When War Breaks Out Over the Family Business, They Get the Call | Brent Crane
body
A.I. may be changing legal jobs. But Chris Hoofnagle, a professor at the U.C. Berkeley School of Law, doesn’t want the technology to transform legal education.
“It’s just like you wouldn’t give a second grader a calculator,” Hoofnagle said. “One has to learn the basics of learning before using high power tools that appear to do reasoning for you.”
Advertisement
He helped develop a hard line A.I. policy for the law school, which it adopted this month. The new policy prohibits the use of A.I. for conceptualizing, outlining, drafting, revising or editing class work, as well as for exam use and translating.
Hoofnagle, who also teaches python programming and seminars on new technologies (“I’m radically ‘pro’ this technology,” he says of A.I.), talked with DealBook’s Sarah Kessler about the decision.
Is this an A.I. ban?
This is a default policy that any faculty member can opt out of. There’s no decanal review, so one doesn’t have to justify opting out.
The idea is to protect the formation of first-year law students by requiring them to do the reading, analysis and argumentation that is constitutive of legal work. Once that happens, Berkeley offers many courses explicitly about A.I. lawyering.
What are acceptable uses of A.I.?
Self-tutoring, getting ready for class. That would be a permissible use. It could also be used on law journals and so on.
Advertisement
Were you seeing something that made you feel this sort of rule was necessary?
We had an uptick in misconduct cases. What the administration told us was that our previous policy, where we allowed students to use A.I. for brainstorming and for grammar correction, ended up being a kind of “get out of jail free” card.
Students would be fingered for A.I. use, for plagiarism. And then come back and say, well, I didn’t plagiarize, I just grammar-corrected. And what that grammar correction did is that it actually changed the paper.
The other thing is that the LLMs became far more capable. It’s possible now to write a graduate level research paper in minutes.
Can you really enforce this?
The assessments I write test unresolved issues in tort law and present facts that exist nowhere in reality. Another thing I’m seeing are assignments where the student has to show the process of each step of their work.
Other faculty are doing in-class writing assignments. I’d say the number of faculty who do an in-person authored exam instead of a take-home has about doubled in the last year.
Advertisement
But how would you know if my idea was conceptualized with the help of ChatGPT?
We all know our literature really well. So I know the difference between an original idea and one that has been hijacked from an existing commentator.
Is it to some extent an honor system?
Misconduct rules always have holes. Law students are paying a lot of money, and there’s a lot of opportunity cost to going to law school. There is an earnestness to law students of wanting to improve their skills, and they can’t do that if their assignments just come out of Claude.
It sometimes feels like half the world is saying that you need to use A.I. for everything, or you’ll be behind, and the other half makes this argument that if you’re not thinking it all through yourself, you’re losing something. So if I’m just trying to make myself into the best lawyer, is it hard to know which path is right?
I think what we’ve come to is that one has to have the basic skills of case analysis and cogent argumentation. Otherwise one will not be able to recognize a quality argumentation from an L.L.M.