If AI has any role to play in the “learning” aspect of education systems, that is changes to student long-term memory, research is needed into what extent it can replicate what evidence suggests are the best pedagogical/instructional approaches.
Great food for thought. I like the idea of the wisdom exchange. Substack here is a good start on that, and hopefully it evolves further in that direction.
Great article Tom, very provocative. My own research is focused on what is learning, but you can't ask that without asking why. Many thanks for your work on this.
This is a great explanation of why machine learning development feels so different from traditional software engineering. I like that you framed it in terms of experimentation rather than just “training models,” because that really is the core of the work. The reminder that multiple training runs per model are the norm and not the exception also helps explain why scaling ML is so costly. It is the kind of perspective that engineers coming from a pure software background really need to hear.
Thanks Tom, this is very thought-provoking. What mechanisms do you think we need to ensure AI becomes more pedagogical. Do you need the equivalent of educational public health officials to guide and steer AI (and tech in general) since recent tech has probably ignored pedagogy to the detriment of learning!
Relatedly, why is there so much more focus on AI for healthcare in terms of infrastructure (bespoke models, benchmarks etc.)? Notable exceptions are LearnLM and Ai-For-Education pedagogy benchmark. But still it's notable that there is less infra for steering AI to be more pedagogical.
Lastly, the fact that LLMs are text-based might also pose a challenge. We know from Richard Mayer's Multi-Media Learning principles that Images + words are better for learning and the increased use of LLMs is likely to make learning more text-based rather than less.
If AI has any role to play in the “learning” aspect of education systems, that is changes to student long-term memory, research is needed into what extent it can replicate what evidence suggests are the best pedagogical/instructional approaches.
+1, some early work on what it might mean to incorporate pedagogical approaches into AI here https://services.google.com/fh/files/misc/improving-gemini-for-education_v7.pdf
Great food for thought. I like the idea of the wisdom exchange. Substack here is a good start on that, and hopefully it evolves further in that direction.
Great article Tom, very provocative. My own research is focused on what is learning, but you can't ask that without asking why. Many thanks for your work on this.
This is a great explanation of why machine learning development feels so different from traditional software engineering. I like that you framed it in terms of experimentation rather than just “training models,” because that really is the core of the work. The reminder that multiple training runs per model are the norm and not the exception also helps explain why scaling ML is so costly. It is the kind of perspective that engineers coming from a pure software background really need to hear.
https://arcq.ai/
Thanks Tom, this is very thought-provoking. What mechanisms do you think we need to ensure AI becomes more pedagogical. Do you need the equivalent of educational public health officials to guide and steer AI (and tech in general) since recent tech has probably ignored pedagogy to the detriment of learning!
Relatedly, why is there so much more focus on AI for healthcare in terms of infrastructure (bespoke models, benchmarks etc.)? Notable exceptions are LearnLM and Ai-For-Education pedagogy benchmark. But still it's notable that there is less infra for steering AI to be more pedagogical.
Lastly, the fact that LLMs are text-based might also pose a challenge. We know from Richard Mayer's Multi-Media Learning principles that Images + words are better for learning and the increased use of LLMs is likely to make learning more text-based rather than less.
I think we must use AI to somehow faster human learning and apply. Can AI imagine or inspire? I dont think so.