What lands most is the distinction between offloading what surrounds the learning and offloading the learning itself. That is the seam many institutions still have not named clearly enough.
The “performance paradox” is especially useful because it explains why so much current evidence feels contradictory: better output, weaker retention; smoother task completion, thinner understanding. And the equity point is the turn of the screw. Students with strong prior knowledge may use AI to accelerate. Students without it may be nudged into a form of false mastery.
Feels to me like this also reaches beyond education. More and more professional settings are now facing the same question: what can be safely delegated, and where is the struggle actually the point?
“Remove the struggle and you remove the learning.” That really is the whole thing. We had a morning with a tangled ball of yarn. I told my daughter, “There’s no way around it, you just have to sit with it and work on it until it’s untangled.”
I love learning about your time in Japan: “The conversations forced the intrinsic work. The discomfort was the learning.” I can only imagine how difficult it would be to learn a language like Japanese, which uses characters, and to have to make conversation.
I would have loved to have had your experience. My family spent time in Japan (before me), so I’ve learned a little about the culture. My mother and grandmother like throwing around Japanese words during games of Scrabble. Not fair! 😂
That’s the quote that stood out most to me too: “Remove the struggle and you remove the learning.” I homeschool my children with a heavy influence from Charlotte Mason. She talks about education as spreading a feast before our children. A textbook chews the food for them and gives them a spoonful of mush to swallow. Maybe it gets the job of a few nutrients done, but they neither had to work for it nor is it delicious. Reading first-hand sources is more difficult, but man, does it stick with us longer and nourish us more deeply!
This is so true, Sarah. Oftentimes I wish, in higher education circles as well, that we went straight to the primary source rather than rely on the textbooks.
Sam, "the struggle is where the learning lives" is close enough to something I've been arguing philosophically that it stopped me cold — I wrote an essay making almost this exact claim, and reading it backed by Bastani's PNAS study and the equity findings from Lodge and Loble feels like finding the empirical floor under something I could only argue from first principles. The Matthew Effect framing is the piece I didn't have: not just that offloading the wrong thing costs you learning, but that the people most at risk of losing it are the ones who could least afford to. Grateful for this — subscribing.
Not trying to overshare my articles with you, but just having seen your about page and your most recent piece, I think you might resonate with our report of a talk we attended by Derek Muller on AI and the future of Education at the Perimeter Institute of Theoretical Physics last April.
AI is just the latest technology that is being inappropriately used. We overuse our cars rather than cycling or walking short distances, and we overuse our mobile phones to the point where (amongst many other undesirable effects) we now have a generation of “smartphone zombies”.
Any young person with the type of very shallow knowledge/understanding that over-reliance on AI produces will get ‘found out’ eventually, but educators are fighting a natural human tendency to take shortcuts at every opportunity.
Literacy in many parts of the world is falling at an alarming rate. In the U.S., many factors other than AI are deemed responsible [1], but AI certainly isn’t helping.
The problems are obvious. The solutions will be difficult to implement.
1. America Is Sliding Toward Illiteracy - Declining standards and low expectations are destroying American education (October 2025).
This article did not disappoint. I came here to understand and you have excellently laid out all the things I have been grappling with.
This morning I finally gave in and hooked up Claude to my Notion in order to expedite task management between three different workflows. In the past, I used to go through my calendars and prioritize which I needed to do, when. As I clicked the "connect" button, I had this eerie sensation of giving away an integral part of my decision making process, but in the back of my head, I also know how much time I spend dilly dallying over what needs to get done, finding the files and then getting "set up" to actually do the task. Claude has been SO efficient with cleaning up my workflows, that a grim sort of terror has set in. How much am I giving away? Do I have a clear boundary?
I pride myself in being critical and tech-agnostic. I loathe the concept of dependency on ANYTHING. It's terrible enough to be a slave to my biological processes and the needs of the people I love. I don't need to be enslaved to technology as well. As a knowledge worker, whose passion and literal job title is facilitating knowledge in other human beings, the blurry boundary between mastering and enslavement is a scary thought. I believe the complacency in thinking that "I'm safe. I'm not like them" can only lead us astray.
Thank you Melissa, and what you are describing is exceptional critical AI literacy. Thinking through what we feel comfortable handing over to AI and why is absolutely essential if we want to be able to use this technology to thrive rather than wither. 🙏
This resonates as a K-12 educator. One question: your Japanese language example may also highlight the power of social learning vs. sitting alone with books or a screen. Have any of these studies looked at differences in learning outcomes between social and solo learning with AI?
From classroom experience, I suspect social learning carries more beneficial cognitive load, partly for the reasons you identified about learning more in conversation with other humans.
Thank you Ian, I want to answer by building on Melissa's answer (see below), as it is excellent! Misinformed peers are indeed destructive to the learning process, especially when we assume that they are better informed than they actual are, which is often the case with AI.
There are several studies that approach social knowledge and learning from a variety of different angles. While collaborative learning has proven to show great results, it's proven that "misinformed peers" are destructive to the learning process if the learner is mid-journey and the learning experience has not been appropriately scaffolded.
This piece confirms everything I have gathered studying AI use in learning and solidifies my stance on it. I am trying to be very intentional with my use of AI in learning new things. When to use it and when not to. The learning comes from the struggle and AI is making it harder to struggle.
To quote the great philosopher of our time, DJ Khaled: "Another one!"
Your point about how AI text being coherent, confident, and articulate reminded me of an interaction I had with AI a while ago. I fed it my article and asked it to go through it. It was pretty early days for me with AI, so maybe it's because my prompting sucked, but it added some bits to my article. The added bits read wonderfully, but there were 2-3 sentences that caught my eye, mainly because I couldn't quite decipher what they meant. I mean, I understood the words, but in the context of my overall article, I didn't think it added anything and was even a little confusing. To me, those sentences felt like it broke the flow of the reading experience too much. So I asked why the sentences were added, and I will never forget its answer: "vague filler." I'm not even making those words up. It literally used them. It gave a longer answer that I can't remember, but I remember those two words. They're seared into my brain. And then, hilariously, it gave me two options for how to improve those sentences with the first option being "delete it."
In hindsight, I'm glad I had this experience because it heightened my skepticism of AI output, seeing how its output can steer me away from what I'm looking to accomplish, which is continual learning in my area of interest (gender equality).
How much of its output is garble that sounds nice but holds little-to-no value, and it goes unchecked? Probably a lot.
Thanks Sam - yes - using my Desktop AI dedicated computer - and a new email too - the verification requires phone & Anthropic system gives an Australian 🇦🇺 phone symbol…
BUT repeated requests get NO phone code arriving? Even on repeated days?
Will let you know if an email request works?? Many thanks! 👍👍
Ps what do you think about Qwen - I think it is an AI that uses an LLM from China?
Another strong piece, Sam.
What lands most is the distinction between offloading what surrounds the learning and offloading the learning itself. That is the seam many institutions still have not named clearly enough.
The “performance paradox” is especially useful because it explains why so much current evidence feels contradictory: better output, weaker retention; smoother task completion, thinner understanding. And the equity point is the turn of the screw. Students with strong prior knowledge may use AI to accelerate. Students without it may be nudged into a form of false mastery.
Feels to me like this also reaches beyond education. More and more professional settings are now facing the same question: what can be safely delegated, and where is the struggle actually the point?
Thank you Stephen. And absolutely this extends beyond education.
“Remove the struggle and you remove the learning.” That really is the whole thing. We had a morning with a tangled ball of yarn. I told my daughter, “There’s no way around it, you just have to sit with it and work on it until it’s untangled.”
I love learning about your time in Japan: “The conversations forced the intrinsic work. The discomfort was the learning.” I can only imagine how difficult it would be to learn a language like Japanese, which uses characters, and to have to make conversation.
Thank you April! I love the yarn story. And Japan was some of my best times. 🥰
I would have loved to have had your experience. My family spent time in Japan (before me), so I’ve learned a little about the culture. My mother and grandmother like throwing around Japanese words during games of Scrabble. Not fair! 😂
That’s the quote that stood out most to me too: “Remove the struggle and you remove the learning.” I homeschool my children with a heavy influence from Charlotte Mason. She talks about education as spreading a feast before our children. A textbook chews the food for them and gives them a spoonful of mush to swallow. Maybe it gets the job of a few nutrients done, but they neither had to work for it nor is it delicious. Reading first-hand sources is more difficult, but man, does it stick with us longer and nourish us more deeply!
This is so true, Sarah. Oftentimes I wish, in higher education circles as well, that we went straight to the primary source rather than rely on the textbooks.
Sam, "the struggle is where the learning lives" is close enough to something I've been arguing philosophically that it stopped me cold — I wrote an essay making almost this exact claim, and reading it backed by Bastani's PNAS study and the equity findings from Lodge and Loble feels like finding the empirical floor under something I could only argue from first principles. The Matthew Effect framing is the piece I didn't have: not just that offloading the wrong thing costs you learning, but that the people most at risk of losing it are the ones who could least afford to. Grateful for this — subscribing.
Thank you angelA. And welcome to the community. 🙏
Thank you for the warm welcome, Sam — glad to be here. Looking forward to reading more.
Grades up / schemas flat is the detrimental-offloading pattern in one line.
Fluency on demand is a metacognitive trap: smooth text ≠ durable encoding. Keep generation + retrieval with the student; let AI critique after.
That’s the product constraint I use for reasoning practice. Founder: https://intelligencemax.ai/guide
Not trying to overshare my articles with you, but just having seen your about page and your most recent piece, I think you might resonate with our report of a talk we attended by Derek Muller on AI and the future of Education at the Perimeter Institute of Theoretical Physics last April.
https://schooloftheunconformed.substack.com/p/learning-fast-and-slow-why-ai-will
Thanks Ruth, reading and liking now. 🙏
AI is just the latest technology that is being inappropriately used. We overuse our cars rather than cycling or walking short distances, and we overuse our mobile phones to the point where (amongst many other undesirable effects) we now have a generation of “smartphone zombies”.
Any young person with the type of very shallow knowledge/understanding that over-reliance on AI produces will get ‘found out’ eventually, but educators are fighting a natural human tendency to take shortcuts at every opportunity.
Literacy in many parts of the world is falling at an alarming rate. In the U.S., many factors other than AI are deemed responsible [1], but AI certainly isn’t helping.
The problems are obvious. The solutions will be difficult to implement.
1. America Is Sliding Toward Illiteracy - Declining standards and low expectations are destroying American education (October 2025).
https://www.theatlantic.com/ideas/archive/2025/10/education-decline-low-expectations/684526/
Thanks Philip. This is a fantastic article. 🙏
This article did not disappoint. I came here to understand and you have excellently laid out all the things I have been grappling with.
This morning I finally gave in and hooked up Claude to my Notion in order to expedite task management between three different workflows. In the past, I used to go through my calendars and prioritize which I needed to do, when. As I clicked the "connect" button, I had this eerie sensation of giving away an integral part of my decision making process, but in the back of my head, I also know how much time I spend dilly dallying over what needs to get done, finding the files and then getting "set up" to actually do the task. Claude has been SO efficient with cleaning up my workflows, that a grim sort of terror has set in. How much am I giving away? Do I have a clear boundary?
I pride myself in being critical and tech-agnostic. I loathe the concept of dependency on ANYTHING. It's terrible enough to be a slave to my biological processes and the needs of the people I love. I don't need to be enslaved to technology as well. As a knowledge worker, whose passion and literal job title is facilitating knowledge in other human beings, the blurry boundary between mastering and enslavement is a scary thought. I believe the complacency in thinking that "I'm safe. I'm not like them" can only lead us astray.
Thank you Melissa, and what you are describing is exceptional critical AI literacy. Thinking through what we feel comfortable handing over to AI and why is absolutely essential if we want to be able to use this technology to thrive rather than wither. 🙏
That will backfire badly
I fear you are right Farida. 😢
This resonates as a K-12 educator. One question: your Japanese language example may also highlight the power of social learning vs. sitting alone with books or a screen. Have any of these studies looked at differences in learning outcomes between social and solo learning with AI?
From classroom experience, I suspect social learning carries more beneficial cognitive load, partly for the reasons you identified about learning more in conversation with other humans.
Thank you Ian, I want to answer by building on Melissa's answer (see below), as it is excellent! Misinformed peers are indeed destructive to the learning process, especially when we assume that they are better informed than they actual are, which is often the case with AI.
Makes a lot of sense. I'll dig into this more. Thank you both!
There are several studies that approach social knowledge and learning from a variety of different angles. While collaborative learning has proven to show great results, it's proven that "misinformed peers" are destructive to the learning process if the learner is mid-journey and the learning experience has not been appropriately scaffolded.
This piece confirms everything I have gathered studying AI use in learning and solidifies my stance on it. I am trying to be very intentional with my use of AI in learning new things. When to use it and when not to. The learning comes from the struggle and AI is making it harder to struggle.
Exactly this. 💪
To quote the great philosopher of our time, DJ Khaled: "Another one!"
Your point about how AI text being coherent, confident, and articulate reminded me of an interaction I had with AI a while ago. I fed it my article and asked it to go through it. It was pretty early days for me with AI, so maybe it's because my prompting sucked, but it added some bits to my article. The added bits read wonderfully, but there were 2-3 sentences that caught my eye, mainly because I couldn't quite decipher what they meant. I mean, I understood the words, but in the context of my overall article, I didn't think it added anything and was even a little confusing. To me, those sentences felt like it broke the flow of the reading experience too much. So I asked why the sentences were added, and I will never forget its answer: "vague filler." I'm not even making those words up. It literally used them. It gave a longer answer that I can't remember, but I remember those two words. They're seared into my brain. And then, hilariously, it gave me two options for how to improve those sentences with the first option being "delete it."
In hindsight, I'm glad I had this experience because it heightened my skepticism of AI output, seeing how its output can steer me away from what I'm looking to accomplish, which is continual learning in my area of interest (gender equality).
How much of its output is garble that sounds nice but holds little-to-no value, and it goes unchecked? Probably a lot.
Another great piece, Sam.
Thank you, Mike. This is excellent. Also, vague filler doubles up as both the perfect description of AI and also a 90s indie band name.
I absolutely would've bought Vague Filler's CD and listened to it from start to finish, reading the lyrics from the insert the whole way through.
Are we the same person? 🤓😎🔥
Loved this - ONLY based on your title! 👍👍♥️
I will read tomorrow - maybe - Dr Sam - you can tell me why Claude won’t send me a Verify message to a valid Australian 🇦🇺 mobile phone number?
Just had my first interaction with Gemini & learned SO MUCH??
Almost 9.30pm here & almost ready to sleep - hope to hear from you in the morning (MY time)! 🤣😂👍👍
Thanks Gail. And honestly I have no idea. Have you tried messaging Anthropic. Was this to try out using Claude on desktop?
Thanks Sam - yes - using my Desktop AI dedicated computer - and a new email too - the verification requires phone & Anthropic system gives an Australian 🇦🇺 phone symbol…
BUT repeated requests get NO phone code arriving? Even on repeated days?
Will let you know if an email request works?? Many thanks! 👍👍
Ps what do you think about Qwen - I think it is an AI that uses an LLM from China?
Thanks Gail. I have only used Qwen a little but have some thoughts here:
https://theslowai.substack.com/p/the-models-you-can-inspect-are-the
https://theslowai.substack.com/p/the-models-you-can-inspect-are-the