What is left for teachers in the age of AI?
...
Let me know why I am blisteringly wrong here. Or, if I am right, how you are adjusting your curriculum.
Consider this a prediction of both (1) how players in the education space will solve for the equilibrium in a world of very cheap intelligence and (2) advice on what you can do to retool if you are a teacher or a student.
My credentials here are admittedly limited. I am still a PhD student in economics, and I have taught just two college classes this past fall and am teaching two more currently (Spring 2025). This means I have not evolved yet into an excellent teacher, but it also means that I am abundantly conscious about what is and what is not working as I am building my course repertoire. Besides that, I have had fellow new-to-teaching friends ask me what I do and plan to do in the age of AI, so here goes. Oh, and to toot the horn of my profession: it is economics that has produced coherent theory on industrial change.
I will work through what we should expect to happen in each of the three “runs”: the short run, the medium run, and the long run. Technically, the short run is the period of time where there are at least some fixed factors of production, and the long run is the period of time where all factors are variable—there is no technical medium run. Think: the difference between adding another worker and opening a new plant. What I mean by the three runs is separate but related. The short run is the period of time when incumbent economic players fight to increase barriers to entry and preserve their current rents. The medium run is where the shakeout happens: new entrants try a bunch of combinations—new processes, new technologies, new organizational forms—some of which stick and erode the rents of the incumbents. The long run is when the process settles into a new pattern of specialization and trade, the entrants are now the incumbents. These are logical, not chronological categories and are all always happening.
I will also end with what can be done in practice today.
The Short Run
Incumbent teachers—think of them as the people most invested in preserving the current paradigm—are the teachers who have high earnings for being super smart, relaying valuable information to students, synthesizing material, and building institutions and relationships. These people rose to their positions in the classroom, have built out a toolkit and repertoire of classes that they are an expert in; it is no fault of their own, but they are interested in maintaining the status quo.
But now there is a very cheap piece of technology available to students that is very good at being super smart, relaying valuable information, and synthesizing material. Students can use this technology to cheat their way through school, and thus teachers who pour out information are redundant. If it is hard to identify which teachers “just pour out information,” check to see if they ban LLMs on their syllabus. In the most basic economics of industrial organization, successful-in-the past firms try to keep the competition out of the market. Healthy competition, however, happens with price competition, product differentiation, and production improvements.
The first-order predictions here are simple. Teachers closer to retirement will likely ignore it or outright ban it and overall, not change their ways. Without curving, we should expect grade inflation in these classes. Younger teachers will have to confront their system of teaching. The advice part of this is to do it sooner rather than later.
The second-order predictions are also relatively clear. We should not have very short timelines on the upheaval of education systems. Science progresses one funeral at a time; the time-until-funeral should be our floor here for institutional changes, with a much higher ceiling. Given that the older, more entrenched teachers hold more administrative positions on average, they will have a period of time to affect department, school, and eco-system-wide policies that protect the status quo.
The demand side for education is relatively inelastic, partially for signaling reasons. But whatever is left of human capital development, expect it to decrease in quality in the short run. Or rather, the variance will be higher: the students who don’t want to learn anyway will take the teachers who don’t want to adjust (assuming they get a choice), while the students who do want to learn will take the teachers who do want to adjust.
The Medium Run
Young teachers will experience the wave of creative destruction whether they like it or not. Some will be able to play the incumbent’s game and pretend nothing is happening; that is not the winning strategy.
The main dynamic is simply that the rents will flow to the relatively scarce factors. Here we should ask what else teachers are good at besides pouring out information—it is no longer enough to have high IQ, sorry. Showing up in person, motivating, inspiring, entertaining, looking in the eye, facilitating and adjudicating, advising, recommending, motivating, guiding, talking to, and motivating students; these are just some of the possibilities. Did I say motivating? My preferred model here (both what worked for me as a student and what I aspire to) is to show students that the world is bigger and cooler than they thought it was—a motivating carrot of wonder at the universe, if you will. But I do not dismiss the very real and effective stick motivation mechanisms. Social approbation and disapprobation from a superior are a great force.
Most information is self-obtained. The self-motivated students are already a leg up then, even more so in the age of an AI. The marginal value product of a teacher therefore is to motivate self-motivation. (By the way, before AI, internet, books, etc., I would also argue that motivation is the main value of a teacher; just that there were enough rents to go around that other people who also called themselves teachers for having intelligence or information could stick around. Ask yourself which teachers mattered most in your development: surely it wasn’t the ones who had strictly the highest IQ or best memory.)
The predictions here: the teachers who have an edge in inspiration and motivation will rise in status and importance. Whether we see more matching between high-motivating teachers and highly self-motivated students, or whether we see the high-motivating teachers educating the more laggard—that will depend on elasticities on both supply and demand side.
The advice for students is twofold: firstly, take teachers, not classes (that’s always been true). Secondly, ask yourself which areas you are not motivated to learn by yourself, but wish you were the kind of person who could—then submit yourself to the accountability of the people who will motivate you. Often that is a “teacher” in some sense. If you can’t think of any areas where that is true, find someone to submit yourself to that will guide the direction of your wishes. Rinse and repeat if necessary.
The advice for my fellow teachers then is to consciously specialize in motivation. Fortunately for the teachers who have specialized in knowledge accumulation—or who have gotten good at literally anything—they must have been self-motivated at some point and can very easily make the switch to modeling that motivation.
The Long Run
In the long run, our educational institutions will look very different. AI and its downstream products will do most of the narrowly defined information transmission. But I do anticipate the continued use of institutional categories like school, college, and university. These institutions will still hire teachers who directly interact with students, manage some set of AI agents, and, above all, motivate students to engage and direct their own set of AI agents to complete projects. “Tests” will be replaced with project outputs; students will show off their portfolios instead of their transcripts.
But I also anticipate the brand name of the schools, colleges, and universities to continue to matter. Partly because signaling, status, and positional goods will always matter if humans are involved. But also, we should expect a gradient of quality or at least a diversity of specialization among the schools. In a world of very cheap intelligence and information, there will still be differences in tastes, values, and emphasis.
In Practice Today
The best practical skill to have is a good sense of taste in whatever it is that you deeply care about. For me, that is economic arguments of a certain kind: I believe I can detect good and bad ones in a certain subset of economics.
This means in a pre-AI world I am also specialized in constructing such arguments, and I intend to ride that wave for a little bit longer. But I do not expect the construction to be the bottleneck with very cheap intelligence. It is the recognition and taste for certain kinds. This will continue to matter in a post-AI world because AIs are agents in the simplest sense of the principal-agent model. They can take on any beliefs, tastes, and attitudes you like; the scarce input is you, the principal, having beliefs, tastes, and attitudes that are worth the AI taking on.
As teachers in the short and medium runs, this means that we should be training our students to recognize good work—in my case good economics. Right now, I do not care if my students use AI to help with their assignment (and often encourage or require it) if they can identify why the output is “right” enough to rest their grades on it. The immediate question you should be asking is “why couldn’t the students just ask their AI to give help with the assignment and the reasons why it is the right answer and if they should rest their grade on it?” That is the right question. To the extent any class is gameable, those are skills it is teaching: the purpose of a system is what it does. Did I mention that in the long run teachers will be motivating students on how they can manage their own group of AI agents?