AI written lesson plans part 2
A few months ago I wrote about using GPT3 to create lesson plans automatically… https://medium.com/oermath/ai-written-lesson-plans-d66132052c7b
But there’s more!
Automatic Drawing
While GPT-3 is continuing advanced I recently have spent time playing with ‘craiyon.com’ which uses an open source engine to generate drawings based off text descriptions. While there are already engines that can create assessment items based off of template item descriptions and the like. Seems like it’d be interesting for rapidly generating formative assessment especially if it could be used in conjunction with GPT-3.
Eg, I could ask GPT3 to give me a lesson plan for congruent triangles and with some guidance, images could be automatically generated. As I was writing this, I typed in ‘congruent triangles’ into craiyon and got this:
It’s getting there, but clearly needs some work — and there is a new version of DALL-E out that is much better, but I couldn’t find a public sandbox to play with yet.
So, right now since AI still needs a lot of training data (which I still feel we can accomplish since there are millions of teacher/company created lesson plans/example problem sets out there…), but if we can focus on the type of content that needs to be created, I think we’re getting there
Other ideas
I’ve always found there is something inherently engaging about conversation — how the conversation ebbs and flows, how a comment in a story can create more conversation and change the course of the evening.
Personalized Learning Paths assisted by conversational AI
Learning paths are good, but they’re often somewhat static or teacher/district created, not student-driven. I feel like if students were able to drive more control of the learning path, technology (though linking information expressed in machine-readable competencies information as well as digital curriculum metadata than what we have now) can be used to fill in the blank. That is, if students are given a three act math task like the Taco Cart problem, as they ask questions could a chatbot give sufficient feedback without giving away the answer to help them come to their own solution? (Note:
I will always be in favor of that guidance happening from a real teacher in a classroom in small groups… but engaging in the hypothetical here. Even better, if that information was collected and analyzed, the teacher could know that some students wanted to learn more about specific topics.
Conclusion
The more I think about this, the more I see AI-based interaction creating content, but rather if enough common curricular offerings are tied together (using common CASE identifiers), the better we can analyze student questions to then recommend (something machine learning is very good at) content and curricular paths to take.
