In 2016, in the pre-transformer times, Georgia Tech professor Ashok Goel gave a prescient TEDx Talk on an AI teaching assistant for his large online Artificial Intelligence course. Students would ask questions to an online forum, and fellow students and TAs would answer these question. Unbeknownst to the students, Ashok had an AI system answer questions from the forum with the moniker Jill Watson (it was built on technology used for Watson's Jeopardy win). Jill answered questions about administrative issues in the class, so well that most students were surprised it was an AI system. Goel still needed human TAs to handle questions about course content.
In the talk Goel made comments ahead of his time.
Today, we have classes with a hundred or more students, and few get personal attention. We even have massive open online courses, where students learn from prerecorded video lessons. Some of these classes attract 100,000 students, but only 10 percent or fewer complete them—because of the lack of teaching assistance and personal attention.
This raises a fundamental issue: can we have personal attention at scale? What if we could give personal attention to every student, when and how they needed it? That would create an educational revolution—because learning and teaching would become personal again...I envision a future where everyone has access to AI teaching assistants like Jill Watson—anytime, anywhere, for any task. A future where education is affordable and accessible to all, yet still personal and fun.
Fast forward to today and AI assistants can answer questions about course content pretty well for nearly every undergraduate course in all disciplines. A well-trained AI tutor outperformed in-class active learning. While I haven't seen formal studies, students seem to go to ChatGPT before reaching out to TAs, instructors or their fellow students. AI can grade assignments and exams.
Teaching assistants still have an important role to play. They teach recitation sections and run student labs. A great TA can provide a human connection with a student that no computer could. A teaching assistantship gives important training to students who might become academics, as well as providing an important funding mechanism, especially in the humanities and social science which has limited grant support for graduate students. But in a time of budget constraints at many institutions, how can we argue for the same TA support at the levels we had back in 2016?
Is it too much to ask: if AI is replacing the TA's, and soon the professors, then do students indeed need to know this? A person can ask if the biggest favor we can do students is by requiring courses in Butlering? (Of course I'm being intentionally provocative. And this is one study, etc. But in a more sober note I'll say that I have not really seen much sense, at least in the teaching side of the profession, of being in the path of an onrushing train.)
ReplyDeleteSaying that "a well-trained AI tutor outperformed in-class active learning" is quite misleading compared to what the linked study actually showed. They did the study for only two weeks of class (so they are only measuring a very limited scope of learning), at Harvard (not representative of the wider population), and without the context of other course activities like homework, exams, readings, outside of class learning, projects, etc. I'd wager that students that use generative AI on most of those other tasks learn significantly worse than those that work through the material themselves or with their peers or course staff.
ReplyDeleteAlso, side note: students seem very against being graded by AI