Philosophyethics and agency, artificial and human.

My research centers on normative ethics (especially contractualism) and moral psychology, especially as they pertain to AI. Three principles animate my work:

Morality consists—centrally if not entirely—in doing only what one can justify to those they affect; and justification involves the exchange of reasons between two rational agents, in light of what each believes about how each will be personally affected.
One is morally responsible for actions produced by their rational agency, since the capacity to be guided by reasons is what makes one responsible for what one does.
We have a decent but imperfect grasp of when a human’s actions originate in their rational agency, but a poor one in the case of AI agents, since the concepts and criteria on offer are designed for humans.

My PhD qualifying papers—on ex ante contractualism and fairness, the ethics of attention, and subjective and objective perspectives on agency—address each principle in turn. My dissertation, which asks what it would take for an AI agent to be morally responsible, takes up all three at once.

My work in the philosophy of AI is supported by my background in computer science and work as a software engineer before philosophy. I also read Ancient Greek and Literary Chinese (文言文) and enjoy studying historical moral psychology in both languages.

Dissertation

Is it ever fair to hold an artificial agent morally responsible for what it does? I argue that nothing in principle rules this out. Whether an agent—human or artificial—is responsible depends on whether it has the capacity for moral agency, which can be specified in species-independent terms. Specifically, I defend a necessary condition for moral responsibility that I call the Practical Integration Condition, which requires that an agent have the capacity for its moral beliefs and actions to answer to its own deliberation, understood following Moran in Authority and Estrangement. I then argue that we have good reasons, both a priori and empirical, to doubt that current LLMs meet this condition because of how they are trained. I proceed to speculate about what it would take to build an artificial agent that we could fairly hold responsible. Finally, I discuss implications including personal relationships with AI agents and the moral responsibility of parties who deploy them.

Selected papers

Apr 2026 Ex Ante Contractualism and Fairness Ex ante contractualism is a plausible account of fairness because the social risks it permits are natural lotteries that give each person an equal epistemic chance of being harmed.
Apr 2026 The Problem of Double Vision: Subjective and Objective Perspectives Double vision may stick around in the case of belief, but in the case of action we resolve it all the time, since we cannot suspend intentions as we can suspend beliefs.
Apr 2026 Attention, Agency, Morality Because attention is selection for action, lapses of attention can be failures of due care. Loving attention is not necessarily owed, but it is constitutive of the relationships in which things are owed.
Jul 2026 The Disturbance Problem: How Predictive AI Models Reshape Preferences Models that predict our preferences can change them, by making the implicit reasons behind them explicit and statistically justified, so even an accurate and otherwise just model can undermine our responsibility for our preferences.
Jan 2020 “Deepfakes” and the End of the Photographic Age GAN images, like photographs, can be skillfully made to embody artistic intentions, and once the two are indistinguishable, photographs will lose their epistemic authority but keep their aesthetic value, as paintings do today.

For philosophers and everyone else

Shorter pieces on the same questions, written so that an engineer, a lawyer, or a policymaker can follow the argument without the literature.

Mar 2025 Dear Mr. Muskrat: or, How to Defeat Bad Faith Appeals to Non-Identity Mr. Muskrat’s non-identity argument for destroying the Amazon no longer goes through once someone guarantees the existence of one future child whose life the racetrack would make worse.

Other papers, talks, and essays

Mar 2026 To Be an Ideal Theorist, Do Non-Ideal Theory Ideal theory is correct about the goal of political philosophy—pure normative principles that depend on no contingent facts—but only the methodology of non-ideal theory can attain that goal.
Mar 2026 The Normative Structure of Natural Law in Hobbes’s Leviathan In response to a scope ambiguity, I argue that each Hobbesian law of nature has a ‘Foole-first’ structure, requiring that one appear to have an unconditional desire to comply and a disposition to comply if there were a sovereign—as opposed to a structure that requires a desire to appear.
May 2025 Sleeping Beauty Meets Descartes’s Demon: Thirding Leads to Skepticism Thirding, together with substantively rational priors about being ‘skeptically tortured’ by a demon, leads to global skepticism, and halfing does not.
Dec 2024summary in progress Blame at the Margins: Two Objects of Anger If anger is moral protest, it has two objects—a target and an audience—and anger at something that is not an agent can still be partially appropriate as protest to a receptive audience.
May 2024summary in progress Rescuing Fodor and Lepore’s Floundering pet fish Large language models use an augmented prototype theory of concepts and nonetheless achieve compositionality, so prototypes can be compositional and concepts might be prototypes.
Apr 2024summary in progress Estimating Utility Functions for Predictably Irrational Agents with Preference Change An AI that learns what we want from our choices should model our predictable errors and changing preferences instead of assuming them away as the von Neumann–Morgenstern theorem and inverse reinforcement learning do.
Dec 2023summary in progress From Resentment and Forgiveness to Acceptance: A Confucian Response to Structural Inequality Acceptance, which turns attention from ‘I myself was wronged’ to ‘someone was wronged’, keeps victims of structural inequality oriented against it while minimizing their emotional disturbance.
May 2023summary in progress Regret, Blame, and the Self We can regret past actions without being able to blame ourselves, most disturbingly when our ends change so substantially that we can no longer reason with our past selves.
Apr 2023summary in progress Testimonial Injustice and Credibility Excess Credibility excess can be a testimonial injustice to the very person to whom it is attributed, since ascribing someone credibility excess psychologically necessitates ascribing them a credibility deficit about their self-knowledge.
Apr 2022summary in progress Music and the Emotions: A Linguistic Analogy The debate between arousal and cognitivist theories of musical expressiveness is parallel to the debate between intention and convention theories of linguistic meaning, since scores are to performances as sentence-types are to sentence-tokens.
Apr 2022summary in progress A Waltonian Account of Aesthetic Originality Walton’s notion of perceiving a work in a category gives a substantive account of the ‘relevant difference’ that Sibley leaves undefined, and a basis for his ‘revolutionary’ sense of originality.