用AI模拟普通人判断合理性，为法律标准提供可量化的实证依据。

This Article introduces the generative reasonable person, a new tool for estimating how ordinary people judge reasonableness. As claims about AI capabilities often outpace evidence, the Article proceeds empirically: adapting randomized controlled trials to large language models, it replicates three published studies of lay judgment across negligence, consent, and contract interpretation, drawing on nearly 10,000 simulated decisions. The findings reveal that models can replicate subtle patterns that run counter to textbook treatment. Like human subjects, models prioritize social conformity over cost-benefit analysis when assessing negligence, inverting the hierarchy that textbooks teach. They reproduce the paradox that material lies erode consent less than lies about a transaction's essence. And they track lay contract formalism, judging hidden fees more enforceable than fair. For two centuries, scholars have debated whether the reasonable person is empirical or normative, majoritarian or aspirational. But much of this debate assumed a constraint that no longer holds: that lay judgments are expensive to surface, slow to collect, and unavailable at scale. Generative reasonable people loosen that constraint. They offer judges empirical checks on elite intuition, give resource-constrained litigants access to simulated jury feedback, and let regulators pilot-test public comprehension, all at a fraction of survey costs. The reasonable person standard has long functioned as a vessel for judicial intuition precisely because the empirical baseline was missing. With that baseline now available, departures from lay understanding become transparent rather than hidden, a choice to be justified, not a fact to be assumed. Properly cabined, the generative reasonable person may become a dictionary for reasonableness judgments.