Normally, arguing a case in front of the Supreme Court involves all nine justices. And normally, those justices are human.

Students, librarians, and faculty at UNC School of Law shifted the script in a mock oral argument with AI justices on October 22. As part of the School of Media and Journalism’s ConvergeCon, “May it Please the Machine” was a collaborative effort across disciplines that put artificial intelligence on the bench. Two student advocates (3Ls Elijah Baetz and Reda Ansar) took sides on a real immigration case, Monsalvo Velázquez v. Bondi, that questioned the technicalities of a 60-day voluntary departure when the final day lies on a weekend or legal holiday. Three AI justices, developed by Humanities Data Librarian Rolando Rodriguez, heard the arguments arguments and presented an opinion.

Both the students and the AI justices were given the same case packet in preparation for the argument. With no access to the internet, the AI justices could only synthesize their opinions based on the materials provided, the advocate’s arguments, and their fellow justice’s commentary. Before the arguments began, we heard from Nicole Downing (Assistant Dean for AI at UNC’s law library) about the motivations behind the experiment and the case in question. After the judges gave their opinions, the floor was opened for questions and commentary that led to discussions about the exercise’s future applications.

Here’s how it went: each student advocate presented their argument, received questions from the bench and was given a chance to respond before the justices deliberated and delivered their final opinions. The results mimicked the ruling in the original case, with the Court upholding that the 60 days do not have to be consecutive.

Takeaways

Collaboration was key

With many moving parts, the event was a joint effort between the School of Media and Journalism, UNC Libraries, and the Law School. Law librarian Nicole Downing provided the legal expertise and a carefully selected case packet as data librarian Rolando Rodriguez programmed the justices, adjusting and re-adjusting as he went. Input from Nicole and the student advocates helped Rolando arrive at the final iteration of the models.

Before the big day, the students and librarians did a test run of the argument. They found that the justices were not quite as assertive or disputative as a human bench might be. Realism was the goal here, so Rodriguez adjusted their programming to reflect that. There was even collaboration between the justices themselves. Throughout the event, they responded to one another’s comments, posed questions, and referred back to each other’s previous statements to bolster their own arguments.

Something was missing

Oral arguments in front of the Supreme Court can feel like an exercise in quick thinking. Attorneys are constantly interrupted by justices and with only 30 minutes to argue their points, the pace moves quickly.

This time, that wasn’t the case. We had been forewarned that the AI would need some time to “think” before responding, and it certainly did. At times, we sat for 30 seconds or more, a silence that would be unheard of before nine human justices. During the discussion, UNC Law faculty Annie Scardulla called oral arguments “a professional conversation” and noted that this element was somewhat lost with AI. The student advocates also noted that being able to read body language and tone is a huge factor in adjusting their argument on the fly. Without those indicators, they were unsure how their comments would be received by the AI justices.

But there is a silver lining to the longer pauses: because the event was part of a larger tech conference, some of the audience likely lacked a legal background. The time between responses allowed for a moment to digest the arguments and responses in a legal setting that would otherwise be dizzying for the regular layperson.

Clear roles

When programming the AI justices, Rolando Rodriguez gave them each a “personality” that might have reflected the distribution of values and political opinions present on a real-life bench. There was one textualist, one pragmatist, and one progressive. These personalities shone through in their word choice and ultimately, their decisions.

However, the justices maintained their original positions throughout the argument until the final decisions were shared; none had a change in opinion. This begs a few questions: which played a larger part in the justice’s final decisions? The students’ arguments or the case packets? Or did their programmed personalities have a greater influence than the law they were provided?

Serves as a simulation

After the arguments were concluded and the case was decided, the discussion centered around what the real-world applications might be for an exercise of this kind. Both the law students and faculty agreed that a mock argument with AI would be useful as tool for attorneys and students, but shouldn’t be the only method used to train or teach. Instead, they felt it was best used to prepare for an argument by testing out how justices might respond to certain claims. It also provided insight into the possible questions or hypotheticals that justices might pose in response to an argument. These applications were especially useful for attorneys preparing to argue a real case, but were also helpful for law students looking to hone their skills in a low-stakes environment.

Further reading

To read more about the event, take a look at these articles published by UNC School of Law and Thompson Reuters.