We know they can beat human champions at chess and Jeopardy, but can the algorithm gurus in Silicon Valley program a computer to beat my 1L law students on my Property Law exam. I doubt it.
This challenge goes to the heart of the “reinvent law” and “law+tech” movements. There’s no doubt that plenty of the work that lawyers traditionally have performed can be substantially taken over or made far more efficient by computers. E-discovery is the obvious example. And there are domains of law steeped in technical rules and linguistics amenable to algorithm programming. The bizarre world of estates and future interests, for example, could very well be reduced to a program that could crunch through problems on my exam, spitting out the correct descriptions of present possessory estates and future interests about as effectively as any lawyer trained in the field (I would buy that program, give it to my students, and drop that section from the course!). But that’s because it is a field consisting entirely of rules and linguistics, with precisely correct answers to each problem and little room for higher-level reasoning.
Where I think the computers would flunk my exam is on the written essay portion. Bear in mind I do not construct insane factual scenarios on my exams–the kind with aliens invading Earth. I use practical scenarios taken from current news or my practice experience and put the students in situations not unlike those practicing attorneys face. To be sure, domain knowledge is essential for success on these questions, and the doctrine behind it could be stuffed into a computer program. But then what? Some of my questions go something like: “How likely is X to prevail on the Y claim?” or “Is there any problem with what the government has done to Y?” Of course, there is an attack strategy I teach my students for such questions–an algorithm of sorts–which I suppose could also be stuffed into a computer. That is what some Silicon Valley legal shops are trying to do for certain fields, as Lex Machina is for patent litigation. The problem is that the fact scenarios on my exam, as in the real world, can be quite nuanced, or they can be incomplete, requiring a decision tree approach with multiple branchings. Well, maybe Silicon Valley can program that too. But then there are questions asking students to advise clients what to do to solve problems, requiring that they explore and compare a variety of options and devise a game plane. Also doable for computers? Maybe so, but I am getting more skeptical as we go along. The most difficult type of question for me to imagine a computer solving effectively is one requiring students to invent new rules for new kinds of property issues, such as how to treat wind as a property interest given the rise of wind power. These questions require consideration of the theory and policy of property law as well as analogical reasoning to identify rules that work well in similar situations, transport them into the new context, and test how well they fit. Try that, Watson.
I can’t reveal the contents of my exam–it’s not being administered until this Friday and some of my students read this blog. But if anyone in Silicon Valley is up to the challenge, I’ll gladly send it to you and grade the computer’s answers.