Imagine a large online lecture class taught by a tenured professor with the assistance of AI tutors. The professor writes and delivers her lectures once; they are infinitely re-playable as long as the course content does not change. Meanwhile, AI agents can design assessments, deliver course material beyond lectures, hold students’ hands through the completion of the assignments, and even evaluate students’ work, eliminating the need for human grading. The professor can simply review the AI’s work and enter the grades, though even this is essentially a formality—there’s nothing stopping the AI agent from entering the grades with the registrar directly. What I have just described is a class that costs almost nothing in labor, especially after the professor recorded the lectures in a previous semester. A really ambitious university might assign a single grad student or teaching assistant to the class just to oversee the automated processes, but even this isn’t strictly necessary. The machinery runs on its own.
If you, like me, are attached to the old model of education, which presupposes that teachers, in delivering instruction, enter into personal relationships with students, you might find this scenario depressing. After all, if no one is really teaching, is anyone really learning? If, on the other hand, you are a college administrator desperately trying to align your instructional obligations to your budgetary limitations, you might find such a model enticing. As a student, you might find it alienating, but then again, if you are already used to Zoom school and distance learning, it might not feel all that different. And perhaps the course doesn’t matter to you all that much. You’re just trying to get a requirement or a prerequisite out of the way. This class is not going to change your life.
Much of the discourse around generative AI in education revolves, fairly or unfairly, around the ways that students can use the technology to do their work for them. AI is frequently connected to what commentators have called a “literacy crisis”—a longstanding decline in Americans’ levels of reading fluency affecting adults as well as children.11xAndrew Roberts and Felicia C. Smith, “Can You Read This? US Low-literacy Crisis Is Costing Us $2 trillion per Year,” The Hill, September 4, 2025; https://thehill.com/opinion/education/5484190-literacy-crisis-national-literacy-month/. While some early childhood educators and policy experts are optimistic about the potential for AI tools to improve childhood reading acquisition alongside traditional classroom instruction and reading with adults at home, many college instructors have been more skeptical, suggesting that students who rely on AI tools lack the motivation to absorb written texts and do not build the skills to understand them effectively.22xDenver Public Schools, for example, has adopted AI-enabled technologies to teach reading. See Gordon Ebanks, “America Is in a Literacy Crisis. Is AI the Solution or Part of the Problem?” CNN Business, October 16, 2025; https://www.cnn.com/2025/09/22/tech/america-literacy-ai-schools; Philosophy professor Alexandra Berg, writing in the New York Times, however, laments the ways that AI negatively affects students’ “cognitive fluency,” harming not only their capacity to understand longer texts but their ability to think at all. Anastasia Berg, “Why Even Basic AI Use Is So Bad for Students,” New York Times, October 29, 2025; https://www.nytimes.com/2025/10/29/opinion/ai-students-thinking-school-reading.html. But if you listen to its boosters in academia and industry, AI is much more than a cognitive shortcut or a text-generating machine: It is a technology that is rapidly threatening to transform the entire economy, as companies make massive bets on its potential to increase productivity and even solve previously insoluble problems such as cancer and climate change.33xTo choose just one example, JPMorgan Chase CEO Jamie Dimon claimed, without evidence, that AI would “cure cancer” in a 2026 interview. “Jamie Dimon Says AI Will Shorten Work Week to 3.5 days, Cure Cancer,” CBS News, March 31, 2026; https://www.cbsnews.com/video/jamie-dimon-says-ai-will-shorten-work-week-to-3-5-days-cure-cancer/. Meanwhile, a 2026 report by clean energy analyst Ketan Joshi found that, while tech companies tout the climate benefits of AI technology, they conflate traditional AI (machine learning) with generative AI, which uses much more energy and has not led to “material, verifiable and substantial level of emissions reductions.” Ketan Joshi, “The AI Climate Hoax: Beyond the Curtain of How Big Tech Greenwashes Impacts,” Beyond Fossil Fuels, Climate Action Against Disinformation (CAAD), Friends of the Earth U.S., Green Screen Coalition, Green Web Foundation, and Stand.earth, February 2026. It could even amount to a new industrial revolution.44xGeoffrey Hinton, “AI Is the Next Industrial Revolution,” Time, December 11, 2025; https://time.com/7339628/geoffrey-hinton-ai/. So far, the promises of AI have been slow to materialize: Although the technology offers applications for tasks like coding, there remains an abiding sense of confusion about whether—and how—to integrate AI products into other lines of work. In fact, among the general public, AI skepticism abounds.55xAccording to a recent NBC news poll, a majority of Americans (57 percent) say that the risks of widespread AI adoption outweigh the benefits, and AI is viewed unfavorably by 46 percent of Americans, compared to only 26 percent who view it favorably. Allan Smith, “Majority of Voters Say Risks of AI Outweigh the Benefits,” NBCNews.com; https://www.nbcnews.com/politics/politics-news/poll-majority-voters-say-risks-ai-outweigh-benefits-rcna262196. This probably has something to do with the fact that AI boosters tell us, repeatedly, that AI is going to put many of us out of work.
When our conversations about AI in college remain tethered to the idea that students might use it to cheat, we miss the larger context in which executives view AI, first and foremost, as a labor-saving (and labor-disciplining) device. Many university administrators have sent the message that learning to live with AI is not optional: The University of Pennsylvania, for example, declared April 2026 “AI Month.” Still, many instructors—particularly those in the humanities and social sciences—have taken an extremely skeptical line, banning all use of AI and even, in some cases, resorting to in-class assignments and handwritten essays. Grit Matthias Phelps, a senior lecturer in German at Cornell, has even brought typewriters back into the classroom as a way to combat AI use. Phelps and others are responding to a real change in how students work: A 2026 Gallup study found that a majority (57 percent) of college students use AI tools for coursework on a weekly basis, with 21 percent reporting daily use.66xStephanie Marken, “AI is Routine for College Students, Despite Campus Limits,” Gallup, April 2, 2026. Marken’s article summarizes the findings of the Lumina Foundation–Gallup 2026 “State of Higher Education” study, https://www.gallup.com/file/analytics/704279/Lumina-Foundation-Gallup-SOHE_AI_Report.pdf
There are, of course, those who view AI as a new, and necessary, supplement to the college curriculum. A policy page titled “Benefits of AI,” hosted by the Office of the Provost of Washington State University, explains, “By integrating AI systems such as ChatGPT into the university environment, we can reshape the educational and research landscape, fostering a more personalized and dynamic experience for both students and faculty.”77x“Benefits of AI,” Office of the Provost & Executive Vice President, Washington State University; https://provost.wsu.edu/policies/artificial_intelligence/benefits-of-ai/. And that experience is, crucially, scalable: If effective, it could work just as well in large classes as in small ones, transforming learning outcomes even without an increase in faculty-to-student ratios. Others point out that if AI adoption transforms the workplace as its boosters predict, then training on these tools—developing an understanding of how they work, as well as what they are good and not so good at—will quickly become an essential part of employers’ expected workplace skill sets.88xThe Kogod School of Business at American University, for example, promises that “Kogod’s education is designed to turn this [AI preparedness] gap into an opportunity—giving all students the skills and confidence to thrive in an AI economy”; https://kogod.american.edu/news/ai-and-career-readiness-meeting-the-moment-in-higher-education. There is, too, the question of research, specifically whether AI will be useful not only to students but also to PhD-level researchers pursuing projects across a range of disciplines. In some academic fields, the use of AI is already widely accepted, while in others it remains taboo (business schools are among the most enthusiastic adopters).99xLindsay Ellis, “Business Schools Are Going All In on AI,” Wall Street Journal, April 3, 2024. When discussing research implications, it is worth remembering that—as Ketan Joshi points out—there is a difference between traditional machine-learning tools that power some academic research and newer generative AI programs.
The AI Adoption Gap
It’s easy to understand why students might want to use AI to make their coursework more manageable: Cheating, cribbing, and other shortcuts are as old as education itself. What demands an explanation, I think, is the rush among many administrators to aggressively adopt AI for everything from university business to instruction. AI in the classroom is only one use case: As Jake Burley of the University of Massachusetts Boston’s Applied Ethics Center shows, AI use in universities spans three categories: “non-autonomous systems,” which perform discrete tasks under human direction, and which mostly involve back-end technology for tasks like “purchasing, registration, and scheduling”; “hybrid systems,” which include “AI-assisted tutoring, personalized feedback tools, automated writing support, and chatbots”; and “autonomous agents,” an emerging category of AI tech, that “have the potential to act as research collaborators, teaching assistants, or even principal investigators in scientific settings” and which “operate without moment-to-moment human intervention.”1010xJake Burley, “AI Agents in Higher Education 2025,” Institute for Ethics and Emerging Technologies, 4; https://www.umb.edu/media/umassboston/editor-uploads/applied-ethics-center/docments/AI-Agents-in-Higher-Education-(3).pdf. Each type of technology poses its own challenges and ethical risks (as well as its own potential benefits), and with the increasing autonomy of systems, the author argues, those risks grow.
The gap between student and faculty use of AI reflects a sense among faculty that AI is a threat to the entire business model and could make old-fashioned teaching and learning much more difficult to achieve, if not obsolete. But what if that’s the whole point? The relative enthusiasm for AI adoption among administrators bespeaks a growing concern about the practical side of the business model: Are colleges offering students the kind of education they need? Or, at least, the kind of education they want? And are they doing it in a way that is sustainable and affordable?
The answer to these questions, so far, seems to be no. For decades now, the rationale for attending college has been the human-capital theory: the idea that, by leveling up their skills and competencies through education and training, workers can command more opportunities and better compensation, and achieve class mobility.1111xFor a discussion of human-capital theory in education, see Cristina Viviana Groeger, The Education Trap: Schools and the Remaking of Inequality in Boston (Cambridge, MA: Harvard University Press, 2021), 6–8. The college wage premium—the earnings benefit conveyed by possessing a college degree—on average, remains powerful, suggesting that (with some caveats about kind and quality of institution) pursuing a college education is still a good financial choice.1212xSee this report from the Federal Reserve Bank of New York, for example: https://libertystreeteconomics.newyorkfed.org/2025/04/is-college-still-worth-it/.
The problem is that the human-capital theory is now under strain, if not totally discredited. As the researchers Marshall Steinbaum and Andrew Elrod write in a recent report, “Research on intergenerational determinants of economic status finds that the single most important predictor, by a wide margin, is the economic status of your parents.”1313xMarshall Steinbaum and Andrew Elrod, “Rebuilding American Higher Education: From an Engine of Inequality to a Pillar of the Public Interest,” Higher Education Labor United and the Jain Family Institute, December 3, 2025, 19; https://marshallsteinbaum.org/wp-content/uploads/2025/12/Rebuilding-American-Higher-Education_FINAL.pdf. They go on to observe, citing new work by the economists Zachary Bleemer and Sarah Quincy, that
the college earnings premium has become significantly more unequal since 1960, meaning that the earnings differences between people who do and don’t go to college are smaller for the children of less-well-off parents and have increased substantially for the children of the elite.… Precisely as that lower-prestige stratum of the higher education system has oriented itself exclusively to “workforce development,” its actual value in the labor market to the students it educates…has fallen off a cliff.1414xIbid., 18; see also Zachary Bleemer and Sarah Quincy, “Changes in the College Mobility Pipeline Since 1900,” Working Paper no. 33797, Working Paper Series, National Bureau of Economic Research, May 2025; https://doi.org/10.3386/w33797.
In other words, college students who are born to wealthy parents tend to attend the kinds of colleges whose degrees confer substantial economic benefits, confirming their economic status rather than changing it, while students from less-affluent backgrounds seek educations that do not translate into economic security. As Steinbaum and Elrod show, this is not merely a function of the quality of education provided at different tiers, but, even more, a function of the highly stratified nature of higher education itself. Prestige (and price) function as a stand-in for quality, conferring a credential that signals that a graduate possesses both strong potential and demonstrated skills. Your ability to acquire human capital through education turns out to depend, in large part, on the circumstances you were born into. We might understand Americans’ declining faith in college education as a symptom of our country’s extreme economic inequality, and as a sign that the country no longer sees education as much of a bulwark against it.1515xAs measured by Gallup, public trust in higher education among Americans declined for almost a decade, from 2015 to 2024. Although the number of Americans expressing trust in higher education saw a modest recovery—from 36 percent to 42 percent in 2025—it remains well below its initial measure of 57 percent in 2015. See “US Public Trust in Higher Ed Rises From Recent Low,” Gallup, July 16, 2025; https://news.gallup.com/poll/692519/public-trust-higher-rises-recent-low.aspx. In other words, as the historian Cristina Viviana Groeger has argued, education is not in itself a cure for economic inequality and can even help exacerbate it.1616xGroeger, The Education Trap: Schools and the Remaking of Inequality in Boston, 255–56.
All of this portends yet another split in the structure of higher education: Sooner or later, elite schools are going to realize that an analog education carries more prestige than an AI-assisted one and will retool their curricula and expectations to reflect that prejudice. This has largely already happened with screens in elementary and secondary education: Private schools and well-heeled public districts offer screen-free or minimal-screen experiences to children, while resource-starved school districts maximize screen time, in part to stretch limited resources further. The prevalence of technology in education is a class marker, but perhaps not in the direction that anyone could have guessed.1717xNellie Bowles, “The Digital Gap Between Rich and Poor Kids Is Not What We Expected,” The New York Times, October 26, 2018.
When I was growing up, it was just the opposite: Schools competed to offer more robust technological educations. When I was in middle school, at the turn of the millennium, my school—an experimental New York City public school—was one of the first to issue each student a laptop. They were light blue, transparent Apple iBooks, designed specifically as affordable computers for educational settings. The fact that we had been given this extraordinary resource served as proof that our school was taking seriously its responsibility to prepare us for a future in which computing skills were going to be required in any line of work. There was some truth to that, but as technology became increasingly integrated into every facet of life, that kind of formal training in its use became unnecessary.
At the university level, some administrators at the most prestigious schools are beginning to see the writing on the wall about AI’s effect on education. At a recent panel discussion at Stanford (Stanford!), for example, former president John Hennessy, now chairman of the board of Alphabet (Alphabet!) said (as summarized by journalist Alex Kekauoha): “Many students who rely heavily on AI tools struggle with critical thinking and basic math while falsely believing they’ve mastered the material.”1818xAlex Kekauoha, “College Leaders Reflect on the Future of Higher Education,” Stanford Report, March 10, 2026; https://news.stanford.edu/stories/2026/03/siepr-economic-summit-future-higher-education Others agreed, pointing out that major employers, like the management consulting firm McKinsey & Company, have begun seeking out humanities majors for their “critical thinking skills.” An elite consensus that AI-assisted learning might be fine for hoi polloi but not for future leaders at the most selective institutions cannot be far-off now.
Why the Rush to Embrace AI?
For all that, the question remains: Why have so many colleges and universities embraced AI, even as educators at the top of the wealth and prestige hierarchy have begun to question it? We might find a clue in Peril and Promise: College Leadership in Turbulent Times, a recent book by former Spelman and Mount Holyoke President Beverly Daniel Tatum, who argues that the existing higher-education model is ripe for disruption, and that AI will be a major driver of change, thanks to education-specific AI tools such as Khan Academy’s “Khanmigo,” which promises to automate much of the person-to-person work of teaching. Much of the appeal of AI for Tatum comes from its promise to reduce labor costs:
While such sophisticated uses of GAI [generative artificial intelligence] may not eliminate the need for the primary classroom instructor, it certainly could reduce the need for instructional support positions like teaching assistants and subject matter tutors. Can AI reduce the cost of education and improve instruction in meaningful ways?1919xBeverly Daniel Tatum, Peril and Promise: College Leadership in Turbulent Times (New York, NY: Basic Books, 2025), 287.
Tatum’s enthusiasm about saving money on labor derives from her belief that a main driver of college costs is something that economists call the “cost disease” of personal services. The underlying theory, introduced by the economists William Baumol and William Bowen in the 1960s, offers a way to explain the rising costs of health care, education, and the performing arts, alongside other services, in advanced capitalist societies.
The Theory of Cost Disease
Understanding the economic impact of technology is crucial to grasping cost disease. Technological innovation allows for ever-rising productivity, particularly in the production sector, allowing smaller numbers of workers to make ever-greater quantities of goods, in part through automation. As productivity rises, so do wages, reflecting each individual worker’s greater value to the economy. Rising wages also allow for greater consumption, generating demand. But workers engaged in personal services do not see the same productivity growth that workers in other sectors of the economy do, because their jobs are resistant to automation. A hairdresser cannot give more haircuts today per hour than in the 1960s, for example; nor can a personal-care attendant take on more patients without giving less attention to each. For Baumol and Bowen, the effect is especially pronounced in fields where workers are highly educated, since such workers already command higher wages than unskilled workers, and so the resistance to productivity gains in their sector makes costs increase even faster.
It’s certainly true that Baumol effects, as economists call them, contribute to rising costs in health care and education. But the persistent deployment of Baumol’s theory to explain the rising price of higher education has an ideological component as well. If we really believe that the primary driver of price increases is labor costs, then we cannot make college more affordable and pay instructors fairly. It’s a convenient dilemma for administrators seeking the upper hand in increasingly pitched battles with academic labor, which has organized massively over the past several years. As Steinbaum and Elrod report, 31 percent of newly unionized private-sector workers in 2023 were employees of an institution of higher education, while 38 percent of all graduate-student employees in the United States are now unionized. The years between 2018 and 2023 saw at least forty-nine higher-education strikes, eighteen of them by faculty and another thirty-one by graduate students.2020xSteinbaum and Elrod, “Rebuilding American Higher Education,” 36.
The wave of labor organizing that has swept across higher education reflects neither the sector’s uniquely generous compensation nor uniquely good working conditions but the opposite. University administrations have chosen, over the past several decades, to degrade the quality of academic employment, replacing full-time, tenure-track appointments with casual, occasional, and contract labor, and offloading more of the responsibility for student instruction to graduate students and other trainees. Administrations cite shifts in enrollment from year to year, and even from semester to semester, as a rationale for introducing what amounts to a just-in-time staffing system designed to perfectly calibrate instruction to demand.2121xFlexibility for universities can mean income instability for instructors. Even at Georgetown University, for example, where adjuncts enjoy the benefits of a union contract, assigned classes can be canceled up to twenty-one days in advance without any payment due to the assigned instructor. If a course is canceled within three weeks of the start date, adjunct instructors receive 15 percent of the fee. The minimum adjunct rate for teaching a three-credit course at Georgetown is $7,000; recently a new round of contract negotiations began at Georgetown, see https://thehoya.com/news/adjunct-faculty-begin-contract-negotiations/. See also Adjunct Teaching FAQs, Georgetown University; https://provost.georgetown.edu/faculty-affairs/part-time-adjunct-faculty/adjunct-faculty-faqs/. Accessed April 8, 2026. It’s worth acknowledging that the degraded conditions of academic instructional labor tend to complicate a simple deployment of Baumol’s theory: College has become more expensive, but not because teachers’ wages are climbing.
So what’s driving college costs? As the historian Bruce Kimball writes, some economists have increasingly turned to the theories of Howard Bowen (not to be confused with Baumol’s collaborator William Bowen), who developed an alternative explanation to that of Baumol.2222xBruce A. Kimball and Sarah M. Iler, Wealth, Cost, and Price in American Higher Education: A Brief History (Baltimore, MD: Johns Hopkins University Press, 2023), 239–242. Bowen highlights the revenue-cost effect, in which both college revenues and costs increase together. Bowen, writing in 1980, offered five laws for understanding why college costs rise:
1. The dominant goals of institutions are educational excellence, prestige, and influence.
2. In the quest for excellence, prestige, and influence, there is virtually no limit to the amount of money an institution could spend for seemingly fruitful educational ends.
3. Each institution raises all the money it can.
4. Each institution spends all it raises.
5. The cumulative effect of the preceding four laws is toward ever-increasing expenditure.2323xHoward R. Bowen, The Costs of Higher Education: How Much Do Colleges and Universities Spend Per Student and How Much Should They Spend? (San Francisco, CA: Jossey-Bass Publishers, 1980), 19–20.
In Bowen’s account, schools are motivated primarily by excellence (the quality of their educational product), prestige (the amount of social capital that one’s association with the university affords), and influence. What’s more, the amount of spending that could productively be devoted to education is functionally unlimited: A school could always attract better or more accomplished faculty, upgrade lab facilities, build state-of-the-art buildings, acquire new collections for the library, expand subscriptions to academic data bases, and so on. Institutions raise as much money as they can, in part by raising tuition. Administrators also pursue revenue streams beyond tuition: alumni donations, direct state appropriations for public universities, public and private research funding, corporate partnerships, technology transfer, facilities rentals, to name a few. By maximizing revenue, colleges maximize their ability to spend on educational priorities.
And they do spend it all. This might seem a counterintuitive claim: Colleges sometimes run surpluses, and often direct revenues into permanent endowment funds, in which capital is not spent but hoarded. Still, as Kimball and his coauthor Sarah M. Iler point out, endowment investment is a form of spending, since colleges take money that might otherwise function as ready cash reserves and lock it into funds designed, ultimately, to provide more revenue over the long term (which leads to the otherwise puzzling phenomenon of wealthy, well-endowed universities turning to short-term cash borrowing to cover financial obligations in moments of crisis or upheaval, like the 2008 crash, the 2020 COVID-19 pandemic, and Trump’s freeze on federal research spending).2424xKimball and Iler, Wealth, Cost and Price in American Higher Education, 241.
By spending all the money they raise, universities ensure that there is always a need for more revenue—allowing them to keep fundraising and developing revenue streams, which are then spent in turn. All these circumstances lead to Bowen’s final law: When revenues are maximized and spent, college expenditures, both for individual institutions and for the sector as a whole, grow inexorably. And because tuition constitutes the major revenue stream for almost every college or university (except the very wealthiest), the ultimate effect of all these practices is to drive the cost of college attendance ever higher.
The fact of declining rates of academic labor compensation (which, even for tenured and tenure-track professors, has remained essentially stagnant for at least a decade) means that Bowen’s theory now has at least as much purchase as Baumol’s.2525xSee Michael T. Nietzel, “College Faculty Raises Have Not Kept Up With Inflation for the Past Decade,” Forbes, March 30, 2026; https://www.forbes.com/sites/michaeltnietzel/2026/03/30/college-faculty-raises-have-not-kept-up-with-inflation-for-the-past-decade/. See also “AAUP Reports Third Consecutive Year of Faculty Wages Falling Short of Inflation,” American Association of University Professors, June 14, 2023; https://www.aaup.org/news/aaup-reports-third-consecutive-year-faculty-wages-falling-short-inflation. But that doesn’t mean Baumol effects don’t exist. After all, it’s true that education is resistant to productivity gains in precisely the way Baumol describes. So the cost of college rises as a function of both tendencies. How much do each of these effects contribute? Opinions differ. According to a study conducted by the economists Robert E. Martin and R. Carter Hill, the Bowen effect is much larger. Martin and Hill found that for “every $1 in Baumol cost effects there are over $2 in Bowen cost effects”; in some cases, they found Baumol effects responsible for 18 percent of college cost increases while Bowen effects drove 48 percent of those increases.2626xRobert E. Martin and R. Carter Hill, “Measuring Baumol and Bowen Effects in Public Research Universities,” Working Paper 2012-05, LSU Department of Economics Working Paper Series, 2012, 1.
Robert B. Archibald and David H. Feldman, by contrast, caution against adopting a higher-education-specific explanation for rising college costs instead comparing college to other personal-service sectors employing a highly educated workforce.2727xRobert B. Archibald and David H. Feldman, “Explaining Increases in Higher Education Costs,” The Journal of Higher Education, vol. 79, no. 3 (May/June 2008), 289. While they endorse the cost disease, they posit two other effects as similarly important, including higher wages commanded by a highly educated workforce and the expenses associated with what they call an evolving “standard of care”—the idea that colleges and universities must keep current with educational and technological offerings that change over time, including serving students with learning disabilities and offering psychological counseling.2828xRobert B. Archibald and David H. Feldman, “Drivers of the Rising Price of a College Education,” Midwestern Higher Education Compact Policy Report, August 2018, 9. These are critical services, not luxuries, and the mission of higher education has undoubtedly expanded in recent decades to encompass them all.2929xIbid., 9. Yet even Archibald and Feldman see “changes in the distribution of income and decreases in state subsidies” as the key factor in the affordability crisis.3030xIbid., 7.
To summarize, then: Both Baumol and Bowen effects help drive up the cost of college attendance, though different analyses weight them differently. To make matters more complicated, the actual price of college—which often differs substantially from the advertised sticker price—reflects aggressive individual tuition discounting by institutions as universities seek to maximize revenue from the wealthiest students while remaining within reach for everyone else. Finally, it is worth pointing out that, since 2020, despite the widespread perception that college costs continue to rise, the price of college has stabilized, reflecting administrators’ reluctance to raise tuition during a period of crisis and the presence of pandemic-era relief programs that provided new revenue streams to colleges and universities.3131xTristan Stein and Arianna Fano, “College Costs Explained: Why ‘High Tuition’ Headlines Don’t Tell the Full Story,” Bipartisan Policy Center, November 7, 2024. Bowen’s theory might suggest new public subsidies would drive the revenue-cost spiral higher, but revenue-maximization strategies do not happen in a vacuum: College administrators weigh the decision to raise tuition prices against the possible negative effects on enrollment and, increasingly, public perceptions of educational value. In other words, there is no one answer to why college costs so much, and no one strategy that will make it cheaper.
It’s obvious, however, why administrators might prefer a simplified cost-disease explanation of college prices: If the problem is primarily that instructors are overpaid, then the urgent question is one of reducing labor costs. That means that administrative practices of making academic labor less secure and less remunerative are actually valiant attempts at controlling the main driver of college costs. Replacing every tenured or tenure-track instructor with a poorly paid adjunct or contract faculty member helps make college more affordable for students. If, on the other hand, the primary problem is unchecked growth of college expenditures, paired with a relentless search for new revenues, then the solution to skyrocketing costs might involve finding a way to reduce college revenues, a task no sane administrator would relish.
But there’s a danger here too, especially for those of us on the left who want to see college access expanded and who support public institutions of higher education—the most affordable and accessible schools that do the best job of moving students into the middle class. After all, one major part of the story of higher education under neoliberalism has been the declining state subsidies for public higher education, which have forced administrators to turn to other sources of revenue, including international students (who pay full tuition) and, at the biggest and best public institutions, endowments that rival those of the wealthiest private institutions. Yet here, too, the problem comes back to tuition: Federal student loan guarantees allow the price of tuition to rise much higher than most students can pay, meaning that most student loans function as a direct transfer of federal money to colleges and universities. Student borrowers, are, in theory, the ones on the hook to reimburse the feds with interest, but most federally guaranteed student loans are never paid back. One Congressional Budget Office study estimated that 70 percent of the $100 billion in student loans originated by the government each year before 2026 will not be repaid.3232xCited in Steinbaum and Elrod, “Rebuilding American Higher Education: From an Engine of Inequality to a Pillar of the Public Interest,” vol. 32, no. 33. For full CBO study, see Congressional Budget Office, “Reconciliation Recommendations of the House Committee on Education and Workforce,” May 15, 2025, 9; https://www.cbo.gov/publication/61412. It’s clear, then, that it’s less accurate to speak of an end to public support for higher education but rather of its transformation into a kind of funding-by-debt peonage mechanism that routes public dollars to colleges and universities through borrowers who remain shackled to the debt long after the colleges have spent their tuition revenue.
Can AI Cut College Costs?
Anyone who is serious about saving US higher education has to get serious, too, about dealing with the cost of college. What I have been suggesting is that we can best understand administrative adoption of AI as an effort to do just that: AI is, fundamentally, a technology of minimizing human labor and improving productivity. That’s why the idea of an AI-assisted education is so appealing to some administrators.
Think back to the fully automated college course I described at the very beginning of this essay. Would such a course be worth taking? Tatum, for one, seems excited about the possibility, repeating claims that educational chatbots will improve student learning outcomes by offering constant, personalized pedagogical support. The college leaders who spoke at Stanford in March seem less convinced that such a process can teach students at all, let alone do it better than traditional teaching. But it’s best to understand these automated tools as part of a continuum of technologies designed to reduce labor autonomy, increase labor discipline, drive down labor costs, and deskill the academic workforce.3333xHarry Braverman describes the deskilling process as workers seeing a “decline in their command over the labor process” that “is more than compensated for by the increasing command on the part of managers and engineers.” These workers “lose craft and traditional abilities without gaining new abilities adequate to compensate the loss.” Technology is key here: “The more science is incorporated into the labor process, the less the worker understands of the process; the more sophisticated an intellectual product the machine becomes, the less control and comprehension of the machine the worker has.” Braverman’s insights hold true across the economy, not least for teachers today. Harry Braverman, Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century (New York, NY: Monthly Review Press, 1998, 294–295; originally published in 1974. Technologies like online platforms and learning management systems, even when they involve no AI automation, function in exactly this way, as Annie McClanahan and Louise McCune note in their work on “Ed Tech.”3434xAnnie McClanahan and Louise McCune, “Ed Tech,” University Keywords, ed. Andy Hines (Baltimore, MD: Johns Hopkins University Press, 2025), 175–187, 183–184. Also, personal interview with the author, March 12, 2026. Moreover, as McClanahan explained in a recent interview with me, even though ed-tech contracts represent significant expenditures—generally decided on exclusively by administrators, without faculty oversight—the cost savings still outweigh the expenses, especially when combined with employment practices that maintain most instructional workers in low-paid precarity. Baumol effects might be exaggerated, but they are still real, and even if instructional labor is not the main driver of college costs, reducing its price will still help to control them. And, crucially, as McClanahan remarked, all these changes are made possible by the fact that many of the people associated with higher education, from students who view their role as purchasing credentials to administrators who see their job as selling them, have given up on educational quality, especially at the largest institutions that serve the great majority of students.
Here’s the reality: If you, as an administrator or professor, care little about whether the education you offer is a quality product, and if you have little faith in the transformative power of that education, then these kinds of changes are going to be acceptable to you. (And those professors who view their primary role, rightly or wrongly, as being high-level researchers are more than happy to see teaching become easier and cheaper, if worse.) Similarly, if you, as a student, view the good you are purchasing as a credential that allows you to enter the job market at a higher level and nothing more, then getting a pro forma education might be boring or annoying, but it won’t bother you much. And given the continuities between technologized primary and secondary education and the technologized college classroom, it may not even seem all that different. For too many students, that’s just what education is now: a disembodied teacher on a screen, a course website with endless little busywork assignments, boxes to check, grading that doesn’t offer students much beyond a sense of how many points they earned on the assignment. You can sell that to people (especially when you convince them they need it if they ever want to get a job), but you can’t sell it to them for $50,000 or more a year.
But what if we don’t accept these changes? Many administrators would have us believe that the alternative is worse. Already, some educators—as well as many universities3535xCarly Fredericks, “What Is a Microcredential? Here’s What You Should Know,” Oregon State University; https://ecampus.oregonstate.edu/news/what-is-a-microcredential/.—are selling an alternative to college that purports to offer many of the benefits without the extras. These include, as McClanahan noted, microcredentials that certify students in discrete skills, available in a modular format involving a few weeks of study per unit, educational platforms that seek to turn instructors into individual contractors like Uber drivers, and Substack influencers who promise to give students a thorough grounding in the classics. These schemes target different kinds of students: The student who buys a microcredential might not care about the fact that a technical skill is likely to become obsolete not long after its acquisition, forcing them into a career of endless self-financed training, while the student who wants a traditional education but isn’t getting it is liable to give up on higher education entirely and turn to the influencers or to alternative schools such as the University of Austin. Finally, with growing interest in trade schools, the Trump administration has sought to use federal student loan dollars to reward some of that sector’s most unscrupulous operators, neglecting pro-worker alternatives, such as union apprenticeships, in favor of more employer-friendly training. Given these alternatives, isn’t even a somewhat degraded college education worth trying to save, even if it means accepting some amount of AI incursion?
If the Baumol effect was the primary driver of college costs, then the case for accepting some AI instruction might be more persuasive. But at best, it’s one of many factors driving up college costs. The best schools won’t be consigning most of their teaching to AI anytime soon: The expectation of quality, or at least of its appearance, is still too strong among their students. A totally bifurcated educational system, in which the vast majority of students pretend to learn while a computer pretends to teach them while an anointed few command the endless ministrations of attentive pedagogues, isn’t really a system of education at all. Or, it’s not a system of mass education, and it’s not going to save the industry.
Microcredentials and other alternatives to the traditional degree aren’t going away, and, over time, they might continue to siphon off the students for whom obtaining a skills certificate is the goal. But it seems to me this is the moment to insist on the fact that college is different. A real education is different, and rather than equipping students to do something, it gives them the ability to learn to do just about anything. That’s worth preserving.
But what about costs? Martin and Hill are helpful here as well: Their analysis found that public universities relied increasingly on contract and part-time faculty over the period between 1987 and 2008, even as the total cost of attendance increased by 47 percent, running counter to the idea that faculty salaries drive cost increases.3636xMartin and Hill, “Measuring Baumol and Bowen Effects in Public Research Universities,” 7. They also found that a major driver of Bowen cost effects is the decline of shared governance, or, put another way, administrative decision-making without faculty input.3737xIbid., 3. This challenges conventional wisdom about spendthrift faculty and penny-pinching administrators. What’s more, Martin and Hill identify a golden ratio that helps control costs: When there are three tenured faculty members to every administrator, costs rise at the slowest rate. As the proportion changes in favor of administration, costs go up dramatically.3838xIbid., 19. You may not be surprised to learn that most institutions of higher education are quite far from that ratio now, with administrators even outnumbering tenured faculty. Other kinds of instructional jobs have proliferated, but because such workers typically lack a role in shared governance, they have no opportunity to veto or even exercise oversight over administrative spending decisions. They are proletarianized academics who carry out orders based on directives from above. A crucial part of deskilling (the replacement of workers with broad skillsets and autonomy with those with narrow ones) isn’t just creating workers with fewer skills but eviscerating worker control over the work process, reducing the worker to a cog in the machine designed by a company brain located far from the shop floor. That’s clearly what’s happened in academia, and, surprisingly, while it has controlled some labor costs, it has led to an increase in every other kind of spending.
Sometimes it can feel like too big of a task to “fix” higher education. Steinbaum and Elrod’s report, for example, outlined an ambitious overhaul of the entire sector, including the establishment of a federal university and the creation of a large, tiered, state university system that serves local students and offers every student who qualifies a free four-year college education. Admirable goals, and worth working toward, but probably impossible over the short term. Archibald and Feldman, too, caution that the problems of college affordability are tied to larger problems within the economy. Martin and Hill, however, suggest a good intermediate goal: winning back shared governance. Let me add: We can do this now—not by returning to the old models that concentrated decision-making power only in tenured faculty whose ranks are too shrunken to co-govern effectively. Rather, we need to fight to reverse deskilling by empowering the academic proletariat to again make meaningful decisions—about their own work, if not about all institutional spending—thus returning the “brain” of educational work to the classroom from the central office. Let classroom instructors teach in the best way they know how rather than in the way that administrators think saves the most money.
This is easier said than done. One of the effects of institutional precarity has been a move by administrators to exclude temporary and precarious faculty from the campus community, arguing that because such faculty are not permanent, they cannot be allowed to make decisions that could change the university’s long-term fortunes. But on the other hand, when you think about who spends time in the classroom, who knows the students, who carries out the educational mission each day, it begins to look more and more like it’s the administrators who stand outside the campus community.
I can already hear the objections that, like so many disgruntled faculty before me, I’m denigrating administrators as a class, ignoring the many who work hard to build strong institutions and deliver quality education to students. True enough. It’s tough to talk honestly about what’s going on in universities without seeming to turn administration into a term of derision. And it’s important to understand that many of the workers who are classified as administrators carry out critically important work (not least because the term blurs the distinction between powerful, decision-making leaders and clerical and technical workers who are classified as administration despite having minimal decision-making authority). Returning some of that power to the workforce would be a smart first step.
Unionization and strong union contracts that enshrine working and instructional autonomy are a major part of the solution. For those who believe in a simplified cost-disease model, unionization only drives up college costs by making labor more expensive. But if you believe Martin and Hill’s assertion that the mismatch between administrative decision-making power and faculty autonomy drives costs far more, then unions become not a way of making college more expensive but a key mechanism for controlling costs as well as for rebuilding student and public trust. The Luddites, nineteenth-century anti-machine activists, are often caricatured as anti-technology fanatics, and their legacy has been invoked again and again to dismiss those who question the utility of ostensibly historically progressive technological tools, like AI. But the Luddites were not universally opposed to technology as such, but to the kinds of technological innovation that increased productivity at the price of workers’ autonomy, workers’ wages, workers’ lives. For those of us who make our living teaching college, AI is such a technology. And in that case, a little more machine breaking might not be the worst idea we’ve ever had.