That sense of being drained by the presence of AI is something I hear a lot when I do workshops with teachers and college faculty and solicit their thoughts and feelings about what I'd been like over the last 12-18 months. The feeling that AI is omnipresent in the classroom is just wearying and that collective, growing exhaustion is worrisome because it's hitting so many people simultaneously. This isn't just a threat for burnout, but something even worse, demoralization.
But I think you've paved your path of hope with your observation that we're recognizing some essential human capacities that cannot (and should not) be given over to automation. I emphasize learning as an "experience" in building a practice because experiences are, and always will be, a way we differ from these AI models. The models can simulate research, but they don't researcher. They can automate text production, but they can't write.
This is why from the first appearance of ChatGPT I've been trying to insist that this is an opportunity, not a threat and the technology can't kill anything worth preserving. (https://biblioracle.substack.com/p/chatgpt-cant-kill-anything-worth) I think I was ahead of some other folks because I'd broken bad on "schooling" years before AI showed up, so for me, the fact that even these earliest models could simulate school artifacts was simply proof that stuff wasn't worth doing in the way we were doing it.
But...students are curious, they do want to know how to learn and do things. They do need help seeing how school fosters those desires, but they aren't entirely unwilling. The challenge is to root what they so in schools in the realm of experiences and then assess those experiences in ways that value the experience rather than just the outcome, which can be outsourced and automated.
The hopeful part for me in reading this is that it seems like more and more people are recognizing the nature of the challenge. Given the nature of the systems we're working within it will not be easy to change what needs changing, but I think we have a strong idea of the kinds of changes that need to be made.
I agree with all of this. It's going to be hard and there will continue to be disagreements and challenges but that's not unique or unusual with almost every other contentious issue when it comes to teaching and learning. But AI has definitely upended the educational model in many respects and has tapped into an emotional place we haven't quite seen before - and none of it is helped by the way in which companies and governments are approaching it. Thanks for the comment.
Always appreciate your keepin’ it real perspective, Stephen. But with respect, doesn’t the history of “the Internet” indicate the folly of trying to guess at what a technology portends for the future? A class on “the Internet” in the year 2000 would have had nothing to say about mobile phones or social media. And even the Big Tech CEOs leading today’s “frontier” AI commercial models can’t agree on the use case — is it chatbots? No, now it’s agents?
Thanks Ben, as always. Of course I agree we can't predict where this is headed - I said as much in the piece. My argument is narrower - when students are already using something daily, showing them what it does and how it works is just good teaching, regardless of where the technology ends up. Looking forward to hashing this out in more detail at some point later this summer!
Another great reflection. The part about content really resonated with me and I don’t know why more people aren’t discussing it. If you follow the anti-AI set, they quietly admit that they have to cover less content in class in order to conduct their teaching as one might have in the 19th century. Typically, these folks are teaching skills-based classes, not knowledge-based ones. But I refuse to deprive my students of the full suite of international relations theories relative to what their forebears received because of a new technology. There has to be a better way than paring down how much knowledge we can convey.
Another excellent and insightful article. Do we have any reason to expect that next year is going to be less exhausting? My prediction, at least based on what I’m seeing locally, is that a lot of schools are going to be pushing full steam ahead into artificial intelligence without any guiding principles or understanding— at either the administrative level or the departmental level— as to what this should look like.
I’m beginning to get a sense as to how to use it to enrich my research, my teaching, and my daily workflow. But I can’t say with any confidence that I know how to use it in the classroom.
Teaching business intelligence online at the college level, I have watched this play out from a different angle than most. My students arrive assuming AI makes the technical skills less necessary. What I have found is the opposite. The students who use AI most effectively are the ones who already know how to ask a precise question and recognize when an answer is wrong. The ones who cannot do that produce confident, plausible, and often deeply flawed outputs they cannot evaluate. The skills you name, curiosity, the ability to sit with a question that does not resolve cleanly, are not soft. In my asynchronous classroom they turn out to be the hardest technical skills of all.
I also noticed I did not get through the same amount of content this year. However, I did shift some of that load to the students who gave unique presentations on more content than I have ever “covered” before.
Another excellent and insightful article. Do we have any reason to expect that next year is going to be less exhausting? My prediction, at least based on what I’m seeing locally, is that a lot of schools are going to be pushing full steam ahead into artificial intelligence without any guiding principles or understanding - at either the administrative level or the departmental level - as to what this should look like.
I’m beginning to get a sense as to how to use it to enrich my research, my teaching and my daily workflow. But I can’t sit with any confidence that I know how to use it in the classroom.
The exhaustion you're describing resonates — but I think it's worth naming what's actually causing it. It's not AI itself. It's that every AI interaction starts from zero. No memory of what you decided last week, what your assessment philosophy is, what your school's constraints are. You're not just using AI — you're re-briefing it constantly, which means the cognitive load never reduces.
The teachers I've seen get genuine relief from AI aren't the ones who found better prompts. They're the ones who built enough persistent context that the AI stops asking them to explain themselves from scratch every session.
That's a solvable problem — though not an easy one, and not one individual teachers should have to solve alone. The institutional gap you're naming is real: there's no shared infrastructure for this. But the exhaustion itself has a specific cause, and naming it correctly might help.
I agree on the exhaustion, and think it points at something structural rather than transitional.
We've been organising education around the wrong unit of analysis. Knowledge and skills look like stable individual possessions, but they aren't—they're relational, distributed across people, tools, and contexts— the cockpit doesn't have knowledge, the cockpit *is* a cognitive system. The exam was always a fiction in this sense, isolating the student from the cognitive system within which they actually function. With AI we appear to have reached the point that this fiction has become untenable.
I've been thinking about this for a while. In 2018 I wrote *To code or not to code: From coding to competence* with Tim Patston at Geelong Grammar, where we tried to frame it in terms of digital competence. We argued that what students need isn't knowledge or skills so much as *predilections*: the attitudes and behaviours that let them engage productively with unfamiliar tools in novel situations.
> “Learned helplessness is a consequence of this. In the roundtables in the previous project, this was known as the ‘beer problem’ as one participant observed 'Most employees can solve practical problems in the workplace, such as needing beer and snacks for Friday afternoon drinks–if I give them fifty dollars they can find beer and snacks. However, if there is a digital technology problem, even a basic problem with their device, they cannot solve the problem unless there is an app for it.'"
You can't teach 'knowing when and why' as a capability because it isn't one. The Khanmigo note is a clean example: students using AI constantly but not the way adults designed it for them, because the predilections that would make the designed use natural aren't there.
The implication for assessment is uncomfortable: if competence is contextual and dispositional, you can only observe it in practice, embedded in real work. Which looks a lot more like apprenticeship than examination. The PhD and the medical residency kept that model because the ensemble nature of the knowledge was too obvious to abstract away. The question is whether we can recover something like it at scale—and whether AI, by expanding what's available between sessions with an expert, changes that calculus at all.
That sense of being drained by the presence of AI is something I hear a lot when I do workshops with teachers and college faculty and solicit their thoughts and feelings about what I'd been like over the last 12-18 months. The feeling that AI is omnipresent in the classroom is just wearying and that collective, growing exhaustion is worrisome because it's hitting so many people simultaneously. This isn't just a threat for burnout, but something even worse, demoralization.
But I think you've paved your path of hope with your observation that we're recognizing some essential human capacities that cannot (and should not) be given over to automation. I emphasize learning as an "experience" in building a practice because experiences are, and always will be, a way we differ from these AI models. The models can simulate research, but they don't researcher. They can automate text production, but they can't write.
This is why from the first appearance of ChatGPT I've been trying to insist that this is an opportunity, not a threat and the technology can't kill anything worth preserving. (https://biblioracle.substack.com/p/chatgpt-cant-kill-anything-worth) I think I was ahead of some other folks because I'd broken bad on "schooling" years before AI showed up, so for me, the fact that even these earliest models could simulate school artifacts was simply proof that stuff wasn't worth doing in the way we were doing it.
But...students are curious, they do want to know how to learn and do things. They do need help seeing how school fosters those desires, but they aren't entirely unwilling. The challenge is to root what they so in schools in the realm of experiences and then assess those experiences in ways that value the experience rather than just the outcome, which can be outsourced and automated.
The hopeful part for me in reading this is that it seems like more and more people are recognizing the nature of the challenge. Given the nature of the systems we're working within it will not be easy to change what needs changing, but I think we have a strong idea of the kinds of changes that need to be made.
I agree with all of this. It's going to be hard and there will continue to be disagreements and challenges but that's not unique or unusual with almost every other contentious issue when it comes to teaching and learning. But AI has definitely upended the educational model in many respects and has tapped into an emotional place we haven't quite seen before - and none of it is helped by the way in which companies and governments are approaching it. Thanks for the comment.
Always appreciate your keepin’ it real perspective, Stephen. But with respect, doesn’t the history of “the Internet” indicate the folly of trying to guess at what a technology portends for the future? A class on “the Internet” in the year 2000 would have had nothing to say about mobile phones or social media. And even the Big Tech CEOs leading today’s “frontier” AI commercial models can’t agree on the use case — is it chatbots? No, now it’s agents?
Thanks Ben, as always. Of course I agree we can't predict where this is headed - I said as much in the piece. My argument is narrower - when students are already using something daily, showing them what it does and how it works is just good teaching, regardless of where the technology ends up. Looking forward to hashing this out in more detail at some point later this summer!
Another great reflection. The part about content really resonated with me and I don’t know why more people aren’t discussing it. If you follow the anti-AI set, they quietly admit that they have to cover less content in class in order to conduct their teaching as one might have in the 19th century. Typically, these folks are teaching skills-based classes, not knowledge-based ones. But I refuse to deprive my students of the full suite of international relations theories relative to what their forebears received because of a new technology. There has to be a better way than paring down how much knowledge we can convey.
Another excellent and insightful article. Do we have any reason to expect that next year is going to be less exhausting? My prediction, at least based on what I’m seeing locally, is that a lot of schools are going to be pushing full steam ahead into artificial intelligence without any guiding principles or understanding— at either the administrative level or the departmental level— as to what this should look like.
I’m beginning to get a sense as to how to use it to enrich my research, my teaching, and my daily workflow. But I can’t say with any confidence that I know how to use it in the classroom.
Agreed. I also think we'll see some schools go in the opposite direction and move to go more tech-free. I'm looking forward to recharging this summer!
Teaching business intelligence online at the college level, I have watched this play out from a different angle than most. My students arrive assuming AI makes the technical skills less necessary. What I have found is the opposite. The students who use AI most effectively are the ones who already know how to ask a precise question and recognize when an answer is wrong. The ones who cannot do that produce confident, plausible, and often deeply flawed outputs they cannot evaluate. The skills you name, curiosity, the ability to sit with a question that does not resolve cleanly, are not soft. In my asynchronous classroom they turn out to be the hardest technical skills of all.
I also noticed I did not get through the same amount of content this year. However, I did shift some of that load to the students who gave unique presentations on more content than I have ever “covered” before.
Another excellent and insightful article. Do we have any reason to expect that next year is going to be less exhausting? My prediction, at least based on what I’m seeing locally, is that a lot of schools are going to be pushing full steam ahead into artificial intelligence without any guiding principles or understanding - at either the administrative level or the departmental level - as to what this should look like.
I’m beginning to get a sense as to how to use it to enrich my research, my teaching and my daily workflow. But I can’t sit with any confidence that I know how to use it in the classroom.
The exhaustion you're describing resonates — but I think it's worth naming what's actually causing it. It's not AI itself. It's that every AI interaction starts from zero. No memory of what you decided last week, what your assessment philosophy is, what your school's constraints are. You're not just using AI — you're re-briefing it constantly, which means the cognitive load never reduces.
The teachers I've seen get genuine relief from AI aren't the ones who found better prompts. They're the ones who built enough persistent context that the AI stops asking them to explain themselves from scratch every session.
That's a solvable problem — though not an easy one, and not one individual teachers should have to solve alone. The institutional gap you're naming is real: there's no shared infrastructure for this. But the exhaustion itself has a specific cause, and naming it correctly might help.
I agree on the exhaustion, and think it points at something structural rather than transitional.
We've been organising education around the wrong unit of analysis. Knowledge and skills look like stable individual possessions, but they aren't—they're relational, distributed across people, tools, and contexts— the cockpit doesn't have knowledge, the cockpit *is* a cognitive system. The exam was always a fiction in this sense, isolating the student from the cognitive system within which they actually function. With AI we appear to have reached the point that this fiction has become untenable.
I've been thinking about this for a while. In 2018 I wrote *To code or not to code: From coding to competence* with Tim Patston at Geelong Grammar, where we tried to frame it in terms of digital competence. We argued that what students need isn't knowledge or skills so much as *predilections*: the attitudes and behaviours that let them engage productively with unfamiliar tools in novel situations.
> “Learned helplessness is a consequence of this. In the roundtables in the previous project, this was known as the ‘beer problem’ as one participant observed 'Most employees can solve practical problems in the workplace, such as needing beer and snacks for Friday afternoon drinks–if I give them fifty dollars they can find beer and snacks. However, if there is a digital technology problem, even a basic problem with their device, they cannot solve the problem unless there is an app for it.'"
You can't teach 'knowing when and why' as a capability because it isn't one. The Khanmigo note is a clean example: students using AI constantly but not the way adults designed it for them, because the predilections that would make the designed use natural aren't there.
The implication for assessment is uncomfortable: if competence is contextual and dispositional, you can only observe it in practice, embedded in real work. Which looks a lot more like apprenticeship than examination. The PhD and the medical residency kept that model because the ensemble nature of the knowledge was too obvious to abstract away. The question is whether we can recover something like it at scale—and whether AI, by expanding what's available between sessions with an expert, changes that calculus at all.
https://www.researchgate.net/publication/327202565_To_code_or_not_to_code_From_coding_to_competence
Also, I found this piece to be way more insightful than the Times one: https://www.thenewcritic.com/p/the-great-zombification