University of North Carolina School of Law staged an unusual courtroom experiment: an entire “jury” made up of AI systems — ChatGPT (OpenAI), Grok (xAI), and Claude (Anthropic). Three tall displays stood in for jurors as the bots “listened” to a fictional robbery case to probe how AI might influence legal decision-making.
Why this matters
AI tools are already common in legal work for research and drafting. Adoption is rising and many firms report ROI — yet failures persist, from fabricated caselaw to misread facts and hidden biases. Several attorneys have even faced fines for filing AI-generated briefs with bogus citations.
What went wrong (for now)
- AI can’t read body language or tone in the courtroom.
- Lacks lived human experience and empathy in judgment.
- Can badly misinterpret text due to simple typos.
- Risk of racial and contextual bias creeping into “verdicts.”
Post-trial panelists were broadly critical; many attendees left thinking that “trial-by-bot” isn’t a good idea yet.
Technology will recursively repair its way into every human space if we let it — including the jury box.
Could it improve?
Proponents argue these gaps may narrow: video feeds for non-verbal cues, richer context to reduce errors, and steady model upgrades. The harder question is how far AI should be allowed into the justice system — and who is accountable when it errs.
