It’s kind of tedious making big knowledge graphs.
I wonder if I can use AI to help.
It feels icky to use AI for pure content, especially given the hallucinations problems. But if it can suggest iterative steps to fill in foundational gaps in knowledge, it could be very powerful for someone with a lot of internal drive.
It may also point out wide gaps in knowledge that a curriculum-maker leaves as they take personal notes. Experts can underestimate what beginners don’t yet know, and an outside perspective (even if AI generated) may help.
Though, AI usually regurgitates tutorial and jargon speak. Often, it takes a very empathetic and self-aware human mentor to bring out the best “trigger” in beginners to help them make crucial connections. (side note: I kind of hate how a lot of good words like “crucial” are becoming red flags of generated writing.)
It makes me think that expertise and mastery are like knowledge graphs, where you have leaps you must cross to get from a familiar idea to a novel idea. Gaps that are small are like “oh, that’s obvious, it trivially follows”. Gaps that are big are like “that’s a pretty big stretch and I probably wouldn’t stumble upon it naturally”. In this context, adding new nodes in between far-apart nodes would help scaffold a path for beginners to grasp advanced concepts.
In MathAcademy, this looks like breaking down many, many strategies and visualizations one can use to solve a problem (painstakingly curated by their team of experts).
But for something like a TART curriculum, which I’m envisioning to be curated by individuals, they’ll have to be aware of all of those knowledge gaps to make an effective, public curriculum.
I think a really useful tool would thus be an AI that augments your TART curriculum. If you lay out your notes in a TART graph, all bite-sized and interconnected, an AI can note where connections are sparse and suggest intermediate topics or ideas for the curriculum creator to think about.
In fact, a student/reader might even participate in this process by trying to hone down what they’re stuck on. An AI can then take this feedback and dynamically grow/edit the curriculum graph to include new intermediate topics, triggered by the student in a live setting. Of course, they should be able to distinguish between what nodes were AI-generated and what was carefully curated by the human author(s).
That idea is exciting. A curriculum graph that grows dynamically through user feedback, building the node-bridges directly towards an advanced topic. We just really need to find some method to reduce hallucinations (or at least, the negative effects of hallucinations with some super-efficient review process?)