Agent-Regret
Agent-regret is Bernard Williams's most consequential contribution to moral psychology: the recognition that a morally serious person feels the weight of outcomes her agency produced, even when she could not have prevented them and acted without fault. The concept rejects the utilitarian claim that correct calculation dissolves residue and the Kantian claim that fulfilled duty ends the moral story. The lorry driver who kills a child through no negligence feels something a bystander does not — not guilt, but the specific heaviness of causal participation. Williams argued that a person who felt nothing would not be admirably rational but morally deficient. The AI transition generates agent-regret among builders whose work accelerates the displacement of the practices that formed them.
In the AI Story
The concept emerged in Williams's 1976 essay Moral Luck, where he used the case of the blameless lorry driver to expose a gap in every systematic moral theory available. Utilitarianism treats regret as irrational residue once the calculation is complete. Kantianism relegates emotional aftermath to psychology. Williams insisted both responses misdescribed moral life. Agent-regret is not a malfunction of rationality but the correct perception of a morally serious person — the recognition that her agency produced the loss, and that this fact carries weight the justification cannot discharge.
Williams distinguished agent-regret sharply from bystander regret. Anyone can regret that a child was killed. Only the driver can regret that he killed the child. The difference is not psychological intensity but structural: agent-regret attaches to the specific fact of causal involvement. It cannot be transferred, shared, or absolved by demonstration that the action was justified. This is why the concept cannot be reduced to guilt — guilt requires fault, and the lorry driver has none — nor to sadness, which is available to everyone who hears the news.
The AI context sharpens the concept's application. Segal's confession in The Orange Pill — standing in Trivandrum unable to tell whether he was watching something born or buried, lying awake unable to distinguish creative exhilaration from the architecture of addiction he had himself helped design earlier in his career — is a textbook case of agent-regret. He did nothing wrong. The transition is justified. And the weight is his, because his hands are on the wheel.
The engineer who benefits from AI while perceiving what it displaces occupies the same structural position. She has not caused the transition. She could not have prevented it. But her participation contributes to the erosion of the conditions under which her own expertise developed, and the recognition of this fact produces a weight that the productivity gains cannot discharge. Williams would have insisted that the weight is evidence of moral seriousness, not a failure to adapt.
Origin
Williams developed the concept in the title essay of his 1981 collection Moral Luck, though the volume itself gathered work from the previous decade. The essay was originally delivered as a joint symposium with Thomas Nagel at the 1976 Aristotelian Society meeting — Nagel defending a broader account of moral luck, Williams focusing on the first-personal residue. The two papers are almost always read together, but they make different arguments, and Williams's is the more uncomfortable for moral theory.
Key Ideas
First-personal structure. Agent-regret is available only to the person whose agency produced the outcome; it cannot be felt vicariously or dissolved by third-party absolution.
Independence from fault. The emotion is appropriate even when the agent acted blamelessly, because its subject is causal involvement rather than wrongdoing.
Diagnostic of moral seriousness. A person who feels no agent-regret after producing serious loss through blameless action has failed to perceive the values at stake, not succeeded at rationality.
Irreducibility to calculation. No aggregate justification dissolves the residue, because the residue is not a claim about whether the action was correct but about what the correct action nevertheless cost.
Applicability to collective agency. Institutions deploying AI without feeling the weight of what their deployment displaces are institutions that have sacrificed moral perception for operational efficiency.
Debates & Critiques
Contemporary philosophers have pressed Williams on whether agent-regret is genuinely distinct from a family of related emotions (guilt, sadness, Nagelian moral luck) or merely a phenomenological cluster admitting of functional reduction. Susan Wolf, Margaret Walker, and others have defended and extended the concept; Brad Hooker and consequentialist writers have argued it can be accommodated within sophisticated utilitarianism. The 2025 application of Williams's framework to AI alignment (Dogramaci and others) treats the concept as indispensable for understanding the moral phenomenology of AI deployment decisions.
Appears in the Orange Pill Cycle
Further reading
- Bernard Williams, Moral Luck: Philosophical Papers 1973–1980 (Cambridge University Press, 1981)
- Bernard Williams and Thomas Nagel, 'Moral Luck' (Proceedings of the Aristotelian Society, Supplementary Volume 50, 1976)
- Margaret Urban Walker, 'Moral Luck and the Virtues of Impure Agency' (Metaphilosophy, 1991)
- Susan Wolf, 'The Moral of Moral Luck' (Philosophic Exchange, 2001)
- Daniel Statman (ed.), Moral Luck (SUNY Press, 1993)