The Background
Our conversations at Learning Agent started over the question, "What roles can Generative AI play in learning that no one seems to be talking about?" We've seen the paradigm of a text chat with an all-knowing tutor bot, but what else is possible?
We wondered if instead of kicking off a generic tutoring session, in the vein of "help me with my homework," a chatbot could be prompted to teach you something specific. We quickly felt too limited with text-only, so decided to load up the bot with (pre-made, static) image files it could show you as part of the lesson.
Then there was the issue of how to structure the experience, and we're fans of instructional routines. We tried creating a Worked Example routine as described by Michael Pershan in Teaching Math with Examples. In this routine, the learner studies a problem and its worked-out solution, answers one or more questions about it, and then tries a similar problem on their own.
We also prompted the chatbot to run the routine "Which One Doesn't Belong?" created for math by Christopher Danielson in Which One Doesn't Belong: The Shapes Book with excellent resources curated by Mary Bourassa at wodb.ca. In this routine, you describe a reason why one of four things doesn't belong in the group. Each thing has a reason it doesn't belong, and describing why something doesn't belong uses specific mathematical language.
Beware of Dragons
As educators we feel compelled to share this with a warning label: The chatbot is sitting on GPT4, so especially for math topics, it's not hard to get it to say something wrong or get it confused. Tread carefully with introducing this tool to novice learners---it might be fun with supervision but be wary of needlessly confusing someone. (Though we are curious about the soon-to-be-released models with reasoning capabilities.)
How to Try It
Try learning how to divide fractions with Worked Examples.
Try describing geometric solids with Which One Doesn't Belong?