Skip to content

Moral Luck Revisited

Williams and Nagel introduced the problem of moral luck nearly fifty years ago. In its standard formulation, the problem is this: our moral assessments of agents seem to depend on factors outside their control — the drunk driver who hits no one is treated differently from the drunk driver who hits a pedestrian, even though their conduct and their character may be identical — and this dependence is, intuitively, in tension with the principle that we are properly assessed only for what is within our control.

The literature on moral luck has been productive in distinguishing varieties of luck (resultant, circumstantial, constitutive, causal) and in mapping the implications of each for moral assessment. Less progress has been made, I think, on the underlying problem. We have a better-articulated puzzle but not, on the whole, a resolution.

Two responses

Two responses to the problem have dominated the literature.

The denial response argues that the apparent dependence of assessment on luck is illusory. We do not, on reflection, hold the drunk driver who hits a pedestrian more blameworthy than the one who does not. We merely impose harsher legal sanctions because outcomes matter for institutional reasons. The moral assessment is the same.

The acceptance response argues that the dependence is real, and that we should revise the control principle accordingly. Luck is a feature of our moral practice, not a bug. The control principle, on this view, was always too strict; what we should require is that the agent’s conduct be controlled, not that the outcomes be.

Both responses, in my view, are unsatisfactory.

Against the denial response

The denial response strikes me as empirically false. We do assess agents differently in light of outcomes, even when we know the conduct was identical. The driver who hits no one is the subject of relief, the driver who hits a pedestrian the subject of horror. These are not merely institutional differences. They run through our pre-theoretical reactive attitudes.

The denier can respond that the reactive attitudes are not properly speaking moral assessments — they are something else, perhaps regret or grief, that we have confused with moral assessment. But this move seems to me to give up too much. It saves the control principle at the cost of declaring most of our practice non-moral.

Against the acceptance response

The acceptance response has a different problem. If we accept that luck features in moral assessment, we lose the resources to criticize cases where luck is doing too much work. The agent who is judged harshly for an outcome she could not have foreseen, in conditions she did not choose, has no recourse on the acceptance view. Her bad luck is, on the acceptance view, just part of the moral facts.

This seems to me to give up too much in a different direction. It saves the practice at the cost of the principle’s critical bite.

A third option

A third option, less developed in the literature, is to hold that luck-sensitivity is itself a feature of moral practice that can be evaluated. We can ask, in any given case, whether the role luck is playing in our assessment is appropriate, and we can criticize cases where it is not. On this view, the control principle survives, but in a more nuanced form: we are properly assessed for what is within our control, plus the foreseeable consequences of what is within our control, plus a margin of luck-sensitivity that is itself open to evaluation.

This is messier than either of the dominant responses, and it requires substantive work on what counts as “appropriate” luck-sensitivity. But I think it captures more of our actual practice than either alternative, and it preserves the critical bite that the acceptance response gives up.

What is at stake

The dispute about moral luck is sometimes treated as a technical curiosity. I think it is more than that. How we resolve the question shapes how we hold one another accountable in conditions of uncertainty, which is to say, in most of our actual lives. The drunk driver who hits no one is a case study; the question generalizes to nearly every act whose consequences are not fully foreseeable.

If we get this question wrong, we end up with practices of accountability that are either too lenient (the denial response, generalized) or too punitive (the acceptance response, generalized). Either way, we have failed to do justice to the agents involved.

More from this kitchen

Add your two pesos

Your email address will not be published. Required fields are marked *