Higher education in the United States is in crisis. No news there. While the Trump 2.0 assault on scientific funding and immigrant workers represents a significant escalation, the culture wars and fiscal austerity are long-standing problems for us. What is more novel is the rapid proliferation of a technology that is experienced as both a core assault on our educational mission and a potential savior: generative AI. Since fall 2023, students, administrators, and, to a more uneven extent, academic workers have rapidly adopted ChatGPT and its competitors. This technology is supposedly destroying the art of reading and writing, speeding up research production, and driving an unprecedented wave of investment—all at once. The novelty of the underlying technology, its constant evolution, and its deployment across wildly different institutional settings prevent any final conclusions. But it is possible to draw up a provisional balance sheet, to map what work AI is doing on campus, where, and for whom, especially because the core conflict is an old one. The work of teaching and research, like that of many service professions, is labor-intensive, relatively autonomous, and resistant to automation. Worse still, as I discuss below, because increases in productivity are difficult to realize, this sort of work grows more expensive every year. Under these conditions, the lure of labor discipline, if not outright replacement, will always be strong for our employers.
This is a recurring cycle. Teresa Sullivan, a former president of the University of Virginia, was targeted (ultimately unsuccessfully) for removal in 2012 in part because of her insufficient enthusiasm for massive open online courses (MOOCs)—more specifically, because she doubted the financial analysis UVA’s board of visitors drew from a David Brooks op-ed. So, when we talk about what work AI does on campus, we should be alert to not just what code is written with Claude or what spreadsheets are cleaned with Gemini but also what political and economic functions the technology serves within our institutions—and for whom.
A recent survey of AAUP members conducted by the Association’s ad hoc Committee on Artificial Intelligence and Academic Professions showed that faculty feel locked out of major decisions regarding AI on campus. Because academic workers have little say in procurement at most institutions—the advisory process of shared governance rarely touches the budget—we also don’t have much say in the design, use, and regulation of large-scale AI deployments on campus. They largely proceed on the administration’s terms. This is not to say that we are left out entirely; academic workers are constantly building, fighting, and negotiating AI in our teaching, research, and service. And, of course, as Sullivan’s case showed, even upper-level administrators do not have the last word, because they have bosses too: regents; state and federal politicians; employers and donors; and, most important of all, the bond market that funds our institutions’ new buildings and determines their credit ratings.
Cost Disease
Why are our bosses, in their many guises, so enthusiastic about AI on campus? The answer lies in the nature of our work. Higher education, like hospitals or orchestras or federal bureaucracies, is a labor-intensive sector that is difficult to automate and that serves a social purpose, rather than solely pursuing profit. When sectors like ours are in the black, they generally spend that surplus on those social purposes (more nurses or concerts) rather than buying back stock or investing in new fixed capital for greater returns. Workers making cars or washing machines can produce more widgets per person per hour as new technologies—assembly lines, robots—augment their labor or replace it entirely. Not so with teachers or violinists, whose product is a relationship, experience, or other immaterial good consumed at the point of service. But both sets of workers—in services and manufacturing—exist in the same labor market and, all things being equal, the increased productivity in manufacturing drives wage growth not just in that sector but in services, too; otherwise we’d all leave our jobs for the well-paying factory. But because there’s little corresponding productivity growth in services, our work becomes more expensive over time in real dollars. We’re teaching roughly the same way as sixty years ago, but everything (including labor) costs more.
Major consequences have followed. First, in developed economies all over the world, the majority share of employment has shifted to person-to-person services as automatable industries do more with fewer workers. Second, economic growth in general has slowed, such that the high-productivity, high-growth decades after World War II now look like an exception rather than the rule. And, most important for our purposes, work like health care and education has become more expensive over time even when controlling for inflation. The revenue that was needed to operate the state or the school sixty years ago is no longer sufficient, even if the workers are doing the exact same thing (and, of course, they are doing much more).
In the 1960s, economist William Baumol identified this phenomenon as “cost disease.” It’s as much a political problem as an economic one, and it is why capital has identified not just universities or hospitals but the professions more broadly (law, journalism, software development) as ripe targets for disruption by AI. Our labor process is expensive, prosocial, and somewhat resistant to the labor discipline we associate with the assembly line and Frederick Taylor’s stop-watch—if for no other reason than our final products are texts and experiences rather than discrete widgets. Because large language models (LLMs) have proved adequate at digesting and producing written text, there is great hope among the bosses that they can observe and direct the individualized and nonroutinized work we do to produce text, replace or deskill some of that work, and speed up what remains.
This hope for a magic bullet to use against the professions is widespread. Software companies are fed up with the high wages they had to pay during the boom years of the 2010s and see their employees as having taken advantage of that security to protest against big tech’s collaborations with the Pentagon or Immigration and Customs Enforcement. Microsoft, for example, is requiring employees to use AI, assessing them on it, and eliminating internal libraries in favor of LLM-generated resources. Like software developers, higher education is full of tinkerers who will experiment with AI on their own, though some coercion or persuasion is still needed to bring along the tinkerers’ colleagues. But higher education is unique because AI is brought to bear on our work from multiple directions, through multiple levels of management—financial, administrative, and political. In what follows, I map the different dimensions of AI implementation on campus and what effects it may have on our work.
Research
The first and best use case for AI on campus is also the domain where management has the least coercive power: the research enterprise. This is not to say that there isn’t political pressure to explore or ignore certain topics (especially now) but that it’s difficult for our deans or provosts to direct the research process of each individual worker—too many projects, too many varieties of expertise, too few supervisors. In this regard, academic freedom is a negotiated settlement: Research is researcher-led because we’re the experts, our autonomy produces the best results, and any other option is more expensive. And we’ve done some marvelous things with AI, especially in the biological sciences, where simulations of complex processes like protein folding have gotten cheaper, faster, and much more sophisticated. Exciting possibilities exist in a host of fields. The problem is that our expertise, about AI and everything else, is not respected in the other parts of our job that are more directly supervised and that, at the end of the day, provide the infrastructure and revenue to support our research.
Credentials
Our greatest panic over AI comes from our classrooms. Particularly in composition, but also in courses ranging from programming to chemistry, instructors have struggled to stay afloat amid a tide of AI-generated student submissions that have fundamentally altered the educational social contract. There are some paths forward here, new pedagogies that make technical training more of a dialogue or old ones that return the humanities to venues of oral debate and handwritten essays. But this is fundamentally a problem of institutional design, not local survival strategies, if for no other reason than that many of these solutions do not scale. I can’t do blue book exams in my current 200-level course because it’s just me, seventy-five students, and a half-time teaching assistant.
Students are using ChatGPT to write essays for the same reason I cannot get fewer than seventy-five of them in that classroom: Reduced investment in public higher education at the state level has shifted the cost of education onto students and their families, while demanding that instructors do more with less. Per-student state appropriations for higher education in my state, like most other states, have been basically flat since 2010. Public monies are increasingly being replaced with tuition. The result? Students want to get their labor market credential as quickly as possible to avoid as much debt as possible. From the faculty’s perspective, when our institutions sign multimillion-dollar deals with OpenAI to make their tools available to our students, it can feel like enthusiastic agreement to this pay-for-play credentialism. AI proves sociologist Tressie McMillan Cottom’s point that the privatization of higher education pushes all our institutions to work more like the “for-profits” we professionals deride as “diploma mills.”
Legitimacy
Baumol’s “cost disease,” especially amid state defunding, may explain administrators’ long-term obsession with forcing us to do more with less. But that long-term trend does little to explain why, over the course of barely a year, so many colleges and universities found the money for least one AI institute, maybe an AI degree, definitely an official AI agent to ask questions of, and a proud declaration that they are now AI-first institutions. This is more obviously a political phenomenon.
In institutional sociology, we know this sort of homophily isn’t necessarily a product of market competition. Especially in an environment where money is tight and the path to long-term success is uncertain, institutional leaders look to signals from their peers and their regulators to provide a solution. In this way, all these big, public AI announcements are attempts to secure legitimacy by imitating corporate leaders and brand-name peers. That legitimacy may be granted by state politicians who are themselves afraid of being left behind, by regional employers who are under the same pressures and who discuss them from their seats on our boards, or by more powerful peers who might invite lower-tier leaders to sit on their boards or begin some institutional partnership that spreads the social capital around. This was the threat to Teresa Sullivan at UVA: Her bosses thought she ignored the status signals they were seeing. After MOOCs came big data and machine learning, but the struggle to earn legitimacy through AI puts those earlier fights to shame. It is no coincidence that this short-term political development comes as state and federal politicians defund gender studies departments, ban research into “woke” topics, and end faculty tenure, faculty senates, and other means of faculty autonomy. This attack has lowered the status of institutions of higher education. Our administrations are wagering that enthusiastic AI adoption will increase that status, or at least mollify our attackers.
Surveillance
For many people in higher education, like many workers in a service economy, the most common experience of AI will not be using it ourselves but having it used on us. It is not necessarily our labor that is automated but our supervision. And not just because we don’t have a vote on procurement. The cloud-based platforms our universities buy to manage our payroll or teach our classes are not purchased but subscribed to, and none of us, neither professors nor administrators, have control over AI features that are introduced into enterprise software. While we don’t control our data once they enter these systems, we may be able to avoid using, say, a new AI widget on learning management systems like Canvas. But there is no opting out of backend updates to Workday when it is processing your benefits or your initial job application. University payroll and hiring is a notorious mess, so management perceives some efficiencies to be gained there, though it may come at the expense of human resources staff who aren’t directly delivering our “product.” The pattern-recognition skills of contemporary AI are also put to use in more traditional surveillance systems—video, audio, internet-trawling—that universities under growing threat of violence may feel obligated to purchase as safety measures. Whether and how these local systems and their data connect to their state and federal counterparts are open questions for each installation, one that is difficult to answer given that all software-as-a-service products are permanently in beta.
AI is also a gift to the snitch state: the right-wing project to enroll private citizens in the surveillance of their neighbors and their public servants. Want to comb through syllabi for mentions of slavery or find the faces and home addresses of student protesters? AI is here to help with quick, dirty, large-scale analysis. As legal scholar Salome Viljoen has noted, the Right understands very well that all governance is data governance. AI facilitates tighter regulation of public data than ever before. This, too, is labor discipline in the face of cost disease. Automated text analysis becomes a means to avoid preexisting, even nominal, deference to both institutional deliberation and expert consultation. We can see this in Elon Musk’s techno-cultural revolution at the Department of Government Efficiency, where young men straight out of college (or still in it!) used ChatGPT to rewrite federal regulations, scour grant applications for thought crimes, and comb through the social media profiles and legal histories of students on visas in the hopes of finding excuses to deport them. At the National Endowment for the Humanities, private equity associate turned DOGE warrior Justin Fox canceled grants by asking ChatGPT to say, in 120 or fewer characters, whether a project involved diversity, equity, and inclusion. In states where university management has been captured by antiuniversity activists, faculty members are being forced to publicize syllabi far in advance of their teaching so as to provide more source material for right-wing policy entrepreneurs. None of this is new; all of it is much easier with AI.
Service Speed-Up
Administrative labor seems only to grow from year to year. This is the university worker’s experience of cost disease: doing more with less because the same workforce only gets more expensive. Here management sees AI as a means of solving the problem from the inside. Again, this is less revolutionary than evolutionary, but it is also a trend much less widely discussed than undergraduate cheating—though potentially just as consequential.
Whether as a conversation agent or a tool for large-scale text analysis, AI invites our institutions to replace specific human tasks, such as first-round interviews, or to delegate those tasks, such as compliance reporting, to nonspecialists. For the latter, think of the payroll or expense reporting functions shifted to end users through platforms like Workday. That a machine can read your receipts would seem like an advance, but the distrust of public-sector workers has not gone away, so the result is that little time is saved because end users must do more uploading and annotation to create a machine-readable product that still complies with local regulations.
More distinct to our sector is the process of shared governance and its potential automation. While truly shared governance has been eroded at many institutions, nonmanagerial academic workers still exercise greater control over hiring, promotion, and curriculum development than professionals in other large organizations, even if this is only ever advisory to management. But the workload increases every year. And as adjunctification continues, those in the secure minority are left with what historian Erin Bartram calls a “phantom pain where their colleagues should be.” Here, cost disease forces the interests of labor and management to converge. We both want to reduce the time spent on administrative tasks.
How AI meets this demand remains to be seen. Whether it works is an entirely separate question. But there are early hints. My own university is currently holding “listening sessions” on the application of AI tools to the tenure and promotion process, and I hear anecdotes about this informal automation from faculty members across the country. The possibilities here exist on a spectrum of delegated control: from copyediting to reviewer suggestions to writing summary statements of achievements to decision recommendations. Some automation here may be welcome, allowing us to focus not on formatting but on interpreting. But the lines are blurry, and the actual procurement decisions are out of our hands. Compromise solutions would involve ending the understaffing that contributes to overload in the first place and deploying faculty AI expertise on our own local systems rather than contracting that work out to Google or OpenAI. But cost disease is a powerful disincentive to both. If our services only ever rise in cost, then taxes or tuition must be raised to recruit more faculty members—a thorny political problem. And when more design decisions get placed in our hands, it’s more difficult, and thus more expensive, to change course without our consent.
Investment
Ultimately, AI adoption on campus is a question of cost, both financial and political. That cost is not a local question but a global one, given the trillions of dollars in investment flooding into AI companies right now. In the United States, 80 percent of stock gains in 2025 came from AI investment. Decades of cuts to taxes on incomes, corporations, and capital gains surely provided much fuel to the fire, and the desire for ever more fuel explains the heavy presence of AI investors like Marc Andreesen in the Trump administration. As state funding for public higher education has stagnated, the importance of debt has risen. The university’s creditworthiness is front of mind for all administrators and governing boards—the bond market is the boss of bosses—and today that investment environment orbits around AI.
This investment landscape imposes itself on university organization in several ways. First, ratings agencies like Moody’s may encourage the follow-the-leader dynamic we saw above in the hunt for legitimacy. If sector leaders like Harvard University are building AI institutes and creating AI degrees, that is a sign that the rest of the pack should pursue those options as future revenue sources. If you don’t, your bonds—bank loans in exchange for guaranteed future tuition revenue—and thus your capital costs may get more expensive. Second, university endowments are inevitably invested in the AI boom because the AI boom is such an enormous piece of the stock market. The same goes for the 401(k)s or pension funds on which our retirements rest, if we’re lucky. Indeed, it would be a major violation of fiduciary duty if the private asset managers in charge of our collective investments did not put our money into Google or Amazon. Index funds that simply track the market are the safest investment you can make, and the market simply is AI right now. Finally, in times of public austerity, university administrations become ever more desperate for private philanthropy and ever more attuned to the needs of potential donors. Living donors are particularly sensitive to follow-the-leader dynamics in their cohort and the broader market and so may be on the lookout for innovative new units, degrees, or buildings focused on AI.
Public Goods If You Can Keep Them
Here arises an unsettling dynamic largely independent of what AI, as a tool, actually does for us or anyone else. The wealthiest people in the world have grown ever wealthier through investments in AI. They have used that wealth to install an anti-education presidential administration that uses AI to hunt down thought criminals and immigrants. To survive the assault, university management turns to investments in AI, and in those same people attacking us, in the hopes that doing so might secure funds and legitimacy.
For our administrations, then, AI is another method to privatize the public good of education, not just in the sense of who owns the institution but in the social atomization of the collective educational enterprise. AI is welcomed as a silver bullet precisely because there is no silver bullet for high-quality, accessible education, just the usual boring solutions: small classes, cheap or free tuition, well-resourced students. Cost disease makes college more expensive year over year, and our collective refusal to tax wealth locks in this dynamic, forcing us to try to do even more with even less every year. AI is attractive to employers across the economy because it seems to offer a way to produce expertise at scale without relying on quite so many experts.
Baumol made a similar observation in the 1960s with respect to the urban crisis: Reformist mayors, he wrote, found themselves having to do more with less because most of the public goods they supplied (health care, policing, education, maintenance) are difficult to automate. “Self-help offers no way out for our cities,” he said, arguing that there’s no bootstrapping yourself out of this long-term fiscal dynamic. Given that many universities are like small cities themselves, and our university systems the largest employers in many states, the same analysis clearly applies. We’re not innovating our way out of this. The remedy is simple but difficult. True federal investment in higher education would necessarily tax revenue from the private wealth financing the AI bubble (or whatever comes next) and redirect it to those public goods that make an educated society possible in the first place. True public wealth would mean that public servants like us have the resources to do our jobs as best we can, with the tools we need, under terms that best serve our students, our states, and our science.
Daniel Greene is associate professor of information at the University of Maryland, vice president of United Academics of Maryland AAUP-AFT, and a member of the AAUP’s ad hoc Committee on Artificial Intelligence and Academic Professions. His email address is [email protected].