Moral Computation: Law, AI, and What We Call Ethics

The Supreme Court splits on transgender athletes while arXiv drops a paper on bounded morality. Both reveal ethics as a computational problem nobody wants to admit.

Based on original reporting by Jeannie Suk Gersen (The New Yorker), Max Kanwal (arXiv), Caryn Tran (arXiv) and Patrick Mineault (arXiv)

The same week the Supreme Court issued a 6-3 split on transgender athletes and equal protection, a paper titled Bounded Morality: Defining the Space of Moral Computation landed on arXiv. The coincidence is almost too neat. Both the court and the researchers are grappling with the same impossibility: how do you formalize ethical judgment across cases where the underlying values are genuinely contested, not just uncertain?

The Deontology Trap in Law and in Code

The Bounded Morality paper, by Max Kanwal, Caryn Tran, and Patrick Mineault, argues that treating moral cognition as adherence to fixed ethical theories, whether deontological rules or utilitarian calculus, fails to capture how moral reasoning actually works. Jeannie Suk Gersen's parallel New Yorker piece on why liberals abandoned a moral reading of the Constitution traces a legal version of the same failure: when one side locks in a procedural framework and the other side abandons the moral argument, you get unanimous Title IX decisions and 6-3 equal protection splits in the same ruling. Process and ethics are not the same thing. Neither legal formalism nor AI alignment frameworks have a clean answer for what happens when the underlying value system is genuinely plural.

The Constitution as a Training Dataset

There is a systems framing available here that the legal community resists and the AI community has not yet earned. Constitutional interpretation and AI alignment both involve attempts to infer stable preferences from ambiguous historical signals, apply them to novel cases, and produce outputs that a legitimacy-granting institution will accept. The failure modes are identical: overfitting to precedent, brittleness at edge cases, and the tendency to launder contested values as neutral process. What the Bounded Morality paper proposes, a more dynamic model of moral computation that accounts for contextual and cultural variation, is roughly what Justice Jackson's dissent was reaching for in the equal protection split. The court and the machine are running the same algorithm. Neither is converging.