A possible use of an LLM style model of human learning would be as a way to model teaching approaches and optimize them and to quantify human learning constraints.
Similarly to Einstein’s theory on thermodynamics a more mechanistic model would allow quantitative predictions about effectiveness.
In the same way it might only show new insights at the extremes but showing that bad approaches are inferior seems like quite the challenge today so perhaps that is useful.
It’s a neat idea. I think we have already demonstrated which approaches are ineffective to the satisfaction of anyone interested in evidence and I suspect those who reject this evidence would also reject LLM modelling. Imagine the names they would call it.
Learning model experiments could show the same statistics appear as you vary the model parameters. For example this might show a result is statistically better independent of learner capability.
It might also demonstrate the expertise reversal effect.
Predictive processing is not right--for education. Go nuts with increasing the accuracy of your AI model from 97.3% to 97.4%, but it doesn't scale meaningfully onto human education, which is fundamentally social. (If I may be so bold.) The allure of the individual, the psychologistic, the anti-historical, is simply amazing sometimes.
A possible use of an LLM style model of human learning would be as a way to model teaching approaches and optimize them and to quantify human learning constraints.
Similarly to Einstein’s theory on thermodynamics a more mechanistic model would allow quantitative predictions about effectiveness.
In the same way it might only show new insights at the extremes but showing that bad approaches are inferior seems like quite the challenge today so perhaps that is useful.
It’s a neat idea. I think we have already demonstrated which approaches are ineffective to the satisfaction of anyone interested in evidence and I suspect those who reject this evidence would also reject LLM modelling. Imagine the names they would call it.
Yes not going to convince some.
Learning model experiments could show the same statistics appear as you vary the model parameters. For example this might show a result is statistically better independent of learner capability.
It might also demonstrate the expertise reversal effect.
Predictive processing is not right--for education. Go nuts with increasing the accuracy of your AI model from 97.3% to 97.4%, but it doesn't scale meaningfully onto human education, which is fundamentally social. (If I may be so bold.) The allure of the individual, the psychologistic, the anti-historical, is simply amazing sometimes.