Maximality Is the Wrong Decision Rule for a Vague Representor

This competition entry has been selected for publication by the Forum team.

Epistemic status: Challenging the inference from "our beliefs should be indeterminate" to "therefore compare actions via maximality" not challenging indeterminacy itself. I think this is correct and I've tried hard to find the reply that dissolves it. I've included the closest one I could find and explained why I don't think it works.

The inference in question

DiGiovanni's case for indeterminate beliefs is that a representative set of probability distributions is less arbitrary than a single precise credence. Fair enough. But he then adopts maximality as the decision rule: action A is preferred to B only if A has higher expected value than B under every distribution in the representor. This is presented as the natural, minimal extension of expected-value reasoning to indeterminate beliefs. I want to challenge that specific step not the move to indeterminacy, but the move from indeterminacy to this rule.

The concession that undoes it

Elsewhere in the same post, DiGiovanni concedes that the boundary of any representor is vague. He asks, of a credence he's written as [0.45, 0.65]: why not [0.42, 0.66], or [0.5, 0.7]? There's no privileged answer the exact edge of the set is exactly the kind of thing his own framework says we shouldn't decisively classify. He treats this as a tolerable cost: representors are still less arbitrary than points, even if their edges are fuzzy.

That's a reasonable response to the vagueness by itself. But it doesn't consider what maximality does with that vagueness. Maximality doesn't average over the representor, and it doesn't require dominance over most of it. It requires dominance over literally all of it including whatever distributions happen to fall inside the fuzzy boundary by whatever convention you used to draw it. That makes maximality maximally sensitive, in a discontinuous way, to exactly the part of the model DiGiovanni has already admitted is arbitrary.

One boundary-hugging distribution, whose membership in the representor is not a fact of the matter by his own lights, gets full veto power over an otherwise unanimous consensus among every other distribution in the set.

A concrete case

Suppose actions A and B are being compared. Every distribution in the "core" of a reasonable representor, the well-motivated bulk of it ranks A above B, often by a wide margin. But there's one distribution near the fuzzy edge, included only because some equally-unprivileged boundary convention happens to sweep it in, under which B narrowly beats A. Maximality declares A and B incomparable.

The verdict "we have no action guidance here" now rests entirely on whether that one boundary distribution counts as "in" the representor, a classification DiGiovanni's own vagueness argument says we can't determinately make. So the rule doesn't inherit a reduced, proportionate amount of arbitrariness from the vague boundary, the way a fair implementation of "less arbitrary than a point estimate" should. It inherits total arbitrariness at the level of the final verdict, because a single indeterminately-classified case is sufficient to flip the outcome from "guided" to "unguided." That's a worse arbitrariness problem than the one representors were introduced to solve, not a better one.

The fix

Replace full dominance with dominance over the representor's core call it trimmed maximality: A is preferred to B if A has higher expected value under every distribution in the representor except a small, symmetric margin at the boundary. This isn't ad hoc backpedaling toward precision. It's the natural response to vagueness itself: if you can't determinately say whether a boundary-case distribution belongs in your representor, you shouldn't let its inclusion or exclusion single-handedly determine whether you have any action guidance at all.

A trimmed rule treats boundary indeterminacy the way DiGiovanni's own framework treats indeterminacy everywhere else by declining to let it force a determinate answer in either direction rather than the way maximality treats it, which is to let it force the most extreme possible answer (total incomparability) whenever it appears on the "disagree" side.

Anticipated objections

"Trimming just introduces its own arbitrary parameter, how much do you trim?" True, and I don't think there's a fully principled answer, any more than there's a fully principled boundary for the representor itself. But this replaces one large, unmotivated arbitrary choice (trim nothing, i.e., maximality) with a smaller, localized one (trim some unspecified small margin). "Trim nothing" isn't a neutral default, it's a specific, substantive commitment to giving the least-plausible members of the representor exactly as much veto power as the most-plausible ones, and DiGiovanni gives no independent argument for that being the right amount of trimming.

"This is just the 'maximality is too permissive' objection again, and you've already conceded his reply works for that." No - his reply to Mogensen's version is that low action-guidance is a feature of impartial consequentialism's ambitious scope, not a flaw in the rule.

That defense is about how many cases end up incomparable.

My objection is about why a specific case ends up incomparable to a single boundary intruder overriding a near-unanimous core, which is a structural property of the rule itself, not a scope complaint. It shows up even in narrow, mundane-stakes comparisons, including ones DiGiovanni cites as maximality's success cases (like preferring the sidewalk to the road), if you deliberately construct a representor with one boundary-hugging outlier. The rule doesn't distinguish "principled disagreement across a real range of reasonable views" from "one barely-admissible edge case" and it should.

"Isn't some version of this already addressed by his response to the aggregation objection?" Not quite that section defends not averaging the representor into one distribution, on grounds that we'd lose information about our inability to pin down weights.

Trimmed maximality doesn't average or add higher-order weights; it only declines to let boundary-vagueness convert into full veto power. It's a weaker, more targeted move than the aggregation proposals he already rejects, and I don't see it addressed anywhere in the post.

"Where does this leave cluelessness about the far future?" Narrower than it might first look: I'm not claiming trimmed maximality resolves severe cluelessness or restores confident long-termist verdicts. Many far-future comparisons will likely still come out indeterminate even under a trimmed rule, since the disagreement in those cases runs through the core of the representor, not just its edges.

The claim is narrower and, I think, more defensible: maximality is currently doing more work than the argument for indeterminate beliefs actually earns it, because it converts residual boundary-vagueness into disproportionate loss of action-guidance. Fixing that won't rescue longtermism, but it should change which specific comparisons we're entitled to call indeterminate.