“This challenges educational approaches that mistake information transmission for understanding. When we flood students with content without ensuring they can extract underlying principles, we risk creating the educational equivalent of an overfitted model: impressive performance on familiar problems, but brittle failure when contexts shift.”
While Carl’s obviously advancing an argument about the delivery of curricular “content”, I think it has a lot to say about the overstimulation that takes place when saturated by a great volume of shallow information, curricular or extracurricular. If students struggle to learn well when flooded by relevant content, what hope do they have of learning in highly stimulating digital environments?
I think we need to take much more seriously the architecture of our learning spaces - physically, digitally, and temporally.
Amazing article… thanks a lot. A link between compression and education, a link between my thesis and my passion, this article is a precious gift. Never though about it this way, but while reading everything makes so much sense. You have given me thoughts for the next decade.
I remember as a kid, never being able to learn multiplication table. I’ve never been good at memorizing stuff… so I had to find trick, learn to multiply in my head fast enough so that the teacher would not see I was multiplying instead of recalling. I learned a few pattern, then learn to get value around it fast. At the end, it became a challenge on its own… how can I learn something without memorizing, what rules could I found that would allow me to escape memorizing.
That’s probably why I ended up in Signal Processing :-)
I wonder if “generalising” is a good description. But I admit to be on the fence and arguing with myself. I am thinking that wha is happening is a overfitting local cases and obtaining a more rough interpolation surface. “Generalising” leds me intuitively to think on models and meaning, something this technology does not create. But I accept - uncomfortably - that generalising might describe a meaning free simpler interpolation hyper surface … ? Anyway, good piece, thanks!
"Beyond that ceiling, they stop copying examples and begin to extract patterns." Reading Researcher David Share calls this the 'self-teaching hypothesis'.
This probably isn't what you're talking about with regard to a "self-teaching hypothesis," but a lot of homeschool and self-directed education advocates make a good case that children do teach themselves, including how to read, when they're ready.
Maybe part of the problem is that we're making kids do academic stuff before they're developmentally ready for it.
Here's one guy who seems to have written a lot about it:
You’re right that it’s different but your link is useful. I’m doing a piece on ‘academics’ in the lower grades, and his section on ‘self-learning’ might be relevant. Thanks!
"Musil's insight predates information theory by decades, yet captures a contentious argument made by AI advocates: in a world of infinite information, wisdom lies not in accumulation but in compression. The librarian's catalogs are human algorithms, reducing vast complexity to manageable patterns. His expertise emerges not from storing content but from understanding relationships, which is exactly what happens when AI models move from memorisation to" < exactly why I don't need to make meticulous notes
Re: "Musil is arguing that the librarian who reads everything would drown in specifics and lose sight of the whole. But the one who reads nothing except organisational structures, who compresses millions of books into navigable patterns, becomes the most knowledgeable person in the building."
You don't have to read any of it to effectively catalog it, but who wants to live having read nothing?
Re: "Yet the functional parallels are striking enough to warrant attention."
If by "attention," you mean skepticism, sure.
Why Your Brain is Not a Computer, by Matthew Cobb:
"In The Embodied Mind, internationally renowned psychiatrist Dr. Thomas R. Verny sets out to redefine our concept of the mind and consciousness. He brilliantly compiles new research that points to the mind’s ties to every part of the body."
“Limits are not the enemy of learning, they are its engine.” -Carl Hendrick.
I found this quote particularly revealing:
“This challenges educational approaches that mistake information transmission for understanding. When we flood students with content without ensuring they can extract underlying principles, we risk creating the educational equivalent of an overfitted model: impressive performance on familiar problems, but brittle failure when contexts shift.”
While Carl’s obviously advancing an argument about the delivery of curricular “content”, I think it has a lot to say about the overstimulation that takes place when saturated by a great volume of shallow information, curricular or extracurricular. If students struggle to learn well when flooded by relevant content, what hope do they have of learning in highly stimulating digital environments?
I think we need to take much more seriously the architecture of our learning spaces - physically, digitally, and temporally.
I think we make too much of metaphors. A brain is not a computer.
But you may be right about digital environments; they're made to be addictive.
However, if a child can learn in a forest or a garden, they can learn in a classroom.
Amazing article… thanks a lot. A link between compression and education, a link between my thesis and my passion, this article is a precious gift. Never though about it this way, but while reading everything makes so much sense. You have given me thoughts for the next decade.
I remember as a kid, never being able to learn multiplication table. I’ve never been good at memorizing stuff… so I had to find trick, learn to multiply in my head fast enough so that the teacher would not see I was multiplying instead of recalling. I learned a few pattern, then learn to get value around it fast. At the end, it became a challenge on its own… how can I learn something without memorizing, what rules could I found that would allow me to escape memorizing.
That’s probably why I ended up in Signal Processing :-)
Thanks again… enlightening!!!!
Adenda … Borges short story about Funes, the man that could no forget and, as such, was unable to generalize… comes to mind.
I wonder if “generalising” is a good description. But I admit to be on the fence and arguing with myself. I am thinking that wha is happening is a overfitting local cases and obtaining a more rough interpolation surface. “Generalising” leds me intuitively to think on models and meaning, something this technology does not create. But I accept - uncomfortably - that generalising might describe a meaning free simpler interpolation hyper surface … ? Anyway, good piece, thanks!
"Beyond that ceiling, they stop copying examples and begin to extract patterns." Reading Researcher David Share calls this the 'self-teaching hypothesis'.
This probably isn't what you're talking about with regard to a "self-teaching hypothesis," but a lot of homeschool and self-directed education advocates make a good case that children do teach themselves, including how to read, when they're ready.
Maybe part of the problem is that we're making kids do academic stuff before they're developmentally ready for it.
Here's one guy who seems to have written a lot about it:
https://www.petergray.org/s-projects-basic
You’re right that it’s different but your link is useful. I’m doing a piece on ‘academics’ in the lower grades, and his section on ‘self-learning’ might be relevant. Thanks!
"Musil's insight predates information theory by decades, yet captures a contentious argument made by AI advocates: in a world of infinite information, wisdom lies not in accumulation but in compression. The librarian's catalogs are human algorithms, reducing vast complexity to manageable patterns. His expertise emerges not from storing content but from understanding relationships, which is exactly what happens when AI models move from memorisation to" < exactly why I don't need to make meticulous notes
Limits are not the enemy of learning, they are its engine. <<<very helpful
Great article. Had never heard of 'Double Descent' before.
I wonder if (effective) learning strategies that work well for 'under-parameterised' students fail for 'over-parameterized' ones.
Re: "Musil is arguing that the librarian who reads everything would drown in specifics and lose sight of the whole. But the one who reads nothing except organisational structures, who compresses millions of books into navigable patterns, becomes the most knowledgeable person in the building."
You don't have to read any of it to effectively catalog it, but who wants to live having read nothing?
Re: "Yet the functional parallels are striking enough to warrant attention."
If by "attention," you mean skepticism, sure.
Why Your Brain is Not a Computer, by Matthew Cobb:
https://www.theguardian.com/science/2020/feb/27/why-your-brain-is-not-a-computer-neuroscience-neural-networks-consciousness
And...
"In The Embodied Mind, internationally renowned psychiatrist Dr. Thomas R. Verny sets out to redefine our concept of the mind and consciousness. He brilliantly compiles new research that points to the mind’s ties to every part of the body."
https://www.simonandschuster.com/books/The-Embodied-Mind/Thomas-R-Verny/9781639364626