Wayne Unger, a Quinnipiac University School of Law associate professor, is all for using artificial intelligence.
But not for his final exam.
In his smaller classes, such as Data Privacy & Security Law, he’s brought back a tried-and-true method to make sure his students know the material and are not relying too heavily on AI: one-on-one, 30-minute oral exams. In his larger courses, he switched to “pen-and-paper only,” with all responses handwritten.
“I want to develop a skill set with our students—that’s critical thinking, that’s analogical reasoning, that’s on-your-feet thinking,” Unger says, “and I can’t do that if a tech solution does all that for them.”
Like Unger, some law professors are turning to old-school methods to address a new challenge: How do you know whether your student knows something or has offloaded their thinking to artificial intelligence?
“One of the core questions for legal education is how do you adapt to the presence of a tool that can mimic or mask what looks like analysis?” says Kirsten K. Davis, a professor at Stetson Law. “The worry you have is that a test will not tell you what a student knows.”
Many professors already have ditched or cut back on writing assignments outside of class “because AI can do a fairly decent job of replicating what a novice can do in a way that’s passable,” she adds.
None of the professors contacted by the ABA Journal ban AI, and all purposefully and intentionally use it in their classrooms and assignments. But they’re finding ways to double-check their students’ true judgment and knowledge of the law with oral exams and other traditional formats, such as tests using locked-down software.
These moves come as law schools reconsider their AI policies. In May, the University of California at Berkeley School of Law banned students from using artificial intelligence for class assignments and during exams. Earlier this month, the University of Texas School of Law Dean Robert “Bobby” Chesney called for faculty to use Socratic-style questioning in teaching and to view classroom time as “the sole context in which a professor can be certain” a student is learning without AI.
In Unger’s smaller classes, the final oral exam is 50% of the grade, so “they have to take it seriously,” he says. Other components of the grade include class participation and experiential assignments, like an in-class debate and a data breach simulation, he adds.
About a week before the exam, Unger sends students the exam prompt, allowing them to time to formulate their thoughts, along with his grading criteria, which covers the students’ ability to think on their feet and their persuasiveness.
“Our basic rules of professional responsibility say you must be competent in your practice, and that means if you discover something that might be relevant, you need to read it and understand it.”
Students schedule their exam appointment, which includes 20 minutes for the test and 10 minutes of feedback. “I sit back to back to back, student after student after student,” he says.
During the test, students can bring notes to reference the case, he says, and he asks questions to push their assertions.
“If they use AI to prepare for this oral final exam, it will show by my subsequent questioning, my pushing them on, ‘Do you actually know this or not?’” he says. “If they can’t defend it, if they can’t go deeper into the analysis, then it shows.”
He gives feedback as soon as the exam is over, which students appreciate, he adds. “Even the students who have performed relatively poorly still very much enjoyed and supported this form of assessment,” he says. “It allows them to display other skills than on the standardized test model.”
And there are upsides for professors, Unger says. “I’m not spending two weeks after the exam grading,” he adds, “and I’m more confident that the students know the material than when they turn in papers.”
Some law schools adopted rules early on to include old-school methods while striking a balance between embracing AI and ensuring students can demonstrate their skills.
At Case Western University School of Law, Dean Paul Rose notes “the default rule is that final exams are administered in person and are either closed-book or open-book with access only to hard-copy materials and no access to digital copies, computer hard drives or the internet.”
“We also encourage faculty to include a verbal component if a significant portion of the grade for a course is based on a paper or to do writing assignments in stages so faculty can observe students’ progress over time,” he adds.
Northwestern University Pritzker School of Law’s AI and Legal Education Committee advises faculty to identify challenges in writing courses and suggests considering a variety of approaches, like allowing broad AI use, providing AI tutors at each stage of writing projects, and including an oral assessment, says Daniel W. Linna, director of law and technology initiatives. Northwestern instructors can consider whether students must disclose how AI was used or only allow AI only at early or later stages of the writing process, he adds.
“I am in favor of all of those approaches and encourage faculty to utilize them,” says April Dawson, associate dean of technology and innovation at North Carolina Central University School of Law and a 2026 Legal Rebel. “Changing how we assess is necessary.”
Dawson notes the need to also rethink the structure of class time, incorporating more experiential learning— such as presentations, simulations and client conversations—to develop professional skills in the age of AI.
Still, especially in writing courses, there is a concern that AI use can go too far.
“The three words that I use are ‘sustained critical engagement.’ How do I keep them active on the project when I’m not there?” Stetson’s Davis says. “My mantra has been, ‘You must have lawyer intelligence to use artificial intelligence.’ ”
Linda Anderson, professor and director of Stetson Law’s Flex JD program, recently conducted her “first real test” in her Research & Writing class, a course she’s taught for about 20 years.
“I didn’t want to find myself in a situation where I couldn’t tell whether it was AI-generated or not,” Anderson says. “I didn’t think that was fair to the students. I didn’t think it was fair to me.”
Instead of the typical end-of-semester writing assignment, like a big office memo or appellate brief, she moved to an in-class test conducted on locked-down software. The test’s format, including performance tests, multiple-choice questions and short answers, is based on the NextGen UBE Bar Exam, which set to have its first administration in July.
“AI helped me build the test. I’ll be very clear about that,” she says.
Meanwhile, Linna worries about overcorrection.
“Adjustments to assessments may be useful, but they’re being implemented to avoid AI when we need to embrace it,” he says. “Professors need to be proactive, consider the learning objectives of their courses, and redesign them to make the most of AI to accelerate student learning and skill development” via AI tutors, mentors and simulations.
All agree there is no set formula for AI use, and each course demands specific consideration. “Faculty need to get engaged and figure out what is best for their courses to ensure that students are achieving the learning outcomes,” Linna says.
Leave a Reply