The Silicon Reasonable Person: Can AI Predict How People Judge Reasonableness?
Abstract
In everyday life, people make countless judgments of reasonableness—judgments that determine what speed to drive on a busy street, what an advertisement likely means, and whether sufficient consent was given to their romantic gesture. Predicting these judgments proves challenging for the legal system. Judges proclaiming how people would interpret a message or situation are often accused of relying on elite intuitions or making crypto-political decisions. Juries are beset by selection and adversarial biases, and surveys and focus groups are expensive tools available only to wealthy litigants. This challenge stems from the fact that people’s judgments rely on intuitive but hard to explain mental frameworks that follow complex statistical patterns.
This Article investigates whether large language models (LLMs)—industrial-grade pattern detectors trained on vast text corpora—can learn to identify the subtle patterns that drive human reasonableness judgments. Using randomized controlled trials that compare humans and models across multiple legal contexts and over 10,000 simulated judgments, the Article demonstrates that certain models capture not just surface-level human responses but potentially their underlying decisional architecture. Strikingly, these systems prioritize social cues over economic efficiency in negligence determinations, just as humans do—even though this contradicts textbook treatments. While capability varies across models and contexts, there are initial signs that LLMs have learned to detect the cues that shape human reasonableness judgments, despite not having been trained for that purpose.
These findings, while still preliminary, simultaneously advance scholarly conversations about consent, negligence, and contract interpretation while suggesting immediate practical applications. They offer judges future tools to calibrate elite intuitions against broader societal patterns, enable lawmakers to test how policy interpretations might resonate, and provide resource-constrained litigants means to preview argument reception. Most urgently, as AI agents increasingly make autonomous decisions in the real world, understanding whether they have internalized recognizable ethical frameworks becomes essential for anticipating their behavior in novel situations.
Read the full piece at SSRN.