The University of Chicago Law School announced a new policy for teaching AI with headlines "AI strategy bans phones, laptops in class for first year students." What does history have to say?
A wonderful trip down memory lane. My first programming in college involved writing code to convert Celsius to Fahrenheit. It was clunky, required the use of a mainframe, which was all that was available at that time in the early 70's. It was an experience that convinced me that programming this way would only be frustrating and unlikely to convince anyone but a hard core believer. The rise of better programming language and personal computing eventually changed my thinking, but I was in college at a time that did not allow me to understand how fast the field would change.
Looking back in time a little farther, I bought one of the first HP-35 calculators in 1972 while I was a chemistry major at UC Riverside. It was an amazing little piece of hardware for its day which supported RPN and trig functions, square roots, exponents, etc. and HP actually sold it on an installment plan when it first came out (luckily for me, since I had a part-time job at minimum wage). I remember bringing it to labs and exams and being the focus of both envy and outrage among the other students, who were doing their calculations on pencil and paper. As best I can remember, the faculty ignored the issue, and of course by the time I was a senior, calculators were everywhere.
Amazing. My freshman year the math library had those calculators bolted to special tables where you could get the power supply and use one in exchange for your student ID.
But I feel you are missing a point here; we _do_ forbid the use of calculators in the first math classes, so kids can learn how to do math.
The point here is education and learning: students will take shortcuts, institutions and teachers have to try and make it hard so students _can_ learn. The point of school/college/university is not to _do_ (who cares about results of equations, or about an history paper: it's not like we are going to use the results! We already have great history books, and we already "found X"), but to learn to do.
For me the greatest analogy is this one: you don't use AI in education, the same you won't use a forklift to lift weight at the gym. The point is not lifting weight; it's the process that matters. You want to build muscles, so you go through the trouble. If you want to learn to do something, you have to go through the same process, and do things yourself.
I don't think you are giving UChicago enough credit for trying to encourage AI use in clinical and other practical settings, or the degree to which they polled AI-using law firms in forming their strategy.
I teach in a business school, and so there is no question of "banning AI" or other such tactics. We want students to use LLMs as they would in their jobs. The problem is that to be good at using LLMs in your job you have to have learned enough to evaluate output and know what to prompt and how to shape that prompt. And that is very hard to ensure in an environment where people can turn in plausible, fluent output without reading it themselves.
The lazy student is fine when there is a professor there to assign cases and readings and offer study questions and even (human) prompts that break the problem down into parts. He just puts the readings into the RAG, turns the assignment questions into prompts, and hits return. The output from this method is still, as of last fall, mediocre MBA boilerplate (often accompanied by LLM errors--I have received many similarly worded student papers asserting that advertisers are on the "subsidy side" of Facebook, for example--oops). But once in the working world, there won't be a professor to pre-select and pre-analyze the problem and the worker bee will have to figure all that out himself. And grade his own work. A student who has followed the lazy path will not be able to execute. THAT is the difference between LLMs and calculators or word processors--these older technologies work just as well without the scaffolding of the instructor's syllabus and organization and selection of texts and questions.
I entered 9 years before you. Rooms full of keypunch machines, sometimes with a queue of people waiting for one to free up. Fortunately, I'd learned about card saws (during an internship at IBM) so I often could borrow one from an operator and free up a "down" machine by vigorously going back and forth at the punch station with the saw to get a jammed card out. (The machines were built like tanks.)
You'd hand your punched deck of cards to the operator and wait a while (an hour, sometimes less, sometimes more?) for a printout of your run to come back. No debugger except the one that ran in your head--no "step into", "step over" :-)
Toward the end up my time at Cornell, the Computer Science department acquired a PDP-11 and I got my first exposure to UNIX in ~1977. I still remember my reaction to "ls", "rm", and "grep" as the names of important commands--WTF?
Funny that the Harvard case wasn't about cheating at first. What reached the Dean was other students' anxiety about fairness, and the cheating theories only showed up later, once people had time to imagine what the tool might become. Same ordering as the AI bans now: the ban lands first, the justification catches up after.
A wonderful trip down memory lane. My first programming in college involved writing code to convert Celsius to Fahrenheit. It was clunky, required the use of a mainframe, which was all that was available at that time in the early 70's. It was an experience that convinced me that programming this way would only be frustrating and unlikely to convince anyone but a hard core believer. The rise of better programming language and personal computing eventually changed my thinking, but I was in college at a time that did not allow me to understand how fast the field would change.
Looking back in time a little farther, I bought one of the first HP-35 calculators in 1972 while I was a chemistry major at UC Riverside. It was an amazing little piece of hardware for its day which supported RPN and trig functions, square roots, exponents, etc. and HP actually sold it on an installment plan when it first came out (luckily for me, since I had a part-time job at minimum wage). I remember bringing it to labs and exams and being the focus of both envy and outrage among the other students, who were doing their calculations on pencil and paper. As best I can remember, the faculty ignored the issue, and of course by the time I was a senior, calculators were everywhere.
Amazing. My freshman year the math library had those calculators bolted to special tables where you could get the power supply and use one in exchange for your student ID.
I liked the history part, as usual.
But I feel you are missing a point here; we _do_ forbid the use of calculators in the first math classes, so kids can learn how to do math.
The point here is education and learning: students will take shortcuts, institutions and teachers have to try and make it hard so students _can_ learn. The point of school/college/university is not to _do_ (who cares about results of equations, or about an history paper: it's not like we are going to use the results! We already have great history books, and we already "found X"), but to learn to do.
For me the greatest analogy is this one: you don't use AI in education, the same you won't use a forklift to lift weight at the gym. The point is not lifting weight; it's the process that matters. You want to build muscles, so you go through the trouble. If you want to learn to do something, you have to go through the same process, and do things yourself.
History doesn't repeat itself, but, it most certainly rhymes.
I don't think you are giving UChicago enough credit for trying to encourage AI use in clinical and other practical settings, or the degree to which they polled AI-using law firms in forming their strategy.
I teach in a business school, and so there is no question of "banning AI" or other such tactics. We want students to use LLMs as they would in their jobs. The problem is that to be good at using LLMs in your job you have to have learned enough to evaluate output and know what to prompt and how to shape that prompt. And that is very hard to ensure in an environment where people can turn in plausible, fluent output without reading it themselves.
The lazy student is fine when there is a professor there to assign cases and readings and offer study questions and even (human) prompts that break the problem down into parts. He just puts the readings into the RAG, turns the assignment questions into prompts, and hits return. The output from this method is still, as of last fall, mediocre MBA boilerplate (often accompanied by LLM errors--I have received many similarly worded student papers asserting that advertisers are on the "subsidy side" of Facebook, for example--oops). But once in the working world, there won't be a professor to pre-select and pre-analyze the problem and the worker bee will have to figure all that out himself. And grade his own work. A student who has followed the lazy path will not be able to execute. THAT is the difference between LLMs and calculators or word processors--these older technologies work just as well without the scaffolding of the instructor's syllabus and organization and selection of texts and questions.
Great article
Just FYI because of the Cornell connection:
I entered 9 years before you. Rooms full of keypunch machines, sometimes with a queue of people waiting for one to free up. Fortunately, I'd learned about card saws (during an internship at IBM) so I often could borrow one from an operator and free up a "down" machine by vigorously going back and forth at the punch station with the saw to get a jammed card out. (The machines were built like tanks.)
You'd hand your punched deck of cards to the operator and wait a while (an hour, sometimes less, sometimes more?) for a printout of your run to come back. No debugger except the one that ran in your head--no "step into", "step over" :-)
Toward the end up my time at Cornell, the Computer Science department acquired a PDP-11 and I got my first exposure to UNIX in ~1977. I still remember my reaction to "ls", "rm", and "grep" as the names of important commands--WTF?
Funny that the Harvard case wasn't about cheating at first. What reached the Dean was other students' anxiety about fairness, and the cheating theories only showed up later, once people had time to imagine what the tool might become. Same ordering as the AI bans now: the ban lands first, the justification catches up after.