion is largely unaffected.

26


likelihood that the largest contribution to the objective value-difference V(A1)-V(A2) is due to ‘unforeseeable’ effects of these actions, while (how-ever) that contribution is of unknown sign. I have argued (contra Lenman) that while that is indeed very likely, it poses no problem for a ‘subjective’ criterion of c-betterness, or (in the consequentialist case) for a criterion of subjective moral rightness, framed in terms of the expectation values of V(A1) and V(A2). That was because while the ‘indirect’ effects almost certainly dominate the objective value-difference V(A1)-V(A2), they make zero contribution to the expected value-difference E[V(A1)-V(A2)]. To be sure, the case for that last claim relies on some restriction of the Principle of Indifference. But, I have argued, despite the fact that a fully general Principle of Indifference is paradoxical, it is overwhelmingly plausible that some suitable restricted Principle of Indifference is true. The old problem of cluelessness is no problem, for consequentialists or for anyone else.

Matters look somewhat different, however, in a different type of case. In ‘old problem’ cases, the unforeseeable effects under consideration were ones that, while they could result from (say) some particular act A1, they could equally easily, and in precisely analogous ways, result from any of the relevant alternative acts. It is this precise analogy between the possibil-ity that (say) choosing A1 over A2 would lead to effect E1 rather than E2, and the ‘opposite’ possibility that choosing A1 over A2 would lead to E2 rather than E1, that renders plausible the indifference reasoning that is so intuitive in those cases. In contrast, in ‘new problem’ cases (I stipulated), one has more specific reasons for suspecting particular correlations be-tween acts and ‘indirect’ effects, but too many such reasons: non-isomor-phic reasons that point in different directions, and for which there is no canonical weighing-up operation. In those cases, no form of indifference principle is at all plausible, and the threat of cluelessness is more genuine.

It is not at all obvious on reflection, however, what the phenomenon of cluelessness really amounts to. Here, I have explored various possibilities, the details depending on whether the correct epistemology for the cases in question requires precise credences but permits a wide range of such credences, or instead requires imprecise credences; and, in the latter case, what the normative connections are between imprecise credal states and actions. The routes that seem to offer more scope for capturing a deep sense of cluelessness are (i) a requirement for precise credences but with wide permissions regarding which precise credences, and (ii) a requirement for imprecise credences together with (in the maximising-consequentialist case) a Supervaluational principle linking credal states to moral require-ments.

It is also not at all obvious, however, how deep or important the phe-

draft paper

27


nomenon of cluelessness really is. In the context of effective altruism, it strikes many as compelling and as deeply problematic. However, mun-dane, everyday cases that have a similar structure in all respects I have considered are also ubiquitous, and few regard any resulting sense of clue-lessness as deeply problematic in the latter cases. It may therefore be that the diagnosis of would-be effective altruists’ sense of cluelessness, in terms of psychology and/or the theory of rationality, lies quite elsewhere.

president: Susan James (Birkbeck)

president-elect: Tim Crane (Cambridge)

honorary director: Rory Madden (UCL)

editor: Matthew Soteriou (Warwick)

lines of thought series editor: Scott Sturgeon (Oxford)

executive committee: Corine Besson (Sussex) / Kimberley Brownlee (Warwick) /Rowan Cruft (Stirling) / Sacha Golob (KCL) / James Harris (St. Andrews) /

Adrian Moore (Oxford) / Samir Okasha (Bristol) / Ulrike Heuer (Leeds)

managing editor: Hannah Carnegy (UCL)

assistant editor: David Harris

designer: Mark Cortes Favis

administrator: Josephine Salverda (UCL)

t h e a r i s t o t e l i a n s o c i e t y

w w w. a r i s t o t e l i a n s o c i e t y . o r g . u k

Date post:	01-Aug-2016
Category:	Documents
Upload:	the-aristotelian-society
View:	224 times
Download:	2 times

Hilary Greaves (Oxford): Cluelessness

Documents