The Inaugural Dean’s Distinguished Lecture, College of Computing, Data Science and Society, University of California, Berkeley, California. February 4 2025
Rethinking Clark Kerr: The Uses of the University in the Age of Generative AI
By Gerald Chan
In March of 1963, Clark Kerr delivered the Godkin Lecture at Harvard University. While he entitled his first lecture The Idea of the Multiversity, clearly as a rejoinder to John Henry Newman’s classic The Idea of a University, the overarching title he gave to his three-lecture series is The Uses of the University. To this day, these lectures stand as a major landmark in the history of American higher education.
I would like to begin with Clark Kerr’s choice of the word “uses”. The content of his lectures gives little insight into why he chose this word as the rubric for his talks. Perhaps one should not be surprised considering Kerr was an economist and economists are about utility. Was Kerr just defaulting to his mental habit as an economist or was there a specific intentionality in using the word “uses”? We will never know for sure, but by giving this title to his prominent lectures, Kerr has gently banished the thought that the university is so transcendent an institution that it is taboo to talk about its uses. The tone is set that the university is expected to deliver utilitarian values.
In using the word “uses” in plural, Kerr recognized that the university, by now the multiversity, had many uses, but he had no time for the assorted and sundry uses of the university. His view of use was informed by Francis Bacon, a seminal figure in the English Enlightenment. In saying “knowledge is power”, Bacon meant the power to subdue nature. Understanding nature was not enough; that knowledge had to be used to free man from the scourge of nature or to harness the power of nature for the benefit of mankind. In today’s jargon, knowledge should become a driver of innovation to benefit society. On this point, Kerr was entirely aligned with the historical character of the American university. Thomas Jefferson emphasized science and practical knowledge in founding the University of Virginia. It was an intentional departure from the traditional European university. The largest growth spurt in American higher education – the founding of the land-grant universities pursuant to the Morrill Act of 1862 – was decidedly use inspired, hence its emphasis on teaching, research and community service in agriculture, engineering and military science. Kerr was ahead of his time in coining the term “the knowledge industry” and foresaw the university at the center of it.
Mid-century America was a picture-perfect depiction of how the university can be of use to society. World War II was won by science and technology. From the radar to the Bomb, research done in universities helped America win the War. In the peace time that followed, university research was the wellspring from which flowed a steady stream of groundbreaking advances — from plastics to electronics, from the polio vaccine to organ transplantation, from computers to space exploration. Never in human history has there been a time when the standard of living of a country was so elevated by research and innovation, nor has there been a material prosperity that was so broadly shared by all of society. Society at large benefited from the university’s work and therefore had a stake in the health of the higher education sector. Add to this the Sputnik moment. America’s answer to that crisis was education. Congress swiftly passed the National Defense Educational Act in 1958 to strengthen education in science, mathematics and foreign languages. University education became a critical buttress to national security.
From the Second World War through the Cold War, there was a bond between the American people and the American universities, an implicit social contract. The federal government poured money into research grants for universities. State governments paid for the operation of their state universities. The GI Bill, one of the most creative social innovations of all times, thrust myriads of American young people into the universities. From 1960 to 1970, enrollment in American universities doubled from 3.6 million to 7.5 million.
This bond was shattered by the campus protests of the Free Speech Movement, the Civil Rights Movement and the Anti-Vietnam War Movement. The trust that Americans had for higher education began to fray. Fast forward to today. In a Gallup Poll published in July 2024, only 36% of Americans had a great deal of confidence in higher education, 32% had some confidence and 32% had little to no confidence. This decline of confidence is true for all Americans irrespective of their political party affiliation, education level, race, gender and age. The main complaints of the public were extreme imbalance of cultural and political views on campus, irrelevance of what was being taught and runaway cost of higher education.
If American higher education were a business, the strategists would say that customer satisfaction is abysmal. Even more worrisome is that the addressable market for higher education is contracting. The fertility rate in America today is half that of 1963. The Great Recession of 2008 coincided with America’s fertility rate falling below replacement level. It is no wonder that many colleges are now struggling to fill their classes. In each of the last ten years, a dozen or two non-profit four-year colleges around the country have ceased operations. The number of closures of for-profit colleges was many times higher.
College closing is usually the end result of a lengthy process of adjustments that finally failed. College administrators see how broken are their value proposition and their business model and therefore the imperativeness of change. The faculty, on the other hand, is often the obstacle to change. In a recently published book, Brian Rosenberg, the former president of Macalester College, talked about the resistance that his faculty put up against change. In resisting the downsizing of a department that had more professors than students majoring in that subject, a faculty member denounced the requirement to balance the budget as capitalist and that the practice should have no place in the functioning of a liberal arts college. It is no wonder that Rosenberg entitled his book Whatever It Is, I’m Against It, subtitled Resistance to Change in Higher Education. Sadly, for many colleges, the handwriting is already on the wall. Demographics is destiny and finance is unforgiving.
In the face of this bleak state of affairs entered generative AI which will surely be both a game changer and an opportunity for every facet of society. Higher education is no exception. When ChatGPT can recall and process more knowledge than any person possibly can, we will have to evolve our thinking on what constitutes an educated person and with that, how do we educate.
How will generative AI change education? As the Internet is about content or knowledge, generative AI is about pedagogy. If the Internet had turned the availability of knowledge from scarcity to abundance for the everyday man, AI is turning pedagogy from one-size-fits-all to personalization. Consider this analogy. In medicine we now practice personalized medicine. All breast cancers used to be thought of as one disease. Today, we have a panel of molecular biomarkers for each patient’s breast cancer. Depending on the biomarkers, we use different drugs to treat each patient’s cancer. If medicine had become personalized, why is education stuck in one-size-fits-all?
At the center of one-size-fits-all teaching is the lecture. Lecturing is an educational modality that is most appropriate when the number of learners is large and the number of available teachers small. The teacher, a scarce resource, must be amortized across a lecture hall full of learners. Lecturing is a teacher-centric modality; it delivers passive learning. A lecture amounts to material taught; it is agnostic to whether the material is learned. Its focus is on the input to the student’s education rather than on the outcome.
The behavior of college students today is revealing. If you sat at the back of a lecture hall and looked at the screens of the students’ open laptops, it is clear that most of them are parallel processing some other tasks while the professor is lecturing. For schools that record their lectures, many students no longer show up to class. In the comfort of their dorm room, they play the video of the lecture at a speed of 1.5X or 2X. Lest you think I am exaggerating, the popular introductory courses of Harvard College such as in economics and statistics have 600 to 800 students enrolled in them. Only 100 to 200 turn up regularly for the lectures. In the Fall term of 2024, the editorial board of the Harvard Crimson, the undergraduate student newspaper, ran two editorials. One is entitled “Harvard is a school. We need to go to class.” It calls for making class attendance mandatory. The other is entitled “Nobody pays attention in class at Harvard. It’s time to change that.” It calls for banning laptops in class. The statement the students are making by their action is that lectures are not where their learning takes place.
There is no shortage of thoughtful critics of lectures as the main vehicle for delivering educating. In 1984, the University of Chicago educational psychologist Benjamin Bloom published a controlled experiment. He showed that in a classroom setting, instead of the whole class paced in lockstep, if each student were required to master the material before moving forward, so-called mastery learning, the performance of the students improved by one standard deviation. This is equivalent to the average student in the mastery cohort outperforming 84% of the students taught in the conventional way.
Furthermore, if the students were afforded one-on-one tutoring, their performance improved by another standard deviation. The average student in this cohort outperformed 98% of the students taught in the conventional classroom. In many subsequent studies, the benefit of one-on-one tutoring was well established even if the observed effect size was not as large as that seen in the Bloom study. The problem, however, is that one-on-one tutoring requires resources that no school can afford. This conundrum has come to be known as the Bloom’s 2 Sigma Problem. No doubt, cost is one major reason why we still practice one-to-many teaching.
The answer lies in the use of technology including AI agents to deliver content. This is typically done in small segments followed by assessment of the student’s mastery of the material. The student can ask questions to the AI agent which acts as a personal tutor. It is constantly analyzing the student’s feedback and accordingly delivering bespoke material to help the student overcome any learning impediments. This way of delivery incorporates both Bloom’s first sigma of improvement coming from mastery learning and his second sigma of improvement coming from one-on-one tutoring. Moreover, the AI agent maintains a memory of all past encounters with a student, knows his learning style and teaches accordingly. If a student has an area of weakness either from a lack of prior exposure or from a mental block, the AI agent can repeatedly reinforce those area of weakness and bring the student through to mastery of the subject.
In the school year of 2023, two instructors of the introductory physics course at Harvard University experimented with students taught either by in-person teaching or by agentic AI tutoring. The outcome data show that with AI tutoring, the students learned faster and the learning gains were double that of in-person teaching. Moreover, the students felt more engaged and motivated. Such early evidence is extremely encouraging.
As to scale, the one school that has used technology to deliver education at scale is Arizona State University. The School has invested massively in educational technology, massive even by private sector standards. When this way of teaching was applied to their introductory biology course, the pass rate jumped to 90% from a historical pass rate in the low double digits, affirming that the kind of effect size seen in the Bloom experiment is indeed possible. Considering ASU admits any Arizona state resident with a high school GPA of 3.0 or better, it is evident that there is a real effect here from the new way of teaching and that they are not merely seeing a selection bias. For sheer scale, ASU now has 9000 biology majors. Using the same way of teaching, ASU is now educating 30,000 students majoring in engineering. The School graduates 7,000 engineers every year and has an excellent track record of their job placement and the salaries they command in the job market.
Impressive as these early results are, we are only in the first inning of agentic AI. People like Bill Gates have predicted that the world will have more AI agents than people. Much of what we do in everyday life will be carried out by AI agents. What may seem as outlandish today will be commonplace in the future. The unlimited scalability of intelligence is what makes agentic AI so powerful. Education is a perfect use case for agentic AI because educational resources around the world are both inadequate and unevenly distributed. High quality educational resources are even more inadequate and unevenly distributed. This shortage has created a perverse perception that in higher education, difficulty of access is an indicator of excellence. A university that denies admission to 97% of its applicants is seen as a better school than a university that rejects only 80% of its applicants because it is a more highly selective school. It is time we change the perception of quality from intake to outcome. I appreciate what Scott Galloway of NYU said in his TED talk, “Higher education is about taking unremarkable kids and giving them a shot at being remarkable.” Deploying agentic AI in education offers the best hope for solving the conundrum of access and quality, a problem that had exercised Clark Kerr and for which the California Master Plan for Higher Education was meant to alleviate.
What then becomes of the professor if his primary task is no longer content delivery? If knowledge transfer were all there is to education, agentic AI would indeed render professors superfluous, but this is far from the case. There is still a great need and plenty of room for professors to have impact on their students. Each professor will have to define his own excellence in teaching which may range from being a learning coach to addressing the higher objectives of education. Along with knowledge acquisition, the students need to make sense and be able to make use of what they are learning. This is where a great professor comes in. He makes the subject matter come alive and inspires the students to love it. He provides context, relevance, applications, dots to connect; he showcases potentialities and gives glimpses into the future. He both excites the learner’s mind and shapes his intellect. He leads the student to think, to analyze, to see relationships, to assemble ideas, to develop larger and richer thoughts and to have his own voice. In so doing, long after the student has forgotten what he learned in the course, or the knowledge he learned having become pedestrian or obsolete, he will be in possession of an intellect that is fertile, supple, excitable and energetic, incisive and discriminating, an intellect that will serve him well in his lifelong learning. All good education should leave behind in the student something that is indelible, something that has become part of him. The impact of a great professor goes far beyond mere knowledge transfer; he changes lives.
The challenge in education is therefore one of knitting together what technology is good at and what humans are good at. Education will always be a human and a social activity. Students need teachers. Students need to come together to learn together, but we must reimagine the classroom, reexamining each aspect of the educational process and redesigning the educational experience and outcome.
Mindful of Clark Kerr’s concern for the risk of the faculty becoming preoccupied with research, funding and prestige and practicing what he called “scandalous neglect of the undergraduates”, I’d like to offer two other suggestions regarding the education of undergraduates.
One is to incorporate more experiential learning into the curriculum. This may be in the form of project-based learning. Industry internship offers great experiential learning. It is a bridge between a student’s life before graduation and after. The university should do more to help students access internship opportunities so that they are not available only to the privileged who have connections. The stellar example of this work is Northeastern University in Boston which has spent years building up connections with industry for securing student internships. These internships, called co-ops at Northeastern, are semester long, full time, and paid positions. I met a young lady majoring in biochemistry at Northeastern. She wanted to work in the biotech industry after college. In the course of her four years at Northeastern, she spent one semester working in a wet lab of a major pharmaceutical company outside Boston. She then had a second co-op in the business development department of a biotech company. By carefully choosing these two co-op experiences, she had acquired a thorough understanding of the biotech industry by the time she graduated.
For those of us who are in the private sector, offering internships is a good thing to do for the young people. It is our partnership with universities to prepare the future workforce. Among my friends who run companies in the Boston area, we vie for Northeastern interns. Any intern that is any good is snapped up with an offer of a permanent job before he graduates. In this day and age when employers are competing for talents, offering internships in fact gives the employer a leg up.
Short of actually working in industry internships, there is a weaker form of experiential learning that can be widely available and that is case studies. Even though the case study method is most practiced in business schools, it should not be thought of as a pedagogical tool only for business schools. It has much more general applications.
Advocates of the liberal arts approach to education have long championed the Socratic method. At its core, the case study method is a Socratic method. Its goal is to stretch the learner beyond the facts he has acquired, and in so doing, develop both the rigor and the creativity of the learner’s intellect. He has to engage with the material critically in order to go beyond it. He has to be original or, if I may borrow a term from people who work in entrepreneurship and innovation, ideate, which is the identification of unmet needs or unsolved problems and coming up with novel solutions. Such ideation leads to a course of action. It may be in the form of a business plan which forms the basis for a startup. This process is really no different from what we call hypothesis generation in science. Based on the hypothesis, one designs experiments which may lead to publishable results. This is how scientific progress is made. Whether it is formulating a business plan or planning scientific experiments, the principle is the same. The goal is for the student to go beyond the learned material, to go from the historical to the original, from the known to the unknown, from the bounded to the unbounded.
What I am suggesting here is that the case study method can be a tool to bring us back to the essence of a liberal arts education. It is a rejection of the bastardized interpretation of a liberal arts education to mean anything that is non-vocational. It is also a rejection of the assertion that a liberal arts education can only be delivered in the domain of the classics or the humanities. A liberal arts education is not a matter of subject matter or field of study. It is a pedagogy with the aim of developing the learner’s intellect. A former Dean of the Harvard Business School once said to me that a HBS education is a liberal arts education in business. Similarly, the Dean of Yale Medical School recently said to me that a Yale medical education is a liberal arts education in medicine. I cannot agree with them more. In the age of generative AI, I do not see any less need for students to go beyond mere knowledge acquisition to learning how to think, but we must enlarge the tent of what we call a liberal arts education.
My second suggestion is that the faculty ranks in universities need qualitative diversification. There should be a greater porosity for people to move between academia and industry. Historically, that flow has been overwhelmingly one-way from academia to industry. We must have workable mechanisms for people from industry to come and be part of the work of the university and the educational experience of the students. How much richer it would be if a course could be co-taught by a faculty member who has published in the field with a faculty member who has practiced in that field.
Last year, I sat in on a course on AI at Harvard College. It was a general education course intended to introduce undergraduates of all majors to different aspects of AI. One session I attended was on AI and healthcare. There were several mid-career Harvard Advanced Leadership Initiative fellows auditing the class. One was a radiologist. Another had run an emergency department in a hospital. They contributed actively to the class discussion as the professor ran the class in a very interactive way. The perspectives and the richness the ALI fellows brought to the class discussion would not have been possible if the class were taught by the professor alone. The undergraduates were fortunate to have those auditors in the room.
In another case, a colleague in my office told me that the most impactful course she ever took as an undergraduate was a pharmacology course where a scientist from Merck came to talk about the development of the leucotriene receptor antagonist drug for asthma. It was the first time she heard a complete story of how a drug was developed. She thought it was so interesting that she sought a summer internship at Merck. After graduate school, she chose a career in drug development. An encounter of one hour with an industry scientist set her on her career path.
Curricula in universities today tend to be designed by academics for academics even though a tiny fraction of graduates go on to work in academia. Even those graduates who are interested in research will find that research jobs are more abundant in the private sector where financial support for research is much more generous. At least in the fields of life science and AI, industry has moved upstream to do fundamental research that used to be done only in academia. Witness the breakthroughs of Alphafold and the Transformer coming from research done in industry. Working in industry and winning a Nobel Prize are no longer mutually exclusive. If academia does not build more porosity with industry, it risks becoming marginalized.
Several years ago, on one of my visits to Berkeley, my friends on the faculty showed me a food pantry in a building on campus and told me about the surveys showing 20% to 25% of Berkeley students had some form of food insecurity. It is no wonder that students today are anxious about their future. They are very much looking for relevance of what they are learning — to their own educational journey, to their career aspirations, and to the world that they see ahead of them. The university as an insular ivory tower will no longer suffice to serve their needs. A Harvard undergraduate recently talked to me about the absurdity he saw in this insularity. He asked his professor in a computer science course why he was learning the material. He was not being obnoxious or challenging to the professor; he was genuinely looking for relevance. His professor told him he had to learn the material because it is a prerequisite for the next course in the series. The student was stunned by the trivial reasoning of his professor who clearly lived in a closed system where everything is self-referential.
I liken higher education today to a neural network whose error function is trapped in a local minimum. A more global minimum lies somewhere out there, but one will have to venture out to a larger and uncharted space in order to find it. In optimization work, one practice at such a juncture is to inject some random numbers into the process to extricate the neural net from being stuck. If my talk today had injected some random numbers into your thinking, I will have accomplished what I set out to do.