Thank you for sharing abundantly! Now can you explain it to me so that I can understand it without having an advanced degree in education, computer science and a doctorate in English. I used to program in COBOL, Fortran and RPG II and III and what i learned was that the we make things more difficult that it has to be. When I mentored new programmers I used the KISS method instead of the way I was taught, using words that were easily understandable instead of the jargon the engineers created. AI can be helpful, yet for the most part it is being used to confuse and manipulate people. Please enlighten me if I'm incorrect.
Thanks for the feedback! I did my best to make the writing accessible without oversimplifying it. I believe some concepts are inherently complex, but more often than not if we put enough effort, they can be made simple enough for anyone to understand. I'm sure there are still tons of ways to improve my writing, and I'll keep working on it. Thanks :)
My pleasure. Thank you for your consideration and taking the time to respond. Have you considered an editor?
I asked AI to help and this is what it provided
Here's a simplified version
"Thanks for the feedback! I'll keep working to make my writing clear and simple. Some ideas are complex, but I believe most things can be explained in a way that's easy to understand. I'll keep improving!"
I suggest this
"Thanks! I'll explain in way that's easier to understand.
I like where you're going here and generally agree. Regarding the myth of the perfect AI teacher - this sounds like the debate around home schooling. As you know there are reasons for home schooling and there are school district home schooling resource centers to help mitigate the drawbacks.
Regarding AI detection, I agree with you in theory for mature students. I think what you're suggestion would work as a college honor code. In public school, my children used a tool called "turn it in". I think there would be a lot of plagiarism with out it. You know what they say, locks are to keep honest people honest. There's also the problem of previous students giving their essays to current students, and students using services which write essays. I would imagine the AI detection problem is tractable in well-defined contexts, for example, a paid-for AI could detect the work of a free AI.
I have more thoughts but I'll keep it short.
One little suggestion: where it says "does my declaration accurately describe..." - perhaps that could be worded, have I included a declaration of how AI was used in my work. I didn't notice but perhaps the AI policy should mandate such a declaration?
Your work is really amazing. I really enjoy the evergreen approach. Thank you for sharing this content for free!
Thanks for all the comments, very thoughtful of you and I sincerely appreciate it. The discussion on plagiarism detection is definitely far from done, I see your reasons and they surely have a lot of weight. This is something I have to think about more deeply. I still believe the downsides of AI detection tools are far greater than the upsides, but it's far from a settled debate for sure.
There is indeed a mandatory declaration in my suggestion of AI policy but of course that's just and example and I think your wording makes perfect sense, I'll reword it accordingly.
The chapter's focus on augmentation over automation feels so critical, especially in the context of education.
I love the phrase "evasion engineering". While acknowledging that this might be valuable skill ("red-teaming" in cybersecurity), but that's not what we're trying to learning in (most) classrooms.
May I suggest a minor terminology improvement? In the 4-step plan for teachers, I don't love the phrase "administrative task" to describe lesson planning and question generation. I think most educators would see those as part of their core work more aptly described maybe as "lesson design" or "instructional planning" ?
I am currently doing a research project on AI tutoring for a Masters degree in AI & Ethics. In my draft intro for my dissertation I have just written:
"In his influential study, Bloom (1984) showed that students who received one-to-one tutoring performed, on average, two standard deviations better than those taught in a classroom. Later research has consistently reinforced the substantial advantages of one-to-one tutoring (Nickow et al 2020). Bloom’s famous '2 sigma problem' is the challenge of achieving the same learning gains as individual tutoring in a scalable and affordable way (Bloom 1984)."
I don't think we should dismiss the potential value of creating AI tutors. But the challenge is ensuring they behave in ways that embody good pedagogy, which is certainly about much more than providing accurate answers to students.
And if we do succeed in creating good AI tutors, that certainly doesn't mean we must reduce the time we spend learning in groups and through social interactions.
I'm familiar with the 2 sigma problem, as any educator is. Part of the hidden face of the 2 sigma problem is these kids perform very well on the already gameable standardized testing system, but in turn perform poorly in soft skills like teamwork and collaboration precisely because they lacked that collective of like minded learners.
I totally agree we shouldn't diminish the value of AI tutors, and I make that case (perhaps too briefly) especially for underserved communities, where lack of a good teacher is the main issue. What I argue against is the very common thought of *replacing* the teacher with an AI tutor. Anything that is augmentation rather than pure automation is fair play in my book.
And of course, as you say, the challenge of creating good AI tutors is huge, even more when we consider current AI models are specifically trained to be servile, and to provide that cognitive offloading that is so destructive in education. But of course we can train better models, it's just very hard, and there are few if any economic incentives to do it.
That's why in the spirit of pragmatism I argue for alternative uses of AI in education that work in today's economic and technological framework, even if I agree we should be on lookout for better frameworks.
I don't see many people calling for AI tutors to replace teachers, but my instinct is that personalised AI tutors / learning companions will become hugely important in education. It's definitely a form of augmented intelligence, and as long as the tools are designed to teach/guide, rather than simply provide easy shortcuts, I don't see why the technology shouldn't become an essential part of the educator's toolkit. I think teachers should encourage students to use technology to learn outside of the classroom, and then they can maximise classroom time to focus on activities that are best suited to group interactions. I see AI tutors as potentially supporting the "flipped classroom" model.
And I think it's easy to underestimate the challenge of providing enough "good teachers" for all groups - not just underserved communities. Even the good teachers can't spend enough 1-1 time with their students. As a personal example, my daughter goes to an excellent independent school in London, but her GCSE maths teacher was very poor at explaining concepts and motivating her to learn. YouTube videos prove very useful for GCSE students, but they aren't personalised or interactive, and they tend to circumnavigate the necessary "struggle" part of learning. A good AI tutor (designed to manifest effective pedagogical principles) could make a huge difference for students who are trying to learn more outside of the classroom.
With your point: "these kids perform very well on the already gameable standardized testing system, but in turn perform poorly in soft skills like teamwork and collaboration precisely because they lacked that collective of like minded learners" - are you saying that students who have 1-1 (human) tutoring or engage in self-motivated study online necessarily suffer from a lack of engagement in collaborative learning? I guess my view is that you can (and should) do both.
These are all really good points. I agree with your nuance, I'll do my best to improve the chapter to soften these ideas.
Sadly, I do see a lot of very vocal people claiming AI tutors can replace teachers. These are not pedagogues, for sure, these are the very tech bros that want to sell us a dystopia of individualistic online learning, gamified to the rotten core. And they are doing it very smartly, convincing policy makers this is the way to go.
Perhaps you're more optimistic than I am, and I sincerely hope you're right and I'm wrong :) but this is a point I cannot stress enough because I see we're dangerously moving towards that dystopia.
That being said, I agree with your points. This discussion is nowhere near done.
Thank you for sharing abundantly! Now can you explain it to me so that I can understand it without having an advanced degree in education, computer science and a doctorate in English. I used to program in COBOL, Fortran and RPG II and III and what i learned was that the we make things more difficult that it has to be. When I mentored new programmers I used the KISS method instead of the way I was taught, using words that were easily understandable instead of the jargon the engineers created. AI can be helpful, yet for the most part it is being used to confuse and manipulate people. Please enlighten me if I'm incorrect.
Thanks for the feedback! I did my best to make the writing accessible without oversimplifying it. I believe some concepts are inherently complex, but more often than not if we put enough effort, they can be made simple enough for anyone to understand. I'm sure there are still tons of ways to improve my writing, and I'll keep working on it. Thanks :)
My pleasure. Thank you for your consideration and taking the time to respond. Have you considered an editor?
I asked AI to help and this is what it provided
Here's a simplified version
"Thanks for the feedback! I'll keep working to make my writing clear and simple. Some ideas are complex, but I believe most things can be explained in a way that's easy to understand. I'll keep improving!"
I suggest this
"Thanks! I'll explain in way that's easier to understand.
Yup! Thanks again :)
Great summary of some wonderful key points about AI use! I agree with so much here.
Great post and better discussion in the comments. Thanks for sharing.
I like where you're going here and generally agree. Regarding the myth of the perfect AI teacher - this sounds like the debate around home schooling. As you know there are reasons for home schooling and there are school district home schooling resource centers to help mitigate the drawbacks.
Regarding AI detection, I agree with you in theory for mature students. I think what you're suggestion would work as a college honor code. In public school, my children used a tool called "turn it in". I think there would be a lot of plagiarism with out it. You know what they say, locks are to keep honest people honest. There's also the problem of previous students giving their essays to current students, and students using services which write essays. I would imagine the AI detection problem is tractable in well-defined contexts, for example, a paid-for AI could detect the work of a free AI.
I have more thoughts but I'll keep it short.
One little suggestion: where it says "does my declaration accurately describe..." - perhaps that could be worded, have I included a declaration of how AI was used in my work. I didn't notice but perhaps the AI policy should mandate such a declaration?
Your work is really amazing. I really enjoy the evergreen approach. Thank you for sharing this content for free!
Thanks for all the comments, very thoughtful of you and I sincerely appreciate it. The discussion on plagiarism detection is definitely far from done, I see your reasons and they surely have a lot of weight. This is something I have to think about more deeply. I still believe the downsides of AI detection tools are far greater than the upsides, but it's far from a settled debate for sure.
There is indeed a mandatory declaration in my suggestion of AI policy but of course that's just and example and I think your wording makes perfect sense, I'll reword it accordingly.
Again huge thanks!
The chapter's focus on augmentation over automation feels so critical, especially in the context of education.
I love the phrase "evasion engineering". While acknowledging that this might be valuable skill ("red-teaming" in cybersecurity), but that's not what we're trying to learning in (most) classrooms.
May I suggest a minor terminology improvement? In the 4-step plan for teachers, I don't love the phrase "administrative task" to describe lesson planning and question generation. I think most educators would see those as part of their core work more aptly described maybe as "lesson design" or "instructional planning" ?
I am currently doing a research project on AI tutoring for a Masters degree in AI & Ethics. In my draft intro for my dissertation I have just written:
"In his influential study, Bloom (1984) showed that students who received one-to-one tutoring performed, on average, two standard deviations better than those taught in a classroom. Later research has consistently reinforced the substantial advantages of one-to-one tutoring (Nickow et al 2020). Bloom’s famous '2 sigma problem' is the challenge of achieving the same learning gains as individual tutoring in a scalable and affordable way (Bloom 1984)."
I don't think we should dismiss the potential value of creating AI tutors. But the challenge is ensuring they behave in ways that embody good pedagogy, which is certainly about much more than providing accurate answers to students.
And if we do succeed in creating good AI tutors, that certainly doesn't mean we must reduce the time we spend learning in groups and through social interactions.
See also Kestin (2025): https://www.nature.com/articles/s41598-025-97652-6
Interested to get your thoughts!
Absolutely agree!
I'm familiar with the 2 sigma problem, as any educator is. Part of the hidden face of the 2 sigma problem is these kids perform very well on the already gameable standardized testing system, but in turn perform poorly in soft skills like teamwork and collaboration precisely because they lacked that collective of like minded learners.
I totally agree we shouldn't diminish the value of AI tutors, and I make that case (perhaps too briefly) especially for underserved communities, where lack of a good teacher is the main issue. What I argue against is the very common thought of *replacing* the teacher with an AI tutor. Anything that is augmentation rather than pure automation is fair play in my book.
And of course, as you say, the challenge of creating good AI tutors is huge, even more when we consider current AI models are specifically trained to be servile, and to provide that cognitive offloading that is so destructive in education. But of course we can train better models, it's just very hard, and there are few if any economic incentives to do it.
That's why in the spirit of pragmatism I argue for alternative uses of AI in education that work in today's economic and technological framework, even if I agree we should be on lookout for better frameworks.
Huge thanks for the comments!
I don't see many people calling for AI tutors to replace teachers, but my instinct is that personalised AI tutors / learning companions will become hugely important in education. It's definitely a form of augmented intelligence, and as long as the tools are designed to teach/guide, rather than simply provide easy shortcuts, I don't see why the technology shouldn't become an essential part of the educator's toolkit. I think teachers should encourage students to use technology to learn outside of the classroom, and then they can maximise classroom time to focus on activities that are best suited to group interactions. I see AI tutors as potentially supporting the "flipped classroom" model.
And I think it's easy to underestimate the challenge of providing enough "good teachers" for all groups - not just underserved communities. Even the good teachers can't spend enough 1-1 time with their students. As a personal example, my daughter goes to an excellent independent school in London, but her GCSE maths teacher was very poor at explaining concepts and motivating her to learn. YouTube videos prove very useful for GCSE students, but they aren't personalised or interactive, and they tend to circumnavigate the necessary "struggle" part of learning. A good AI tutor (designed to manifest effective pedagogical principles) could make a huge difference for students who are trying to learn more outside of the classroom.
With your point: "these kids perform very well on the already gameable standardized testing system, but in turn perform poorly in soft skills like teamwork and collaboration precisely because they lacked that collective of like minded learners" - are you saying that students who have 1-1 (human) tutoring or engage in self-motivated study online necessarily suffer from a lack of engagement in collaborative learning? I guess my view is that you can (and should) do both.
These are all really good points. I agree with your nuance, I'll do my best to improve the chapter to soften these ideas.
Sadly, I do see a lot of very vocal people claiming AI tutors can replace teachers. These are not pedagogues, for sure, these are the very tech bros that want to sell us a dystopia of individualistic online learning, gamified to the rotten core. And they are doing it very smartly, convincing policy makers this is the way to go.
Perhaps you're more optimistic than I am, and I sincerely hope you're right and I'm wrong :) but this is a point I cannot stress enough because I see we're dangerously moving towards that dystopia.
That being said, I agree with your points. This discussion is nowhere near done.