North Carolina School of Law Puts AI Supreme Court to the Test

The University of North Carolina School of Law conducted a simulated Supreme Court hearing to see to what extent artificial intelligence can mimic the reasoning processes of U.S. Supreme Court justices.

At an on-campus artificial intelligence-focused conference held at the University of North Carolina School of Law, students and academics staged a simulated Supreme Court hearing based on the case Monsalvo Velázquez v. Bondi. Nicole Downing, the law school's associate dean for AI, stated that the goal was not to see whether AI could replace human judges, but to examine whether AI could exhibit reasoning similar to published Supreme Court decisions.

For the hearing, instead of all nine justices, three AI judges were programmed representing the textualist/originalist wing, the pragmatist wing, and the progressive wing. Developed using ChatGPT by university data librarian Rolando Rodriguez, these AI judges were provided with the case background, petitions, relevant statutes, rules, and guidelines, along with the past opinions of the justices they represented.

While the actual case resulted in a 5-4 vote ruling that the voluntary departure period extends to the next business day if it falls on a weekend or holiday, in the simulation, the progressive and pragmatist AI judges formed the majority decision to extend the period, while the textualist AI judge voted against the decision.

Nicole Downing stated that AI reasons similarly to humans but that there is still a major gap in terms of acting like humans. It was noted that the AI judges asked questions to the students but did not interrupt or break in like real judges do, and the interactions felt robotic.