A generous, well-built framework. I'd like to add the frame from where I teach it. In a course on the economics and ethics of sustainable design, VERIFY reads less like a response to AI than like the recovery of something the discipline has always known. The core move is to stop treating thinking as a service the instructor delivers and start treating it as a stock we steward. Once we do, the whole vocabulary of sustainable design transfers with almost no loss.
Fluent AI output is a subsidized good. Its sticker price, a prompt and a few seconds, hides its true cost, which is not eliminated but displaced: onto the student's future self who never built the judgment, onto a profession that inherits practitioners who can produce work they cannot evaluate, onto a public sphere absorbing confident, unverified claims. That is precisely the externality structure sustainable economics exists to expose, and it is why we make designers do full-cost and lifecycle accounting instead of trusting market price. Read this way, VERIFY is not a study skill. It is an internalization mechanism, the cognitive equivalent of making the polluter pay, forcing the deferred cost back to the point of use.
The objection writes itself: a better model removes the friction, so verification becomes waste. Jevons' paradox is the answer. Efficiency in the use of a resource doesn't conserve it; it lowers the effective cost and expands total consumption. Make thinking frictionless and you don't get the same judgment faster… you get more offloading and less judgment exercised. The sustainable answer is never efficiency; it is sufficiency. The friction in VERIFY isn't a transitional inefficiency awaiting a better model. It is a designed-in limit, because the capacity it protects is destroyed precisely by making it frictionless. The ethics are simply Brundtland applied to cognition: meeting present needs, the deliverable, the deadline, the grade, without compromising the future capacity to think. An education that satisfies the present by depleting that capacity is unsustainable in the structural sense, not the sentimental one. Tony Fry would call it defuturing.
So, I'd put the point more strongly than the modest framing allows. This isn't a clever accommodation to a new tool. Sustainable design has always been about exposing who bears the deferred cost and accepting present friction as the price of an intact future. The cognitive case isn't an analogy to that argument. It is the same argument, run on the one resource students are most tempted to treat as free, because it is their own.
This is a good framework and Dr. Jovic is asking the right question. Most educators are still stuck on ban or allow. She moved past that. That matters.
Where I’d push further: a verification checklist gives students a process, but it doesn’t tell you whether the student actually changed. How do you know they internalized the habit versus performed the steps? Without measurement — before and after, scored, tracked — you’re trusting self-report and classroom observation.
I’ve spent over 20 years teaching chemistry and physics across eleven schools. For the last two years I’ve been building a cognitive development program that measures exactly this — whether a student owns their reasoning or outsources it. Five scored dimensions. Pre and post assessment. A six-week cohort where the student has to defend their thinking out loud, not just check a box.
The diagnosis in this article is right. The architecture needs to go deeper. If OpenAI’s education team ever wants pedagogy from a teacher who’s built the implementation layer — not just the awareness layer — I’m easy to find.
Thank you for the comment. Actually, one conference paper and one journal article measuring the impact are about to be published. Once they are, I will share the links.
Marina — I’d love to read them when they’re out. If your research is measuring the impact of structured verification on student cognition, we may be working on parallel problems from different angles. I’m measuring pre/post cognitive ownership across five dimensions in a K-12 cohort right now. Would be glad to compare notes. smalax0172@gmail.com
I also wrote a book about this — Cognitive Sovereignty Under Compression, live on Amazon. If anyone on the OpenAI education team wants to talk to a K-12 teacher who built the measurement layer, I’m at smalaxos.substack.com
This is the AI conversation I wish more schools were having.
Not “ban it” and not “let students use it however they want,” but give them a way to question it.
The line that really stood out to me was that students “may not yet know what to distrust.” That feels exactly right. AI can sound confident when it is wrong, thin, or biased, so students need practice slowing down and checking it.
I also like that VERIFY keeps the student responsible for the work. The best question is not “did AI help?” It’s “what did you decide, challenge, change, and actually contribute?” That’s where the learning is.
Not impressed, Ok, try doing this in middle school and even in the lower grades of high school, where they are learning to think and write. When students are learning to think critically, it is developmental, but in no way can they analyze AI well enough to make its use beneficial. Teach them how to think and write first -- then maybe partner with them at the college level (not freshmen) in this manner. Besides, they should write their own drafts first before ever engaging with AI-- it needs to be their authentic voice.
The framework was tested in ENGL112, Academic Writing 2 (research writing skills), with students who had already completed ENGL100 and ENGL110. It was a classwork-based pilot project.
I never made any claims suggesting success at the middle or primary school level. I leave that area for others to explore.
Erin's instinct seems right to me, and I'd argue the frame I posted formalizes it rather than contradicting it. You cannot internalize a cost you cannot yet perceive. Verification is the act of bearing a cost back at the point of use, but bearing it requires a developed standard to verify against. A writer who hasn't yet built judgment has no stock to steward, so there is nothing for the externality to be forced back onto. Run the protocol too early and it isn't internalization; it's theater. So, the sustainable design reading doesn't universalize VERIFY. It implies exactly the sequencing you're describing: build the capacity first, own drafts, authentic voice, the slow accrual of judgment, then introduce the tool under a sufficiency constraint, once there is something to protect. That's ordinary lifecycle logic. You don't run a structural audit before the structure exists. Read that way, Marina's scope limit (ENGL112, after 100 and 110) isn't only an empirical boundary of the pilot; it's a principled one. The threshold is where the argument lives, not an afterthought to it. Which makes the disagreement smaller than it looks. Both positions hold that capacity must precede the tool. The frame just supplies the reason, and the same reason is why introducing it too early defutures the very judgment it claims to build.
A generous, well-built framework. I'd like to add the frame from where I teach it. In a course on the economics and ethics of sustainable design, VERIFY reads less like a response to AI than like the recovery of something the discipline has always known. The core move is to stop treating thinking as a service the instructor delivers and start treating it as a stock we steward. Once we do, the whole vocabulary of sustainable design transfers with almost no loss.
Fluent AI output is a subsidized good. Its sticker price, a prompt and a few seconds, hides its true cost, which is not eliminated but displaced: onto the student's future self who never built the judgment, onto a profession that inherits practitioners who can produce work they cannot evaluate, onto a public sphere absorbing confident, unverified claims. That is precisely the externality structure sustainable economics exists to expose, and it is why we make designers do full-cost and lifecycle accounting instead of trusting market price. Read this way, VERIFY is not a study skill. It is an internalization mechanism, the cognitive equivalent of making the polluter pay, forcing the deferred cost back to the point of use.
The objection writes itself: a better model removes the friction, so verification becomes waste. Jevons' paradox is the answer. Efficiency in the use of a resource doesn't conserve it; it lowers the effective cost and expands total consumption. Make thinking frictionless and you don't get the same judgment faster… you get more offloading and less judgment exercised. The sustainable answer is never efficiency; it is sufficiency. The friction in VERIFY isn't a transitional inefficiency awaiting a better model. It is a designed-in limit, because the capacity it protects is destroyed precisely by making it frictionless. The ethics are simply Brundtland applied to cognition: meeting present needs, the deliverable, the deadline, the grade, without compromising the future capacity to think. An education that satisfies the present by depleting that capacity is unsustainable in the structural sense, not the sentimental one. Tony Fry would call it defuturing.
So, I'd put the point more strongly than the modest framing allows. This isn't a clever accommodation to a new tool. Sustainable design has always been about exposing who bears the deferred cost and accepting present friction as the price of an intact future. The cognitive case isn't an analogy to that argument. It is the same argument, run on the one resource students are most tempted to treat as free, because it is their own.
This is a good framework and Dr. Jovic is asking the right question. Most educators are still stuck on ban or allow. She moved past that. That matters.
Where I’d push further: a verification checklist gives students a process, but it doesn’t tell you whether the student actually changed. How do you know they internalized the habit versus performed the steps? Without measurement — before and after, scored, tracked — you’re trusting self-report and classroom observation.
I’ve spent over 20 years teaching chemistry and physics across eleven schools. For the last two years I’ve been building a cognitive development program that measures exactly this — whether a student owns their reasoning or outsources it. Five scored dimensions. Pre and post assessment. A six-week cohort where the student has to defend their thinking out loud, not just check a box.
The diagnosis in this article is right. The architecture needs to go deeper. If OpenAI’s education team ever wants pedagogy from a teacher who’s built the implementation layer — not just the awareness layer — I’m easy to find.
Syd Malaxos
Thinking Labs by Temple Academy
smalaxos.substack.com
Dear Syd,
Thank you for the comment. Actually, one conference paper and one journal article measuring the impact are about to be published. Once they are, I will share the links.
Marina — I’d love to read them when they’re out. If your research is measuring the impact of structured verification on student cognition, we may be working on parallel problems from different angles. I’m measuring pre/post cognitive ownership across five dimensions in a K-12 cohort right now. Would be glad to compare notes. smalax0172@gmail.com
I also wrote a book about this — Cognitive Sovereignty Under Compression, live on Amazon. If anyone on the OpenAI education team wants to talk to a K-12 teacher who built the measurement layer, I’m at smalaxos.substack.com
This is the AI conversation I wish more schools were having.
Not “ban it” and not “let students use it however they want,” but give them a way to question it.
The line that really stood out to me was that students “may not yet know what to distrust.” That feels exactly right. AI can sound confident when it is wrong, thin, or biased, so students need practice slowing down and checking it.
I also like that VERIFY keeps the student responsible for the work. The best question is not “did AI help?” It’s “what did you decide, challenge, change, and actually contribute?” That’s where the learning is.
Very valuable framework and activity but there is no substitute to facing a blank page.
Not impressed, Ok, try doing this in middle school and even in the lower grades of high school, where they are learning to think and write. When students are learning to think critically, it is developmental, but in no way can they analyze AI well enough to make its use beneficial. Teach them how to think and write first -- then maybe partner with them at the college level (not freshmen) in this manner. Besides, they should write their own drafts first before ever engaging with AI-- it needs to be their authentic voice.
Dear Erin,
The framework was tested in ENGL112, Academic Writing 2 (research writing skills), with students who had already completed ENGL100 and ENGL110. It was a classwork-based pilot project.
I never made any claims suggesting success at the middle or primary school level. I leave that area for others to explore.
Erin's instinct seems right to me, and I'd argue the frame I posted formalizes it rather than contradicting it. You cannot internalize a cost you cannot yet perceive. Verification is the act of bearing a cost back at the point of use, but bearing it requires a developed standard to verify against. A writer who hasn't yet built judgment has no stock to steward, so there is nothing for the externality to be forced back onto. Run the protocol too early and it isn't internalization; it's theater. So, the sustainable design reading doesn't universalize VERIFY. It implies exactly the sequencing you're describing: build the capacity first, own drafts, authentic voice, the slow accrual of judgment, then introduce the tool under a sufficiency constraint, once there is something to protect. That's ordinary lifecycle logic. You don't run a structural audit before the structure exists. Read that way, Marina's scope limit (ENGL112, after 100 and 110) isn't only an empirical boundary of the pilot; it's a principled one. The threshold is where the argument lives, not an afterthought to it. Which makes the disagreement smaller than it looks. Both positions hold that capacity must precede the tool. The frame just supplies the reason, and the same reason is why introducing it too early defutures the very judgment it claims to build.