A new paper provides a fascinating economic analysis of how AI disrupts the role, function, and value of universities and offers 10 proposals to help them survive in the age of AI | Edition #324
Dr. Jarovsky I appreciate your insight into AI but have no idea why you like this Potts piece. the 10 items you list in your X post are indecipherable in themselves, and the whole paper follows from an incorrect understanding of universities ("they sell matching and verification."). This fully commoditized view of education has a long history and married itself to ed-tech with the first content management systems and related developments in the early 1990s. It resurged with MOOCs and its back with AI. It excludes (1) the process of learning and teaching; (2) the process of research, collective as well as individual; (3) intellectual goods held in common in intangible forms (4) the disjunction between management and educational processes; (5) knowledge creation and transmission as a community; (6) the personal relationships (teacher-student, teacher-teacher, student-student, frontline staff with all) that are central to the everyday life of any university (or school); (7) the inter-university relations that create knowledge in cooperation (as well as competition) as as social process. It excludes, people, thinking, learning, discovery, knowledge as both private and social processes. Universities don't really have Hart assets, and they certainly aren't alienable from academic life and owned in effect by management. Rey G points out how specific to economics this piece is, and I would add that it's completely enfolded in assumptions rejected by most other social sciences regarding the autonomy of market forces and subordinate institutions that try to signal them. Put in other terms, the economics of education also acknowledges non-pecuniary effects, both private and "external" or social, and their existence makes this "disintermediation" model untenable. The real story of Anglophone universities is that policymakers and senior managers have stripped out the public / social / external ("public good") effects of the model, and in the UK and US in particular they now lurch from funding crisis to funding crisis amidst general social confusion about the university's intellectual purposes. At the heart of this decline lies the new academic precariant, and Potts's vision of "hosted" academics is of a universal gig economy if I am able to understand his writing correctly. People are still paying to go to college because they still in spite of everything find a teacher or two and a group of friends who care about them personally and about the labor of improving their intellectual powers to the point that they can understand the world with enough insight to face its challenges successfully. The university is an intellectual and an existential project. To the extent that universities really do become disintermediating platforms, why would anyone go? why would anyone study for 10 years to get advanced knowledge to develop and transmit in them?
One data point from this week bears directly on the value question. OpenAI's Astra for Law, announced 17 September 2026, was tested on 200 US legal research questions from the private validation set of Vals AI's Legal Research Bench. It passed the overall correctness check on 54.0% of them at the highest reasoning effort, against 38.7% for the general model using web search alone. The retrieval-and-synthesis half of professional training is being automated much faster than the judgment half. My view is that this argues for universities rather than against them, but only for the part of a degree that was always hardest to grade.
Really important topic and useful summary, thank you.
Worth naming upfront: Potts's is writing as an economist, and that gives a different analysis than an anthropologist's perspective would. Knut Sorensen and Sharon Traweek's Questioning Excellence in Academia: A Tale of Two Universities frames a university very differently, closer to lived practice inside institutions than to market mechanisms. Which leads me to translate two of Potts's economic concepts into cultural and historical terms: signal and matching.
Potts's #2, "restore costly signaling mechanisms," quietly concedes something bigger. Signaling theory only works because a signal was never the trait itself, a diploma always stood in for competence rather than being it. But the digital self represented by signal is not the human self, for good or ill. My digital presence is not the holistic human presence I present in a job, definitely not spiritual nor even an academic environment.
Today, AI's/LLM's signal and matching occurs because of modernity, even though universities are not modern in origin. LLMs now read someone's actual output (code, writing, project history) and "verifies" them directly, that's still just a new signal with the same old gap. A digital trace is curated and gameable in the same way a résumé always was, maybe more so, since the tool doing the reading is also the cheapest tool available for producing a convincing fake one.
On matching: though I don't know its nuances, I would think vetting (e.g. power of admissions, grades, graduation, accreditation, grant funding) would not be deprioritized. Additionally, formal matching mechanisms are a twentieth and twenty-first century invention. The university's founding logic, back in eleventh-century Paris, was something closer to co-location plus a scholar's personal, sustained certification of a student. Reading modern market design backward into that origin story is a small anachronism worth naming.
Of course, what I am proposing here is less of a guidance approach for university administrators, because I don't have tenure, and more of a larger societal context like a historian and I prefer to add the experiences and practices of real people in these sometimes-digital environments without risking economic reductionism.
I have many thoughts here: First, I appreciate the approach of deconstructing the business model of higher education. Looking at the pieces, so to speak, offers up some really unique solutions to what we know is an existential reckoning in the early stages.
I've been looking at this problem from the other side of this "marketplace"--the employer side--which offers up some points of alignment and counterargument:
1. Agree that the expense of valuable teaching expertise can be disintermediated through "visiting" status. I've made a similar case in the rise of fractional workers across organizations in transition (startups, larger organizations innovating) who need the expertise before they necessarily have the infrastructure to improve it.
2. The value of apprenticeships is/will be immense and should be baked into the college experience. As we're seeing, entry level roles are being substituted for AI in the workplace and robbing new entrants of much-needed learned experience.
3. While AI can be a "co-instructor" we can't lose more Socratic methods and subject matter than supports judgment, creativity, critical thinking. Not sure how to certify these things beyond a degree, but we also can't lose this form of learning to AI.
When the first paragraph here is about the "economic value" of universities, the academic world is not deep enough in trouble yet.
Dr. Jarovsky I appreciate your insight into AI but have no idea why you like this Potts piece. the 10 items you list in your X post are indecipherable in themselves, and the whole paper follows from an incorrect understanding of universities ("they sell matching and verification."). This fully commoditized view of education has a long history and married itself to ed-tech with the first content management systems and related developments in the early 1990s. It resurged with MOOCs and its back with AI. It excludes (1) the process of learning and teaching; (2) the process of research, collective as well as individual; (3) intellectual goods held in common in intangible forms (4) the disjunction between management and educational processes; (5) knowledge creation and transmission as a community; (6) the personal relationships (teacher-student, teacher-teacher, student-student, frontline staff with all) that are central to the everyday life of any university (or school); (7) the inter-university relations that create knowledge in cooperation (as well as competition) as as social process. It excludes, people, thinking, learning, discovery, knowledge as both private and social processes. Universities don't really have Hart assets, and they certainly aren't alienable from academic life and owned in effect by management. Rey G points out how specific to economics this piece is, and I would add that it's completely enfolded in assumptions rejected by most other social sciences regarding the autonomy of market forces and subordinate institutions that try to signal them. Put in other terms, the economics of education also acknowledges non-pecuniary effects, both private and "external" or social, and their existence makes this "disintermediation" model untenable. The real story of Anglophone universities is that policymakers and senior managers have stripped out the public / social / external ("public good") effects of the model, and in the UK and US in particular they now lurch from funding crisis to funding crisis amidst general social confusion about the university's intellectual purposes. At the heart of this decline lies the new academic precariant, and Potts's vision of "hosted" academics is of a universal gig economy if I am able to understand his writing correctly. People are still paying to go to college because they still in spite of everything find a teacher or two and a group of friends who care about them personally and about the labor of improving their intellectual powers to the point that they can understand the world with enough insight to face its challenges successfully. The university is an intellectual and an existential project. To the extent that universities really do become disintermediating platforms, why would anyone go? why would anyone study for 10 years to get advanced knowledge to develop and transmit in them?
One data point from this week bears directly on the value question. OpenAI's Astra for Law, announced 17 September 2026, was tested on 200 US legal research questions from the private validation set of Vals AI's Legal Research Bench. It passed the overall correctness check on 54.0% of them at the highest reasoning effort, against 38.7% for the general model using web search alone. The retrieval-and-synthesis half of professional training is being automated much faster than the judgment half. My view is that this argues for universities rather than against them, but only for the part of a degree that was always hardest to grade.
Really important topic and useful summary, thank you.
Worth naming upfront: Potts's is writing as an economist, and that gives a different analysis than an anthropologist's perspective would. Knut Sorensen and Sharon Traweek's Questioning Excellence in Academia: A Tale of Two Universities frames a university very differently, closer to lived practice inside institutions than to market mechanisms. Which leads me to translate two of Potts's economic concepts into cultural and historical terms: signal and matching.
Potts's #2, "restore costly signaling mechanisms," quietly concedes something bigger. Signaling theory only works because a signal was never the trait itself, a diploma always stood in for competence rather than being it. But the digital self represented by signal is not the human self, for good or ill. My digital presence is not the holistic human presence I present in a job, definitely not spiritual nor even an academic environment.
Today, AI's/LLM's signal and matching occurs because of modernity, even though universities are not modern in origin. LLMs now read someone's actual output (code, writing, project history) and "verifies" them directly, that's still just a new signal with the same old gap. A digital trace is curated and gameable in the same way a résumé always was, maybe more so, since the tool doing the reading is also the cheapest tool available for producing a convincing fake one.
On matching: though I don't know its nuances, I would think vetting (e.g. power of admissions, grades, graduation, accreditation, grant funding) would not be deprioritized. Additionally, formal matching mechanisms are a twentieth and twenty-first century invention. The university's founding logic, back in eleventh-century Paris, was something closer to co-location plus a scholar's personal, sustained certification of a student. Reading modern market design backward into that origin story is a small anachronism worth naming.
Of course, what I am proposing here is less of a guidance approach for university administrators, because I don't have tenure, and more of a larger societal context like a historian and I prefer to add the experiences and practices of real people in these sometimes-digital environments without risking economic reductionism.
I have many thoughts here: First, I appreciate the approach of deconstructing the business model of higher education. Looking at the pieces, so to speak, offers up some really unique solutions to what we know is an existential reckoning in the early stages.
I've been looking at this problem from the other side of this "marketplace"--the employer side--which offers up some points of alignment and counterargument:
1. Agree that the expense of valuable teaching expertise can be disintermediated through "visiting" status. I've made a similar case in the rise of fractional workers across organizations in transition (startups, larger organizations innovating) who need the expertise before they necessarily have the infrastructure to improve it.
2. The value of apprenticeships is/will be immense and should be baked into the college experience. As we're seeing, entry level roles are being substituted for AI in the workplace and robbing new entrants of much-needed learned experience.
3. While AI can be a "co-instructor" we can't lose more Socratic methods and subject matter than supports judgment, creativity, critical thinking. Not sure how to certify these things beyond a degree, but we also can't lose this form of learning to AI.