"I’m aware that I’m rapidly typecasting myself as some kind of zealot against generative AI" Admittedly, even while agreeing with much of what you have written over the past few months, I have become concerned that you have sounded a bit like an old man living in a cave shouting at passersby about the evil of houses. Even when I agreed with you I worried that you were overreaching.
Until this post. You got through to me, finally, and I'm seeing your concern more clearly. I started typing up my reactions to some of your specific examples, but I realize I wouldn't be adding anything to what you said. I guess I'm just here to say, "Keep your message going, keep pushing us." This is where I should post my theory of learning, but I need to think about that one a bit. I'll try to do that in the next few days or so.
Great points. Resonates deeply with https://www.edsurge.com/news/2023-06-09-to-close-the-math-achievement-gap-we-must-recognize-what-students-bring-to-the-classroom, for example "As former teachers and education leaders, we have a duty to give students opportunities to advance their mathematical reasoning in learner-centered classrooms. To provide such opportunities for learner-centered mathematics classrooms, it is important to understand how learning occurs, recognize students' assets and existing understanding, and create awareness of the differences between teachers’ mathematical thinking and students’ mathematical thinking."
In your piece, I particularly liked "A novice who needs Khanmigo by definition does not have access to the knowledge that an expert would use to critique Khamigo." It reminds me of some of the suggested applications of generative AI to writing feedback. For example, https://sites.google.com/stanford.edu/ai-edu-learningresources/format-b#h.xnc4ru232kng. A writer who is having difficulty revising their work is likely going to have a lot of difficulty interpreting ChatGPT's feedback if it involves reading through wordy, abstract, and possibly hallucinogenic referents. It feels like the proposed solution is to solve a language-based challenge with challenging language. Great for those who are already operating at a pretty sophisticated level of written expression; not so great for those who aren't. Made harder still by the lack of auditable results (where is this stuff coming from?). The productization of generative AI for learning isn't there yet.
And to the Benjamin Franklin example: there are always opportunities to learn from generative AI's hallucinations, but unless that's the purpose of the lesson, it feels like the instructional equivalent of spinning a bug as a feature. The hallucinations, while interesting, aren't likely to effectively advance an instructional goal of learning more about Benjamin Franklin.
Love it. Obv yours and mine are cousins and maybe even immediate family! I wonder what you think about "and" vs "or" in this part: "either affirm their ideas or to create alternative ones".
Interesting comment and reflects perhaps the difference between math learning and science learning (me). Math learning is divergent looking for many ideas for patterns whereas science being linked to the material world looks for convergence to ideas we agree are consistent with our experience.
Hey Dan - curious if you've seen/played around with PlayLab (https://www.playlab.ai). It's interesting as a tool that lets educators build their own niche use cases for AI, basically to start developing an understanding of the technology by putting it to work on their understanding of learning.
I've been prettying with it for a few months now, mostly focusing on use cases where the tool doesn't need to be precisely correct but it needs to be provocative and catalytic. (a collection of examples here: https://routinechaos.mmm.page)
I've thought about it a few times in the context of things like 3 act math, and I kind of wonder if it could be used for:
- taking a hook and differentiating based on student capabilities.
- taking a hook and personalizing based on student interests.
I've got a few invites if you're interested in tinkering a bit.
🏆 Students interacting with Benjamin Franklin in Khanmigo. 🏆
"I’m aware that I’m rapidly typecasting myself as some kind of zealot against generative AI" Admittedly, even while agreeing with much of what you have written over the past few months, I have become concerned that you have sounded a bit like an old man living in a cave shouting at passersby about the evil of houses. Even when I agreed with you I worried that you were overreaching.
Until this post. You got through to me, finally, and I'm seeing your concern more clearly. I started typing up my reactions to some of your specific examples, but I realize I wouldn't be adding anything to what you said. I guess I'm just here to say, "Keep your message going, keep pushing us." This is where I should post my theory of learning, but I need to think about that one a bit. I'll try to do that in the next few days or so.
Great points. Resonates deeply with https://www.edsurge.com/news/2023-06-09-to-close-the-math-achievement-gap-we-must-recognize-what-students-bring-to-the-classroom, for example "As former teachers and education leaders, we have a duty to give students opportunities to advance their mathematical reasoning in learner-centered classrooms. To provide such opportunities for learner-centered mathematics classrooms, it is important to understand how learning occurs, recognize students' assets and existing understanding, and create awareness of the differences between teachers’ mathematical thinking and students’ mathematical thinking."
In your piece, I particularly liked "A novice who needs Khanmigo by definition does not have access to the knowledge that an expert would use to critique Khamigo." It reminds me of some of the suggested applications of generative AI to writing feedback. For example, https://sites.google.com/stanford.edu/ai-edu-learningresources/format-b#h.xnc4ru232kng. A writer who is having difficulty revising their work is likely going to have a lot of difficulty interpreting ChatGPT's feedback if it involves reading through wordy, abstract, and possibly hallucinogenic referents. It feels like the proposed solution is to solve a language-based challenge with challenging language. Great for those who are already operating at a pretty sophisticated level of written expression; not so great for those who aren't. Made harder still by the lack of auditable results (where is this stuff coming from?). The productization of generative AI for learning isn't there yet.
And to the Benjamin Franklin example: there are always opportunities to learn from generative AI's hallucinations, but unless that's the purpose of the lesson, it feels like the instructional equivalent of spinning a bug as a feature. The hallucinations, while interesting, aren't likely to effectively advance an instructional goal of learning more about Benjamin Franklin.
My theory of learning
Students enter a classroom with ideas about how the world is based on their experiences
We should value their thinking and provide new experiences and resources to help them either affirm their ideas or to create alternative ones
Love it. Obv yours and mine are cousins and maybe even immediate family! I wonder what you think about "and" vs "or" in this part: "either affirm their ideas or to create alternative ones".
Interesting comment and reflects perhaps the difference between math learning and science learning (me). Math learning is divergent looking for many ideas for patterns whereas science being linked to the material world looks for convergence to ideas we agree are consistent with our experience.
Hey Dan - curious if you've seen/played around with PlayLab (https://www.playlab.ai). It's interesting as a tool that lets educators build their own niche use cases for AI, basically to start developing an understanding of the technology by putting it to work on their understanding of learning.
I've been prettying with it for a few months now, mostly focusing on use cases where the tool doesn't need to be precisely correct but it needs to be provocative and catalytic. (a collection of examples here: https://routinechaos.mmm.page)
I've thought about it a few times in the context of things like 3 act math, and I kind of wonder if it could be used for:
- taking a hook and differentiating based on student capabilities.
- taking a hook and personalizing based on student interests.
I've got a few invites if you're interested in tinkering a bit.
Hook me up, Seth, thanks! ddmeyer@gmail.com
Zealot? No. Obsessed. Maybe. :)
My theory of learning (not just for my students but for myself): You're learning when your brain says, "Hey! I never thought of it that way before."
I think learners need to be curious. From us, they need understanding and a truckload of patience while they formulate their ideas.
Thanks for this great article, Dan. You inspire me to be better. :)