This competition entry has been selected for publication by the Forum team.
There is a move buried inside the unawareness argument that deserves more scrutiny than it has so far received, because it is doing more work than its brief appearance suggests. It occurs at the transition from P2a to P2b: having established that impartial altruists should not assign literal precise expected values to actions, DiGiovanni goes on to argue that they should not even force themselves toward a "best guess" comparison when their understanding of consequences is sufficiently coarse-grained. The first claim is, as he himself notes, the less contentious one; a large academic literature on imprecise credence already supplies the vocabulary and the arguments. The second claim is the one on which the entire practical force of the unawareness argument rests, because it is the one that forecloses the ordinary response of a person who, lacking a number, nonetheless has to act. This essay argues that P2b is false, or at least radically under-motivated, because it silently assumes that any justified comparison in two coarse-grained options must proceed by aggregation — by summing, weighing, or otherwise combining the possible consequences into something that behaves enough like an expected value to be blocked by the same argument that blocks precise EV. Once we notice that comparative judgments can instead proceed by asymmetric reasons, without at any point purporting to aggregate a comprehensive set of consequences into a single verdict, the coarse-grainedness that motivates P3 stops being the kind of thing that generates incomparability.
Begin with what "best guess" actually means in ordinary deliberation, outside the aggregative frame the argument imports. When a person judges that donating to a well-evidenced anti-malaria charity is better, in expectation, than donating to an untested and unevaluated one, she is not modelling the full cosmic consequence tree of either action and comparing sums. She could not do this even in principle, and DiGiovanni is right that any attempt to do so would run into exactly the arbitrariness he describes: the catch-all category of unconsidered effects would swamp the comparison, and there would be no non-arbitrary way to assign it a value. But this is not, in fact, how the judgment is reached. It is reached by noticing a reason that favors one option and asking whether anything favors the other option strongly enough to defeat it. The reason is local, not comprehensive: there exists good evidence that the intervention reduces child mortality, and the objector has not supplied a comparably weighty reason on the other side. This is what DiGiovanni's own gestures toward when it names "reasons-based choice" as the most promising framework left unexplored, but the sequence does not go on to ask whether such a framework survives P3, and I think the answer is that it does, precisely because it was never in the business of aggregation to begin with.
The distinction matters because the empirical premise, P3, is pitched entirely against aggregative comparison. Its evidence is that our track record of discovering sign-flipping considerations is poor, that we are likely unaware of further considerations we have not yet conceived, and that any attempt to sum up the mechanisms we can model against the mechanisms we cannot will be dominated by an unmodellable remainder. All of this is true, and none of it touches a reasons-based comparison, because a reasons-based comparison claims only that a specific, identified consideration favors A over B, and that no consideration of comparable burden has been identified on the other side. The demand that we also account for everything we have not yet identified is a different and much stronger demand — one that, notice, would equally undermine our confidence that the sun will rise tomorrow, since we cannot rule out unawareness of some catastrophic astrophysical mechanism that would prevent it. We do not, in fact, hold physical predictions to this standard, and the reason we do not is that a reasons-based judgment about tomorrow's sunrise does not purport to have surveyed the full space of possible defeaters; it rests on an identified mechanism (gravitation, stellar physics) that has not yet been defeated by any specific consideration, however open we remain to being wrong. Impartial altruistic judgments about, say, the sign of extinction risk reduction, can have the same structure without collapsing into either precise EV or naïve heuristic-thumping.
DiGiovanni anticipates something like this response in his treatment of the Ngo-style objection that ordinary, small-scale decisions face the same qualitative unawareness problem and yet are not thereby rendered incomparable. His answer is that there is a principled threshold, crossed when a comparison becomes sensitive to arbitrary choices about how to fill in a coarse-grained model, and that impartial altruistic comparisons of cosmic consequences are past this threshold in a way that everyday comparisons are not, because we lack both deep mechanistic understanding and a track record of consistent success at that scale. This is a fair distinction to draw against an aggregative framework, where sensitivity to arbitrary parameter choices really is fatal, because the aggregate is only as trustworthy as its least-constrained input. But a reasons-based comparison is built out of identified considerations that are either present or absent, weighed against each other directly. The threshold DiGiovanni describes governs when an aggregative model becomes untrustworthy. It does not obviously govern when a specific, articulable reason becomes too weak to carry a comparative judgment, because those are different failure modes with different remedies. A reason can be weak, contested, or defeated by a countervailing reason; it cannot be "arbitrary" in the sense that afflicts a coarse-grained parameter, because it was never a parameter to begin with.
It might be objected that this simply relocates the problem: if reasons-based comparison does not aggregate, then on what basis does one reason outweigh another, and is not that weighing itself just aggregation by another name, subject to the same coarse-grainedness? I do not think so, and the disanalogy is worth stating precisely, because it is where the substantive philosophical work has to happen. Aggregative EV comparison requires that every possible consequence, known and unknown, be assigned some value and combined into a total; the catch-all category is its central object, since by construction it contains most of what would determine the sum. Reasons-based weighing requires only that the specific reasons actually on the table be compared to each other, and it is a further, substantive question — not settled by P3 — whether the mere existence of unconceived considerations defeats an identified reason's weight, or whether it merely means our confidence in the comparison should be less than certain. DiGiovanni's own response to the "cancels out" objection (4.1.1) shows why symmetric cancellation cannot be assumed; but the reasons-based framework was silent about the unconceived remainder, in the same way that a physicist's confidence about tomorrow's sunrise is silent about unconceived astrophysical defeaters, without thereby being unjustified.
None of this amounts to a decisive victory over the unawareness argument, and I want to resist the temptation, which the sequence itself warns against, to claim more than the argument delivers. It does not show that impartial altruists can, after all, assign confident sign and magnitude to long-run interventions; a reasons-based comparison between, say, two different approaches to AI safety work may turn out to have no clear winner, because the identified considerations may be genuinely balanced or because no considerations of adequate specificity have yet been identified for either side. What it shows, if it succeeds, is narrower and more useful: that the absence of a trustworthy aggregate is not the same thing as the absence of any justified comparison, and that P2b's inference from "coarse-grained" to "incomparable" smuggles in an aggregative picture of justification that a reasons-based impartial altruist was never committed to holding. The unawareness argument would then retain its force against the most ambitious and the most naive forms of expected-value reasoning about the far future, which is already a significant result, while leaving open a narrower, more modest, and more defensible route by which impartial altruists can say, of some pairs of actions, that one is better than the other — not because they have weighed the whole of the cosmos, but because they have found and not yet lost an argument.
