Hercules (Dworkin's Judge)

In the [YOU] on AI Field Guide

The cycle's gallery of thinkers returns repeatedly to the question of what intelligence is for—what it means to be very capable at a task that matters. Hercules embodies one answer: intelligence in the service of principled justification, of taking each person seriously as an equal, of building a body of law that a community could endorse as its genuine moral commitment. The machine embodies a different answer: intelligence in the service of prediction, of fitting the historical distribution, of producing consistent outputs with high confidence. These two answers are not close to each other. They are separated by exactly the gap Dworkin identified between fit and justification, and the gap does not close as the machine becomes more capable. A more accurate legal prediction system is not a closer approximation of Hercules. It is a more accurate prediction system—a different kind of thing, better at what it does, farther from what Hercules represents.

The cycle's observation that fluency and authority have decoupled finds its most consequential expression in the legal domain. An AI system can produce opinions that read as Hercules's work—that cite the right precedents, articulate a coherent principle, apply it with apparent consistency. This surface is not Hercules's activity. It is the shadow his activity casts across a training corpus, and the shadow can be reproduced without the substance. The danger is not that the shadow is ugly. The danger is that it is beautiful—that it looks like what principled adjudication looks like, while doing what prediction does. Dworkin's Hercules is the standard that makes this distinction legible: the question is not whether the output reads as a reasoned opinion, but whether the system that produced it is responsive to reasons, committed to a principle, and accountable to the person it decided about.

Origin

Hercules first appeared in the chapter “Hard Cases” in Taking Rights Seriously (1977), introduced as a device for working through what it would mean for a genuinely hard case to have a right answer. The strategy was to stipulate a judge who faced no human limitations and ask what he would do—not because real judges could do the same, but because the exercise revealed the structure of the task even when its full execution was impossible. Hercules appeared again, developed further, in Law's Empire (1986), where Dworkin applied him to the law-as-integrity thesis. In each appearance he serves the same function: as the idealized instance of an activity that real judges approximate imperfectly, making visible what the activity essentially consists of by removing the noise of human limitation.

The choice of the name was not accidental. Hercules is the mythological figure of superhuman labor—the one who could do what ordinary heroes could not, not because he used a different kind of strength but because he had vastly more of the same kind. Dworkin's Hercules differs from a real judge only in degree, not in kind. This is the crucial feature: the thought experiment stipulates that the judge has more of what real judges have, not that he has something they lack entirely. His moral judgment is better than theirs—more comprehensive, more consistent, less subject to bias and fatigue—but it is moral judgment, the same cognitive operation, not a different one. This is precisely what the machine cannot claim: it has something real judges lack entirely in their capacity as judges, which is the optimization over a statistical distribution, and lacks something they have, which is moral judgment as a genuine responsiveness to reasons.

Key Ideas

Superhuman in capacity, human in kind. The essential feature of Hercules is that he is the idealization of what judges already do, not the replacement of what they do by a different process. He interprets, weighs fit against justification, brings moral judgment to bear, holds convictions he is responsive to revising when shown better arguments. He differs from a real judge only in that he can do these things perfectly. This means that asking whether AI could build Hercules is asking whether AI could perform the same activity judges perform, only better—and Dworkin's answer is that current AI systems perform a different activity, not a worse version of the same one.

Errors answerable to argument. Hercules can be wrong. Dworkin is explicit about this. When Hercules is wrong, he is wrong in the way a conscientious interpreter is wrong—he has reached, by reasoning, a conclusion that better reasoning would correct. Show him the superior interpretation and, being genuinely responsive to reasons, he revises. This answerability is what makes him a model of justice: he is wrong correctably, in the space of reasons. A machine that is wrong is wrong for different reasons: the data was biased, the distribution shifted, the correlation learned does not hold in this case. These errors are not answerable to argument in the relevant sense, because the machine does not hold its outputs for reasons that an argument could engage. You cannot persuade a model; you can only retrain it, which is a different kind of correction in a different kind of space.

Legitimacy through participation. Hercules is not a sovereign issuing answers from outside the legal practice; he is a participant in the practice, bound by its history, accountable within its community of principle. His authority derives from his fidelity to the community's law and his accountability to its evolving moral sense. The AI decision system sits outside any community of principle: it was trained by a company, deployed by an administrator, and is accountable to neither the litigant nor the law but to the objective function its designers chose. To install it in Hercules's place is not to perfect the judiciary but to replace a participant in a moral practice with an instrument owned by interests external to that practice—a substitution that changes the kind of authority that decides without any of the legitimating features that made Hercules's authority legitimate.