Recently in this journal, Christopher Jay posed a new challenge to one form of subjective consequentialism.Footnote ¹ It is well known that unforeseeable consequences of some of our actions pose a difficulty for making even subjective assessments of their likely results, but Hilary Greaves has recently proposed a principle of indifference to unforeseeable results, making the deontic status of our actions a function only of foreseeable ones.Footnote ² Jay argues that depending on how we interpret Greaves’ approach, it entails either ‘deontic indeterminacy’ in which agent A's proposed action φ has no determinate deontic status prior to A's investigation of the results of φ-ing, or ‘arbitrary deontic variance’, in which A's non-moral choices of how to investigate φ change its permissibility. Jay considers both possibilities unacceptable, but I will argue that the latter is actually completely innocuous.

Jay first treats ‘foreseeable’ results as the potentially foreseeable (which he also calls ‘ultimately foreseeable’) results of φ, that is, what we may not foresee now, or even be able to foresee now with our current evidence, but which we could foresee if we took some other investigatory actions to make available some additional evidence prior to the choice to φ.Footnote ³ Jay then raises the possibility that two (or more) of the actual results of φ could be such that either result could become foreseeable if A takes some investigatory action directed thereto, but are not jointly foreseeable (perhaps A lacks time or resources to investigate both possibilities). This makes it unclear which results of φ we should count as ultimately foreseeable prior to A's investigation. This is the ‘deontic indeterminacy’ concern.

Jay grants that this could be avoided if we define the foreseeable as what is or becomes foreseeable given A's actual investigatory actions up to any given point in time. But this interpretation leads directly to Jay's second concern, for the permissibility of φ-ing may then depend upon A's arbitrary, non-moral choices about whether or how to conduct this investigation. This is the ‘arbitrary deontic variance’ concern.Footnote ⁴ Jay objects to the possibility that such choices made for ‘prudential, pragmatic, aesthetic, or other non-moral reasons’ could change the deontic status of φ, suggesting that such a result is ‘wildly implausible’ and claiming that ‘surely no subjective consequentialist would welcome [this] news’, especially when the ‘investigation’ is merely a choice of which results to think more carefully about or draw to our attention.Footnote ⁵

Here I am very confident that Jay is wrong; this subjective consequentialist does not find the result at all implausible, and welcomes this news openly. If we truly have limited epistemic and cognitive resources for investigating various morally significant results of our actions, and some moral reason to do some further investigation, but no moral reason to prefer one investigation over another, then we should arbitrarily pick some partial investigation forthwith in hopes of marginally improving our epistemic situation, even if we know or suspect that this is radically incomplete, and that a different choice may well have delivered different deontic results. And this is still true even if the ‘investigation’ is merely an arbitrary choice of which possible results to think more carefully or vividly about, making them marginally more foreseeable when we cannot in the same way make all other possible results equally foreseeable.

A specific example may support this claim. Suppose a group of hikers on a mountain ridge has an injured member, so it is morally important for them to return to lower altitude soon. They can descend by either of two gullies, G₁ and G₂, neither of which they were previously familiar with, so they have no reason to think that either is safer or more direct than the other. Each gully has a hill overlooking it, H₁ and H₂ respectively. While most of the group rests in a short preparation for the final descent, their most energetic member has time to scout ahead to exactly one such hill. Given his lack of any moral reason to climb one hill versus another, he arbitrarily climbs H₁ because it has a granite outcropping, and he admires granite, although this has absolutely no bearing on the safety of G₁, or the likelihood of getting a good view of it. And suppose that this reveals that G₁ is a smooth, direct route. Perhaps G₂ is as well, but the hikers don't know that, and so now arguably have an obligation to descend via G₁ rather than G₂. This is true even though, had the scout detested granite, he might have climbed H₂ and instead acquired an obligation to descend via G₂. I see nothing objectionable in this result.

Nor is it problematic if the evidence is mixed and must be resolved by attentional focus. Suppose the scout also sees a black bear in G₁. This complicates but does not decisively govern the choice. The scout knows that black bears are usually a danger only to small groups of careless hikers; his group is neither, and has experience making noise to warn bears of their approach. The bear appears to be solitary, yet there is a slim chance that it is a mother with unseen cubs, and if so, a slimmer chance that she could misperceive even loud hikers as a threat thereto. The bear might be injured and moving slowly, yet appeared at the moment to be descending the gully at a normal pace. And the hikers won't be traveling that fast with their injured member, so even a slow bear should have time to get out of the way. And so forth – there might be dozens of factors with different valences which the scout could now reflect upon. Perhaps even unlimited time for consideration would not allow the hikers to determine with great confidence which route was safest. But their time is limited – the scout returns to the group, and they need to descend now to get their injured member to shelter before the sun sets. They allow the scout to decide the route, as he is most vividly aware of the relevant evidence. It seems to me that this is a case where, like Sartre's student wondering whether to join the Free French overseas or stay and care for his mother, the pre-reflective weights of the options are at least prima facie roughly balanced, and do not clearly favor one choice over another.Footnote ⁶

So the scout turns his attention to just one fact: bears have sometimes injured people. As a result, they descend via G₂. Had he instead thought more about how easily bears can be warned away from humans, he might have gone down G₁. By hypothesis, the scout chose which factor to reflect upon further, out of the many available, essentially at random, or at any rate for reasons unrelated to the likelihood of its moral importance (for if he had evidence of any such reasons, he should guide his reflective act on them instead of choosing randomly). Then via this procedure he at least picks out some relevant factor, when he cannot reflect upon all of them, and has thereby some chance of marginally increasing the safety of their journey. We can even show the value of this strategy mathematically. Suppose that there are 10 factors relevant to the safety of the choice, and the agent has no prior reason to think any of them weightier than the other, or that more factors count for or against φ-ing, but in fact seven of them would count for and three against φ-ing, with equal weight. A random selection of one of them then gives a 70% chance of making the best choice, an improvement over the 50% odds for guessing at random without such reflection. If they are not equally weighty, then this strategy could lead us astray in a given case, for example, perhaps three factors each count strongly against φ-ing while each of the other seven count only weakly in favor of it. But this strategy will still improve our chance of success across a series of similar choices where the factors’ weights vary randomly across their valences. In any case, it is better than nothing, and might very well be the best approach in cases of uncertainty.

The strategy is structurally very much like choosing one hill at random to overlook a gully if ten hills and gullies are available. Or like picking one ball out of a bin of ten, finding that it is either red or not red, and hence making a very slightly improved guess about whether the bin contains more red than non-red balls if your prior probabilities for each are evenly split. Non-moral choices of what known evidence to reflect upon should now seem less different than they otherwise might from non-moral choices of what external evidence to gather, as both ultimately simply change (and can improve) our rational beliefs about the probabilities that various choices will satisfy certain ends. In both cases, for a subjective consequentialism based on practically foreseeable results this strategy is obligatory precisely because it would change our obligations to the making of choices more likely to satisfy our moral goals. Which is again unobjectionable.

Unfortunately, as Jay did not give any concrete example of a case in which ‘arbitrary deontic variance’ produces implausible results, I cannot be certain that I either fully understand his objection or that my example decisively refutes the same. But I do think it is a prima facie rebuttal which at least throws the ball back into his court. To restore the objection, he would have to do more than simply say without argument or examples that such variance is ‘wildly implausible’, but needs instead to show specifically what is wrong with my example. For example, perhaps it does not actually illustrate the kind of variance he has in mind, which other cases would show better, or perhaps he can argue that the deontic variance therein remains problematic despite what I have said about it. Until then, I currently have, and have shared with the reader, some subjective reasons for thinking that his argument is mistaken.