This competition entry has been selected for publication by the Forum team.
There is something deeply unsettling about Anthony DiGiovanni’s unawareness argument. It arrives with the force of a philosophical reductio: if you care about all consequences, impartially, across all of space and time, then you cannot justify preferring any action over any other. Not “you might be wrong about which action is best.” Not “you should update your credences.” No justification whatsoever. The impartial altruist, confronted with any choice, has no reason to pick one option over another.
This conclusion, if true, would be devastating. It would mean that the entire project of using evidence and reasoning to do the most good is built on sand. And DiGiovanni is not making a casual claim; he has built a careful, multi-part argument with three distinct premises. To defeat it, you have to show where the chain breaks.
I will argue that the unawareness argument fails at its very first step. The normative premise sets an unreasonably high bar for what counts as a justified impartial-altruistic preference. It demands that we be able to argue that an action has higher “expected value” in some idealized sense, even when we lack the epistemic resources to do so. But this is not a requirement of rationality; it is a philosophical overreach that conflates ideal justification with actual justification. Once we see this, the rest of the argument collapses. Even if we accept P2 and P3,it does not follow that we have no impartial-altruistic reason to prefer anything. We can have good, non-arbitrary reasons to act without having idealized reasons.
Let’s get precise about what P1 actually claims.
DiGiovanni defines a “c-preference” as a preference that is justified by an impartial altruistic comparison of actions’ possible consequences. P1 then says:
To justify c-preferring A over B, it’s not enough to say (e.g.) that A seems heuristically good. Rather, we need to argue that A has higher “expected value” broadly speaking, meaning: In some sense we “expect” that, if we were idealized agents who could aggregate all of A’s and B’s possible consequences into literal EVs, then we’d say A has higher EV.
Notice what this premise is doing. It is not merely saying that expected value is relevant to justification. It is saying that the only way to justify an impartial-altruistic preference is to have a reason to think that an idealized version of ourselves, one with complete information and perfect aggregation, would assign higher expected value to A than to B. If we cannot say that, then our preference is “unacceptably arbitrary”.
This is a strikingly strong claim. It is also, I will argue, false.
The first thing to notice about P1 is how much weight it places on the idealized agent. DiGiovanni is not asking us to actually compute expected values; he acknowledges we can’t do that. He is asking us to have some reason to think that if we could compute them, A would come out ahead.
But this creates a bizarre epistemic situation. We are being asked to have beliefs about what an idealized version of ourselves would think, under conditions that are radically different from our own. The idealized self has access to all consequences, can assign precise values, and can aggregate across all moral patients. We have none of these things. Why should our beliefs about what this being would think track anything reliable?
Consider an analogy. Suppose I am bad at chess, I can barely see two moves ahead. You tell me that I cannot justify preferring one move over another unless I have reason to think that a grandmaster, with perfect calculation, would prefer that move. But I have no such reason! I don't know what the grandmaster would see. Does that mean my preference is arbitrary? Surely not. I can have perfectly good, non-arbitrary reasons to prefer one move; maybe it defends my king, or develops a piece. These reasons are real reasons, even if they fall short of what an idealized agent would require.
The same applies to impartial altruism. We can have good reasons to prefer one action over another, reasons based on the evidence we actually have, the values we actually hold, and the comparisons we can actually make, without having any clue what an idealized version of ourselves would think. The fact that our reasons are incomplete does not make them arbitrary.
DiGiovanni anticipates something like this. His response is that without the idealized standard, we have no way to avoid arbitrariness. If we just pick actions based on what “seems” good, we are choosing arbitrarily.
But this gets things backwards. The arbitrary choice is not the one made on the basis of the best evidence available. The arbitrary choice is the one made on the basis of a standard we know we cannot meet, and then pretending that this somehow licenses paralysis.
Think about what P1 actually requires in practice. For any action A and B, we are supposed to ask: “Do I have reason to think that an idealized version of myself would say A has higher EV?” For almost any pair of actions with significant long-term consequences, the answer will be “no.” We simply don't know enough about the far future to have any confidence about what a being with complete information would conclude. So P1 tells us that almost all preferences are unjustified.
But here’s the rub: the choice not to prefer is itself a preference. If I decide to do nothing, or to flip a coin, or to do whatever I feel like, I am still choosing. And P1 would say that this choice is also unjustified, because I have no reason to think the idealized self would prefer inaction over action. So P1 doesn't just tell us we can't justify our preferences; it tells us we can't justify any course of action, including the course of not acting. The argument is self-undermining: if P1 is true, then we have no justified reason to accept P1, because accepting P1 is itself an action that we cannot justify under P1’s own standards.
What we need is a different account of what makes a preference justified. Here is a sketch.
A preference is justified, in the relevant sense, if it is based on the best available evidence, reasoned through carefully, and not defeated by known counterarguments. This is the standard we use in every other domain of life. When I choose a career, or a charity, or a political candidate, I don't ask “What would an idealized version of myself think?” I ask “What does the evidence suggest, given my values and my epistemic situation?” The fact that I might be wrong, that there might be unknown unknowns, does not make my choice arbitrary. It makes it fallible.
The impartial altruist can do the same. She can look at the evidence available, consider the consequences she can foresee, weigh them according to her values, and make a choice. She can acknowledge that she might be wrong, that there may be consequences she hasn't considered, that her estimates are rough. None of this makes her choice arbitrary. It makes it human.
DiGiovanni might respond that this is just “going with your best guess,” and that he has already argued against this. But his argument against best guesses relies on the same problematic premise: that without access to the idealized standard, any choice is arbitrary. Once we reject that premise, the case against best guesses falls apart.
Even if we grant P1, the argument still has problems. But let's set that aside and suppose, for the sake of argument, that P1 is true. Does the conclusion follow?
P2 says that if our understanding is sufficiently coarse-grained, we can't compare actions' expected values. P3 says our understanding is that coarse. Therefore, we have no impartial-altruistic justification for preferring any action.
But this inference depends on an assumption that is never explicitly defended: that the only way to have a reason to prefer A over B is to be able to compare their expected values. If we cannot compare them, we have no reason.
This is false. There are other ways to have reasons. I might prefer A over B because A is more likely to be good, even if I can't say exactly how good. I might prefer A because it avoids a particular known risk that B carries. I might prefer A because it is reversible in a way that B is not, giving me more information for future decisions. None of these require me to compare expected values in any precise sense. They are all genuine reasons.
DiGiovanni’s framework assumes that the only kind of reason that counts is one that compares expected values. But this is a stipulation, not an argument. If we broaden our conception of what counts as a reason, then even under conditions of severe unawareness, we can have reasons to prefer one action over another.
If I am right, then the unawareness argument does not show that impartial altruism is non-action-guiding. It shows that impartial altruism is difficult, that we must act under uncertainty, with incomplete information, and with the humility to acknowledge that we might be wrong. But this is true of every moral framework, not just impartial altruism.
So what should the impartial altruist actually do?
First, she should take the problem of unawareness seriously. She should be humble about her ability to predict long-term consequences, and she should build flexibility and reversibility into her plans. She should favor actions that preserve options and generate information, rather than actions that commit to a single path.
Second, she should focus on things that are likely to be good under a wide range of possible futures. Reducing existential risk, improving institutional decision-making, and building a culture of careful reasoning are all examples of actions that seem good, even if we can't predict their exact consequences.
Third, she should adopt a portfolio approach. Rather than putting all her eggs in one basket, she should diversify across different types of interventions, different time horizons, and different theories of change. This doesn't require her to compare expected values across all options. It requires her to recognize that uncertainty is a reason to diversify, not a reason to do nothing.
Fourth, she should engage in ongoing learning and updating. The fact that we are unaware of some consequences doesn't mean we can't become aware of more of them over time. By acting, observing, and reflecting, we can expand our awareness and improve our future decisions.
None of this requires the kind of idealized comparison that P1 demands. It requires something much more modest: the willingness to act on the best available evidence, while remaining open to revision.
The unawareness argument is impressive in its scope and rigor. But it rests on a normative premise that is both undefended and implausible. P1 demands that we justify our preferences by reference to what an idealized version of ourselves would think, even though we have no access to what that being would think and no reason to trust our guesses about it. This sets the bar for justification impossibly high, and it conflates ideal justification with actual justification.
Once we reject P1, the rest of the argument collapses. Even if we accept that our understanding is profoundly coarse and that we are unaware of vast swathes of consequences, we can still have good, non-arbitrary reasons to prefer one action over another. We can act on the best evidence available, we can favor good and reversible actions, we can diversify, and we can learn. None of this requires us to know what an idealized version of ourselves would think.
The impartial altruist is not clueless. She is uncertain, fallible, and humble, but she is not paralyzed. And that is a conclusion worth defending.
I believe this is built into and accounted for with DiGiovanni's ultimate conclusion of the unawareness argument. He never claims that there are no reasons to do things, only that what reasons you may find are not universe-wide, self-independent, time-independent altruistic reasons. He argues that they cannot be both impartial and altruistic reasons. (I disagree, but that's for my own argument).
But I like the article!