Universities Are Obsolete—If All They Do Is Training
The internet, Khan Academy, YouTube and LLMs have made information abundant and much training radically cheaper. I agree that universities should not be protected by government funding.
The internet, Khan Academy, YouTube and LLMs have made information abundant and much training radically cheaper. I agree that universities should not be protected by government funding. But access to information is not the same thing as education, and confusing the two is precisely how universities made themselves so easy to replace.
Keywords: education; training; universities; credentialism; higher education; artificial intelligence; online learning; government funding; apprenticeships; liberal education
A reader, Kurt Overley, responded to my earlier post Education Before Training with an argument that deserves a serious answer. His position, in essence, is that the internet has changed the economics of higher education beyond recognition. Kindle gives us libraries in our pockets. Khan Academy gives away structured instruction. YouTube contains lectures, demonstrations and explanations on almost anything a university can teach. Massive open online courses made elite course material available globally. Large language models can now explain, quiz, summarise, generate examples and provide something approaching individual tutoring at negligible marginal cost. If the informational content of a university education can be acquired without a university, why keep subsidising universities at all?
On the funding question, I am substantially in agreement. I do not think universities should be funded by government: no operating subsidy, no subsidised tuition, and no state-backed student credit or loan guarantees. Institutions that cannot persuade students, families, donors, employers or other voluntary funders that what they provide is worth its cost should not be insulated from that judgement merely because they have historically possessed academic status. Government support also changes incentives: it weakens price discipline, supports credential expansion and turns political priorities into institutional priorities. There is empirical evidence that expansions in federal student credit can pass through into tuition, although the magnitude differs by programme and institution; Lucca, Nadauld, and Shen (2019), for example, found substantial tuition pass-through from changes in federal loan limits, while Cellini and Goldin (2014) found much higher prices among aid-eligible for-profit programmes than comparable nonparticipating programmes. The case against subsidy should not, however, be caricatured as a claim that financial aid never helps anyone. Causal evidence also shows that grants can increase attendance, completion and later earnings for some students (Denning et al., 2019; Dynarski, 2003). A serious argument has to admit both sides.
My disagreement begins one step earlier. The fact that information has become cheap does not establish that education has become obsolete. It establishes that information delivery has become cheap. The distinction matters because universities have spent decades confusing the delivery of information, the certification of competence, the signalling of status and the education of a person. Once those functions are separated, the apparent contradiction disappears. The internet is devastating to institutions whose principal product is content delivery. It is highly disruptive to institutions whose principal product is routine training. It is potentially devastating to institutions whose principal product is a credential. But none of those propositions demonstrates that the formation of judgement, intellectual independence and the capacity to reason well has become unnecessary. If anything, a world in which anyone can obtain an answer instantly makes those capacities more important.
That is the point I think the obsolescence argument misses. I am not defending the contemporary university as it exists. Much of it deserves disruption. I am defending a distinction between having access to knowledge and becoming the sort of person who can use knowledge well.
1. A library is not an education
The easiest way to see the problem is to take the technological argument to its logical conclusion. The internet did not invent abundant educational material. A large research library already contained more information than any undergraduate could consume in a lifetime. Public libraries made enormous bodies of knowledge available for little or no direct charge. Cheap paperbacks radically reduced the cost of owning serious books. Recorded lectures, correspondence courses, educational television and home-study programmes all predate YouTube. The internet changed the scale, speed and convenience of access, and LLMs changed the ease with which information can be reorganised and explained, but the existence of a resource has never been identical to education.
If access were sufficient, the invention of the printing press should have abolished universities. It did not. Nor did the public library. Nor did radio. Nor did television. Nor did the personal computer. Nor did the first generation of online courses. The reason is not that institutions possess some mystical quality that technology cannot touch. The reason is simpler: learning requires activity by the learner, not merely availability of material.
The empirical literature is quite clear that the mode of engagement matters. Chi and Wylie’s (2014) ICAP framework distinguishes passive, active, constructive and interactive engagement and predicts stronger learning as students move from merely receiving information toward generating, explaining, questioning and interacting. Freeman et al. (2014), in a meta-analysis of 225 studies in undergraduate STEM education, found better performance and lower failure rates under active learning than under traditional lecturing. Those findings are not an argument for a particular campus architecture. They are an argument against treating exposure to excellent content as equivalent to education. A world-class lecture on YouTube may be better than a mediocre lecture in a university hall, but watching either one remains a different cognitive activity from defending an argument, solving a problem without a template, being forced to explain an assumption, receiving targeted criticism and reconstructing one’s view in response.
This is where the online-learning evidence is instructive rather than decisive. Bettinger et al. (2017) found that, in the large institution they studied, taking a course online rather than in person reduced grades, subsequent performance and persistence. That does not prove that online education is inherently inferior; the result is context-specific, and well-designed online learning can be excellent. But it does show that placing nominally similar material online does not make the surrounding educational process irrelevant. Reich and Ruipérez-Valiente (2019) documented a related problem in the first great wave of MOOCs: enormous initial reach did not translate into the universal transformation once predicted, completion remained low, and participation growth was concentrated in comparatively affluent countries. The problem was never scarcity of video.
At the same time, the strongest evidence for educational technology cuts against any simplistic defence of universities. Kulik and Fletcher’s (2016) meta-analysis of intelligent tutoring systems found substantial average benefits across controlled evaluations. Technology can teach. It can give feedback. It can adapt practice. It can sometimes outperform conventional instruction. Contemporary LLMs push this much further because they can hold a dialogue, generate examples at different levels of difficulty, challenge a student’s answer and explain a concept repeatedly without impatience. Kasneci et al. (2023) therefore rightly describe both major opportunities and major risks for large language models in education, including personalisation on the one hand and the need for critical checking and human oversight on the other.
But this strengthens rather than weakens the distinction I am making. If technology can perform instruction and tutoring better and more cheaply, then we should use it. We should not preserve inefficient university practices merely to preserve employment. Yet the educational question remains: what is the learner being formed to do? A tutoring system can optimise performance against a defined objective. An LLM can help a student understand calculus, Latin grammar, econometrics or constitutional law. That is valuable. It still does not follow that the endpoint of education is successful task performance.
Training asks, in one form or another, whether a person can reliably perform what a domain requires. Education asks a larger set of questions: Can the person decide which method applies? Can they recognise when the specification is wrong? Can they distinguish evidence from rhetoric? Can they reconstruct an opposing argument before rejecting it? Can they reason under uncertainty? Can they identify the limits of what they know? Can they connect a technical problem to historical, ethical, institutional or scientific context? Can they change their mind for reasons rather than social pressure?
A search engine can provide sources. An LLM can summarise them. Neither act, by itself, makes the user educated.
2. Training is increasingly automatable; education is not reducible to the same object
The distinction between education and training is often treated as elitist because the word training is heard as an insult. It should not be. Training is indispensable. I want the pilot flying an aircraft to be extremely well trained. I want a surgeon to have repeatedly practised procedures until competent performance is reliable. I want an electrician to understand and comply with electrical standards. I want an accountant to know the rules that govern the accounts being prepared. I want a programmer to be able to produce correct code.
Training is not inferior because it is practical. It is different because its validity can usually be expressed in relation to performance requirements. There is some specification, even if sophisticated, against which competence can be judged. Expert practice may involve tacit knowledge, adaptive judgement and years of improvement beyond the threshold of qualification, but certification still has to warrant a proposition such as: this person can safely and reliably perform the relevant work.
Education cannot be exhausted in that way. It contains thresholds—no one should receive a mathematics degree while being unable to do mathematics—but its highest goods are open-ended. There is no finite checklist that exhausts historical judgement, mathematical imagination, philosophical discrimination, scientific creativity or the ability to construct a genuinely original argument. Better performance remains meaningful above any minimum. The purpose is not simply that the student can do X. It is that the student’s capacity to understand, judge and reason has been enlarged.
This distinction is close to an older conception of education that modern universities have steadily obscured. Newman (1996) defended the cultivation of intellect as a good not reducible to immediate utility. Dewey (1916) treated education as growth rather than mere preparation for a fixed future task. More recently, Biesta (2009) has argued that educational purposes cannot be collapsed into measurable learning outcomes and distinguishes qualification from socialisation and subjectification. These traditions differ sharply in philosophy, politics and institutional assumptions, but they share one important refusal: they do not treat education as the efficient transfer of content from expert to novice.
That older insight becomes more, not less, relevant when technology improves. Consider what LLMs do to the ordinary university essay. If the assignment asks a student to produce 2,000 competent words explaining a familiar topic, a modern model can often generate an adequate answer in seconds. The natural institutional reaction has been to ask how to prevent students from using the tool. But that is frequently the wrong question. If the educational value of the assignment disappears once the production of the artefact is automated, we should ask what the assignment was actually measuring.
Perhaps it was measuring prose production. Perhaps it was measuring recall. Perhaps it was measuring the ability to assemble sources into a conventional structure. Those are all trainable abilities, and some remain useful. But if the claimed purpose was to develop judgement, then the proper assessment should expose judgement. Ask the student to defend the causal inference orally. Give them a contradictory source and ask which assumption fails. Require them to choose between two plausible methods. Make them identify what evidence would change their conclusion. Ask them to reconstruct the strongest objection to their own argument. Let them use the AI, but make them answer for the reasoning.
The same applies to statistics. Teaching a student which buttons to press in SPSS was never the same as teaching statistics. Automating those buttons with AI does not destroy statistical education; it exposes how little education there was in the button-pressing exercise. A student who knows why a model is appropriate, what assumptions identify the parameter, what would bias the estimate, how uncertainty should be represented and when the result should not be trusted has an education in the subject. A student who knows how to execute the command has training. We need both, but we should stop naming them as though they were identical.
This is why I think Kurt’s technological argument is strongest against universities at precisely the point where universities have become weakest intellectually. The more a university defines itself through content delivery, standardised assignments, procedural competence and credential issuance, the more easily technology substitutes for it. The answer is not to protect those activities. It is to stop pretending that they define education.
3. Universities made the case for their own obsolescence
The contemporary university is vulnerable because it has increasingly sold itself in the language of employment, credentials and measurable outputs. Students are told to invest in a degree because of the earnings premium. Governments justify expenditure through productivity and workforce development. Employers use degrees as screening devices. Universities advertise employability, graduate salaries and professional accreditation. Courses are described through learning outcomes and competencies. None of these practices is automatically illegitimate, but together they encourage an institutional shift from education toward certification.
Economics gives us two different ways to understand the degree. Becker’s (1993) human-capital account emphasises the productive skills education can create. Spence’s (1973) signalling model shows how education can also function as information in a labour market: the credential can help employers infer characteristics about applicants even when the education itself is not the sole cause of those characteristics. The two mechanisms can coexist. A degree can both teach something and signal something.
The trouble begins when the signal becomes the product. Labaree (1997) describes a long-running tension among democratic equality, social efficiency and social mobility as goals of education, arguing that the social-mobility goal tends to turn education into a private positional good and elevate credentials over learning. That diagnosis becomes particularly important in a system where employers require degrees for jobs that previously did not require them, students purchase credentials partly because others are purchasing credentials, and universities respond by expanding programmes whose value depends partly on the scarcity and status of the credential itself.
In that world, the internet is indeed destructive. Why pay tens of thousands for lectures when lectures are free? Why sit through a semester of software instruction when an AI tutor can teach the relevant commands in an afternoon? Why take a generic writing course if personalised feedback is available instantly? Why pay a university to tell you which textbook chapter to read when the world’s literature is searchable from a phone?
Those are not rhetorical questions. Universities should have to answer them.
My answer is not that the campus is sacred. It is not. Nor is the lecture sacred. Much traditional lecturing is a poor use of expensive expert time, and the evidence for active engagement should have changed teaching practice much more aggressively than it has (Chi & Wylie, 2014; Freeman et al., 2014). Nor is the professor sacred as the exclusive source of explanation. If an AI system explains a difficult theorem better than the person assigned to teach it, the student should use the AI system.
The university deserves to survive only where it provides something that cannot be reduced to the cheaper substitute—or where it combines the cheaper substitute with something educationally more demanding. That might include sustained supervision, adversarial seminars, laboratories, studios, clinical formation, access to research communities, peer intellectual life, high-quality feedback, oral examination, mentorship, protected time for difficult study and the maintenance of standards that cannot be passed merely by producing a polished artefact.
Put differently, the modern university often behaves as though its moat is possession of information. That moat has gone. Good. It should have gone. Information should be abundant. The remaining question is whether there is value in organised intellectual formation once information is abundant.
I think there is, but I do not think every existing university deserves to provide it.
4. Yes, end government protection—but do not pretend there is no cost
I agree with Kurt that universities should not be funded by government. I would end operating subsidy, government-backed expansion of student borrowing, tuition subsidy and the assumption that taxpayers must continually enlarge the higher-education sector. A university that claims to create immense value should be capable of attracting voluntary payment, philanthropic support, endowment income, employer sponsorship, private scholarship or other non-state funding.
The economic case is not difficult to understand. When a third party subsidises demand, the buyer becomes less sensitive to price. When credit expands, institutions can capture part of the increased purchasing power. Lucca et al. (2019) found that increases in US federal student-loan caps were associated with meaningful tuition pass-through, with larger effects in some expensive, for-profit and two-year programmes. Cellini and Goldin (2014) found that aid-eligible for-profit certificate programmes charged substantially more than comparable programmes outside federal aid. These results do not prove that all public aid is capitalised into tuition, nor that eliminating aid would mechanically reduce every fee. They do show why it is naïve to assume that subsidising the purchaser leaves the seller’s pricing behaviour unchanged.
But the strongest argument against my position should be stated plainly. Student aid can also relax real liquidity constraints. Dynarski (2003) found that grant aid increased college attendance and educational attainment in the population she studied. Denning et al. (2019) found that additional Pell Grant eligibility increased degree completion and later earnings among students in four-year institutions in their Texas setting. If government funding is removed abruptly, some capable students who would have attended university will not attend. Anyone advocating the end of subsidy who refuses to acknowledge that is not engaging seriously with the evidence.
My answer is that access and subsidy are not synonyms. If an educational institution is genuinely valuable, there are other ways to finance access: private scholarships, endowed places, philanthropic foundations, employer sponsorship, work-study arrangements, apprenticeship-linked study, deferred-payment mechanisms and much lower-cost institutional forms. A sector exposed to price discipline would also have to confront costs that have been allowed to become normal because students can borrow against future income with state support.
The transition would be difficult. Some institutions would close. Some programmes would disappear. Enrolment would probably fall. I do not regard those outcomes as self-evidently bad. The assumption that a society is better educated merely because a larger proportion of young adults is enrolled in universities is precisely the assumption I reject. A country can have more degrees while having less education if the degree becomes primarily a labour-market licence.
There is also a political reason to separate education from the state. Whoever funds an institution acquires leverage over it, even when that leverage is indirect and well intentioned. Funding formulas create metrics. Metrics create behaviour. Regulatory conditions become institutional priorities. Research agendas respond to grant structures. Universities that depend heavily on political funding inevitably become political objects. The problem is not that government is uniquely wicked. The problem is that dependence creates control.
A genuinely independent university should be able to tell the government that it is wrong without wondering what happens to next year’s allocation. It should be able to teach unfashionable ideas without negotiating a bureaucratic definition of social value. It should be able to fail students without treating completion rates as a financial threat. It should be able to remain small if serious education requires smallness.
This is one reason the funding issue and the education-versus-training issue are connected. State funding often requires legibility: measurable outcomes, completion targets, employment measures, standardised qualifications and comparable indicators. Those things are understandable from the perspective of public accountability. A government spending public money wants evidence of what it purchased. But the easiest educational goods to make legible are often those closest to training and certification. Open-ended intellectual development is harder to compress into a dashboard.
Once universities organise themselves around what funders can count, they gradually become organisations that produce countable things.
5. The strongest objection: perhaps AI can educate too
There is an obvious reply to my argument. Perhaps I am drawing a distinction that technology itself is about to erase. If an LLM can challenge a student’s premises, play devil’s advocate, ask Socratic questions, adapt to misunderstanding, generate counterexamples, simulate an oral examination and provide unlimited individual feedback, why assume that intellectual formation requires a university at all?
I do not assume it.
This is the most important qualification in my position. I am defending education, not the inherited institutional monopoly of the university. If AI systems eventually provide better intellectual challenge, better feedback and better personalised instruction than most human institutions, we should use them. If groups of serious learners can form independent seminar communities around digital tutors, libraries and examinations, some functions of universities may disappear entirely. If employers cease demanding degrees and instead test what candidates can actually do, the credential function may shrink. If professional bodies can certify competence directly, much professional training may leave universities.
That would not refute my argument. It would vindicate it.
The test is functional. Does the arrangement merely provide answers, or does it develop the learner’s capacity to judge answers? Does it merely optimise performance on known tasks, or does it expose the learner to problems where the task itself must be framed? Does it reward agreeable fluency, or force the student to confront disagreement? Does it build dependence on the tool, or increase the learner’s ability to detect when the tool is wrong?
The danger with LLMs is not simply cheating. It is epistemic outsourcing. A student can now produce work whose surface quality substantially exceeds the student’s underlying understanding. Tlili et al. (2023) identify concerns around truthfulness, misuse and educational practice, while Kasneci et al. (2023) emphasise the need for critical thinking, fact-checking and understanding the limitations of large language models. The educational problem is therefore not that the machine can write. It is that the artefact can cease to be reliable evidence of the person.
That forces better assessment. It also forces us to decide what we actually value.
If a lawyer can use AI to draft a contract, then legal education should spend less time rewarding the mere production of standard clauses and more time testing whether the lawyer understands their consequences. If an economist can ask a model to run code, economics education should test model choice, identification, interpretation and error detection. If a programmer can generate code, computer science should place more weight on architecture, verification, debugging, threat modelling and understanding why the system works. If a historian can retrieve and summarise secondary literature instantly, historical education should become more demanding about primary evidence, provenance, interpretation and competing causal narratives.
AI should not lead us to eliminate thought from education. It should lead us to eliminate unnecessary mechanical work from education so that more time can be spent on thought.
This is also why I am sceptical of blanket bans on AI in universities. They protect obsolete assessments. If an assignment collapses the moment the student is allowed to use the tools available in the real world, redesign the assignment. There will still be circumstances in which unaided performance must be tested, just as pilots sometimes have to demonstrate procedures without automation. But the reason should be educational validity, not nostalgia.
The university of the future, if there is one, should be harder intellectually and easier technologically. Students should have extraordinary tools. They should also face standards that tools alone cannot satisfy.
6. What would a university worth keeping actually do?
If we removed government protection and accepted that information delivery and much routine training can be provided more cheaply elsewhere, universities would have to become much clearer about their purpose. I would not attempt to save every institution. I would separate functions that have been bundled together for historical reasons.
Vocational training should be allowed to be vocational training. Apprenticeships, technical institutes, employer-based programmes and specialist academies should not be regarded as failed versions of universities. For many occupations they are a better institutional form because learning is embedded in practice. A person does not need three or four years of nominally academic study merely because the labour market has converted the degree into an entry ticket.
Professional formation should combine rigorous knowledge with accountable judgement. Medicine, engineering, law and similar fields require common competence thresholds, but expert practice cannot be reduced to those thresholds. Their educational structures should therefore combine training, supervised practice and intellectually demanding study rather than forcing everything into one vocabulary of generic learning outcomes.
Universities proper should concentrate on education that is genuinely advanced and open-ended: mathematics, sciences, history, philosophy, literature, economics and other disciplines pursued beyond mere procedural competence, along with serious interdisciplinary work that depends on prior disciplinary knowledge. They should not promise that everyone will achieve the same level. Their obligation should be equal treatment, genuine opportunity, serious teaching and honest standards—not administratively guaranteed outcomes.
This would mean fewer students attending universities simply because the degree is the default route into respectable employment. I see no problem with that. Equality of human worth does not require identical institutional pathways. A master craftsperson, highly skilled technician or successful entrepreneur is not an incomplete academic. The attempt to dignify everyone by sending everyone through university has often achieved the reverse: it has devalued vocational routes, inflated credential requirements and forced universities to provide training they are often poorly designed to provide.
It would also change how universities teach. Content transmission should move to the cheapest effective medium. If a recorded explanation is excellent, use it. If an AI tutor provides better practice, use it. Faculty time should be concentrated where scarce human or expert judgement adds value: critique, supervision, research formation, difficult discussion, oral defence, laboratory judgement, mentoring and the design of intellectually demanding problems.
Students, in turn, should be expected to arrive prepared to work. Education is not a service delivered to a passive consumer. The consumer model is itself part of the problem. A student can purchase access to a teacher, a library, a laboratory and an intellectual community, but cannot purchase the transformation those resources are supposed to enable. That requires effort.
Arum and Roksa (2011) famously raised concerns about limited gains in critical thinking and writing among many students in their study of American higher education. Whatever one thinks of the measurement choices in that literature, the larger challenge is hard to avoid: enrolling in college and receiving a credential are not themselves evidence of substantial intellectual development. If universities want protection from the claim that YouTube has made them obsolete, they need to demonstrate the difference between attendance and education.
This would also be a much more honest compact with students. Instead of promising a credential whose value partly depends on employers continuing to demand credentials, universities would have to say: come here because you want to undertake a demanding course of intellectual development that cannot be reduced to the acquisition of information. If you only need occupational competence, there may be faster and cheaper routes. If you only want the signal, we should stop designing an entire educational system around preserving the signal.
That is a smaller claim for the university, but a more serious one.
Conclusion: technology did not abolish education; it abolished the excuse for confusing education with content delivery
Kurt’s comment is right to be impatient with the economics of modern universities. I share the impatience. The abundance of high-quality material online makes the old model of expensive content delivery increasingly indefensible. LLMs make routine explanation, practice, drafting and procedural assistance cheaper still. Government should not protect institutions from those changes, and I would rather see universities shrink than watch them use public subsidy to preserve functions technology can perform better.
But I do not accept that this makes education obsolete.
It makes bad universities obsolete.
It makes the university as information warehouse obsolete.
It makes the university as lecture-distribution system obsolete.
It may make large parts of the university as occupational-training provider obsolete.
And if employers ever stop using degrees as convenient labour-market filters, it may make much of the university as credential factory obsolete as well.
What remains is the harder thing: forming people who can reason, judge, question, create, discriminate between better and worse arguments, recognise the limits of their knowledge and take responsibility for conclusions they can defend. That work may occur in a traditional university. It may occur in a new kind of institution. Some of it may occur through AI. Some may occur in apprenticeships, laboratories, studios, reading groups or independent scholarship. I am not attached to the building.
I am attached to the distinction.
The mistake is to point to Kindle, Khan Academy, YouTube and LLMs and say: there is education. What we actually have is an unprecedented infrastructure for education. That is an extraordinary achievement. It removes scarcity, lowers the cost of explanation and gives motivated people access to resources that previous generations could scarcely imagine.
But infrastructure is not formation.
A gym full of excellent equipment does not make anyone fit. A library full of books does not make anyone learned. An AI capable of answering almost any question does not make its user capable of knowing which questions matter, which answers are wrong, or what should follow from them.
That is why I would remove government funding while refusing the conclusion that universities are necessarily obsolete. Let them face the market. Let training migrate to institutions that train better. Let credentials lose the artificial value created by compulsory credentialism. Let technology absorb every task it can genuinely perform more effectively.
Then see what remains.
If a university has nothing left once the lectures, routine exercises and credential subsidy are stripped away, it probably should close.
If what remains is a community organised around difficult knowledge, intellectual challenge, serious standards, criticism, research, judgement and the formation of independent minds, then the internet has not made that institution obsolete.
It has merely removed its excuses.
References
Arum, R., & Roksa, J. (2011). Academically adrift: Limited learning on college campuses. University of Chicago Press.
Becker, G. S. (1993). Human capital: A theoretical and empirical analysis, with special reference to education (3rd ed.). University of Chicago Press. (Original work published 1964)
Bettinger, E. P., Fox, L., Loeb, S., & Taylor, E. S. (2017). Virtual classrooms: How online college courses affect student success. American Economic Review, 107(9), 2855–2875. https://doi.org/10.1257/aer.20151193
Biesta, G. (2009). Good education in an age of measurement: On the need to reconnect with the question of purpose in education. Educational Assessment, Evaluation and Accountability, 21(1), 33–46. https://doi.org/10.1007/s11092-008-9064-9
Cellini, S. R., & Goldin, C. (2014). Does federal student aid raise tuition? New evidence on for-profit colleges. American Economic Journal: Economic Policy, 6(4), 174–206. https://doi.org/10.1257/pol.6.4.174
Chi, M. T. H., & Wylie, R. (2014). The ICAP framework: Linking cognitive engagement to active learning outcomes. Educational Psychologist, 49(4), 219–243. https://doi.org/10.1080/00461520.2014.965823
Collini, S. (2012). What are universities for? Penguin.
Denning, J. T., Marx, B. M., & Turner, L. J. (2019). ProPelled: The effects of grants on graduation, earnings, and welfare. American Economic Journal: Applied Economics, 11(3), 193–224. https://doi.org/10.1257/app.20180100
Dewey, J. (1916). Democracy and education. Macmillan.
Dynarski, S. M. (2003). Does aid matter? Measuring the effect of student aid on college attendance and completion. American Economic Review, 93(1), 279–288. https://doi.org/10.1257/000282803321455287
Freeman, S., Eddy, S. L., McDonough, M., Smith, M. K., Okoroafor, N., Jordt, H., & Wenderoth, M. P. (2014). Active learning increases student performance in science, engineering, and mathematics. Proceedings of the National Academy of Sciences, 111(23), 8410–8415. https://doi.org/10.1073/pnas.1319030111
Kasneci, E., Seßler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., Krusche, S., Kutyniok, G., Michaeli, T., Nerdel, C., Pfeffer, J., Poquet, O., Sailer, M., Schmidt, A., Seidel, T., ... Kasneci, G. (2023). ChatGPT for good? On opportunities and challenges of large language models for education. Learning and Individual Differences, 103, 102274. https://doi.org/10.1016/j.lindif.2023.102274
Kulik, J. A., & Fletcher, J. D. (2016). Effectiveness of intelligent tutoring systems: A meta-analytic review. Review of Educational Research, 86(1), 42–78. https://doi.org/10.3102/0034654315581420
Labaree, D. F. (1997). Public goods, private goods: The American struggle over educational goals. American Educational Research Journal, 34(1), 39–81. https://doi.org/10.3102/00028312034001039
Lucca, D. O., Nadauld, T., & Shen, K. (2019). Credit supply and the rise in college tuition: Evidence from the expansion in federal student aid programs. The Review of Financial Studies, 32(2), 423–466. https://doi.org/10.1093/rfs/hhy069
Newman, J. H. (1996). The idea of a university (F. M. Turner, Ed.). Yale University Press.
Reich, J. (2020). Failure to disrupt: Why technology alone can’t transform education. Harvard University Press.
Reich, J., & Ruipérez-Valiente, J. A. (2019). The MOOC pivot. Science, 363(6423), 130–131. https://doi.org/10.1126/science.aav7958
Spence, M. (1973). Job market signaling. The Quarterly Journal of Economics, 87(3), 355–374. https://doi.org/10.2307/1882010
Tlili, A., Shehata, B., Adarkwah, M. A., Bozkurt, A., Hickey, D. T., Huang, R., & Agyemang, B. (2023). What if the devil is my guardian angel: ChatGPT as a case study of using chatbots in education. Smart Learning Environments, 10, Article 15. https://doi.org/10.1186/s40561-023-00237-x