Can Oxford, Cambridge save Harvard from ChatGPT?

Bloomberg has published an opinion piece that says a time-tested tutorial system offers top US universities a way to blunt AI cheating and revive real learning. Caliber.Az reprints the article.

Artificial intelligence (AI) is capable not just of disrupting higher education but of blowing it apart. The march of the smart machines is already well advanced. AI can easily pass standardized tests such as the GMAT (Graduate Management Admission Test) and the GRE (Graduate Record Examination) required by graduate schools. AI received a 3.34 GPA (grade point average) in a Harvard freshman course and a B grade on the final exam of a typical core Wharton Business School MBA course.

What can be done to avoid a future in which AI institutionalizes cheating and robs education of any real content? This question is stirring an anxious debate in the university world, not least in the United States, a country that has long been a pacemaker in higher education and technology, but one that is losing confidence in its ability to combine equity with excellence.

With the return to campus nigh, the Washington Post warns of an autumn of “chaos” and “turmoil.” This debate should also be coupled with another equally pressing one: What does the ease with which machines can perform many of the functions of higher education as well as humans tell us about the deficiencies of the current educational model?

One solution to the problem is to ban students from using AI outright. Sciences Po in Paris and RV University in Bangalore are taking this draconian approach. But is trying to ban a technology that is rapidly becoming ubiquitous realistic? And is it a good preparation for life after university to prevent students from using a tool that they will later rely on in work? The banners risk making the same mistake as Socrates who, in Plato’s Phaedrus, opposed writing things down on the grounds that it would weaken the memory and promote the appearance of wisdom, not true wisdom.

A more realistic solution is to let students use AI but only if they do so responsibly. Use it to collect information or organize your notes or check your spelling and facts. Refrain from getting it to write your essays or ace your tests. But this raises practical questions of how you draw the line.

How do you tell if students have merely employed it to organize their notes (or check their facts) rather than write their essays? And are you really doing research if you get a bot to do all the work and then merely fluff the material into an essay?

The “use it responsibly” argument opens the possibility of an academic future that is a cross between an arms race and a cat-and-mouse game. The arms race will consist of tech companies developing ever more sophisticated cheating apps and other tech companies developing even more sophisticated apps to conceal the cheating. The cat-and-mouse game will consist of professors trying to spot the illicit use of AI and students trying to outwit them.

Neither approach seems to work, particularly for spotting cheating, let alone eliminating it. Open AI, the maker of ChatGPT, unveiled an app that was supposed to expose AI-generated content this January only to scrap it quietly because of its “low rate of accuracy.”

Another company, Turnitin.com, has discovered that bots frequently flag human writing as being AI generated. A professor at Texas A&M, Jared Mumm, used ChatGPT to check whether his students might have been using the system to write their assignments. The bot claimed authorship and the professor held up his students’ diplomas until they provided Google Docs timestamps showing that they had actually done the writing. It turns out that ChatGPT is over enthusiastic in its claims of authorship.

So, what can be done to prevent educational Armageddon? The best answer lies not in fine-tuning machines — the solution to the problems of technology seldom lies in more technology but in adopting a teaching method that goes back to Plato and Socrates and has been perfected in Oxford and Cambridge over the past 150 years: the tutorial method. Call it the Oxbridge solution.

In Oxbridge students meet once a week individually or in a group of two (or on rare occasions three) with their tutors. The tutor sets them an essay question and provides them with a reading list. The students do the necessary reading on their own, write their essays, and then either submit them to their tutors (the preferred method in the days of email) or else read them aloud (the method in my day). The tutors then probe the essays for weaknesses. What did you mean when you said that? What about X, or Y, or Z?

Why didn’t you take Professor Snodgrass’s views into consideration? (Or alternatively, if the student relied too heavily on Snodgrass, why didn’t you recognize that Snodgrass, though a dear colleague, is a blithering idiot?) The tutorial partner is also obliged to join in with the discussion in the same spirit of testing hypotheses, looking for alternative explanations or generally playing with ideas.

The spirit of the tutorial is both gladiatorial and egalitarian. Knowledge is contested. Debate is of the essence. Authorities are there to be dethroned. Tutors happily concede arguments to their pupils if the pupils get the better of them. “A good tutorial should be a sparring match” not a “substitute for a lecture” pronounced Dacre Balsdon, a fellow of Exeter College, Oxford, from 1927 to 1969.

The students’ grade is determined by high-stakes exams that involve writing essays at speed and under exam conditions; these are then marked by an alien caucus of examiners appointed by the university (perhaps Snodgrass will be among them).

The tutors compete to get the best results for their pupils, and the colleges compete to get the best collective performance. There have recently been moves to lighten the burden of examinations — letting pupils type rather than write, and introducing theses as well as examinations. But AI may have the paradoxical effect of strengthening the role of old-fashioned hand-written exams. Sometimes the best way forward is backwards.

It would be hard to think of a system that is better designed to expose the over-reliance on AI. A pupil who had the chatbot compose the essay verbatim — or who had the bot do the reading and simply fluffed up the essay — would immediately be exposed under cross-examination as a fraud.

The point of the essay is not merely to answer the question and get a mark. It is to start a discussion in which your understanding of the reading is examined. Fail to do the reading and you are destined to spend an uncomfortable hour being pulverized by a skillful sparring partner.

Tutorials don’t just expose cheating. They expose the illusion that AI can do the work of real education. Real education is not just about the assembling of facts into plausible patterns. Nor is it about the accumulation of marks and the awarding of certificates. It is about the open-ended exploration of ideas and, as a reward, admission into the world of learning and argument.

The great Oxford historian-cum-philosopher-cum archeologist, R. G. Collingwood, captured the difference between real learning and AI-generated pseudo learning in his 1939 Autobiography, in the context of historical writing. He denounced “scissors-and-paste” history that consisted of the rearrangements of the statements of various authorities as pointless.

The real historian doesn’t engage in such futility. Instead, he concentrates on finding “something that has got the answer hidden in it” and concentrates on getting “the answer out by fair means or foul.” The aim of tutorials is to get beyond “scissors and paste” — the world of AI — and get the answer out by interrogating the literature and debating with fellow scholars.

The (admittedly self-satisfied) history of the University of Oxford (published in eight volumes by Oxford University Press) describes tutorials as “the hyphen which joined, the buckle which fastened senior to junior members.” By fastening senior to junior members, tutorials also add a moral element to education.

This moral element is a safeguard against cheating: There is all the difference in the world between trying to fool an impersonal educational bureaucracy and trying to fool a tutor whom you meet personally in both educational and social contexts. But the tutorial is much more than that — “a gymnasium for the personality,” as the theatre critic Kenneth Tynan put it, or perhaps even “a cure for souls” as the don Kenneth Leys ventured.

The best tutors can serve as both role models and moral guardians. They might also act as life-long mentors, opening doors to jobs, acting as sounding boards, offering advice and getting their proteges out of various pickles.

The opening of doors and unpickling of pickles underlines the ability of the tutorial system to prepare students for later life as well as adorn universities. It teaches people three of the most important skills that they need in most high-profile professions: how to present arguments under pressure, illustrating big points with vivid facts; how to absorb mountains of information in short order; and how to make fine judgments about the plausibility of various explanations. It also teaches people something that is just as useful outside your career as within it: the ability to learn and think independently — to act, as it were, as your own teacher.

The AI revolution may thus have a salutary impact on US education, where the “scissors and paste” approach has conquered even the most elite institutions. American universities emphasize the “sage on the stage” pronouncing from on high (you must wait until graduate school to establish anything like a close relationship with these demi-gods). The transmission of knowledge is tested by routine exams that are usually marked by graduate students, or by multiple-choice questions that can be marked by machines.

Every stage of this process is open to disruption by AI. The lectures can be replaced by better lectures available on the internet. The essays can be churned out by AI. The tests can be taken by machines as well as marked by them. The progressive mechanization of the system by elite professors trying to devote as much of their time as possible to research may finally have reached its Waterloo in the form of AI. The only way forward is to increase the human element in education.

The obvious objection to introducing tutorials into US education is that they are expensive — tutors must devote 12 or more hours a week to teaching and class-student ratios are reduced to 2-to-1. But Ivy League universities make Oxford and Cambridge look like paupers. They can also afford to lavish money on athletic facilities and vast administrative cadres, both of which have nothing to do with education and one of which arguably impedes it.

State universities may have a better case about money — particularly the local universities below the state flagships which specialize in providing a meat-and-potatoes education to less gifted students. But even here AI will demand an increase in the human touch. Flagship universities could introduce tutorials as a reward for the most talented students. Local universities will have to insist that their professors adapt their teaching to the AI age — shifting from lectures to seminars and setting more demanding essays.

American universities became world-beating institutions in the late 19th and early 20th centuries because they combined the best of the two available university systems: Oxford and Cambridge with their residential colleges and tutorial system, and German universities with their obsession with research. Harvard and Yale introduced houses that functioned like Oxbridge colleges and experimented with the tutorial system. Johns Hopkins and the University of Chicago increased the emphasis on research.

The Germanic model eventually won out over the Oxbridge model. Professors were subjected to a regime of publish or perish and thus spent most of their time learning more and more about less and less. Universities became more hierarchical and more bureaucratic: The aim of the ambitious academic was to become a big-name professor who was too busy flying to conferences and cultivating disciples to meet any undergraduates. Many Oxbridge academics looked at these pampered creatures with envy — Max Beloff complained that “we keep our best historians tied to the routine tasks of giving individual tuition to those unworthy of it.” But the price of such pampering was that the pastoral side of universities — mentoring students and shaping their moral lives — was either ignored or left to bureaucrats.

This system not only short-changed the undergraduates who ended up paying more and more for less and less contact with the tenured faculty. It also ended up producing a lot of useless research. Research might be the gold standard of the hard sciences, which end up not only pushing forward the frontiers of knowledge but also producing practical knowledge. But what about literary studies where the primary goal is surely to educate people’s sensibilities rather than produce yet another article for an obscure academic journal? And what about the proliferation of various “studies” whose aim is to promote an ideological agenda rather than either advance knowledge or solve practical problems?

The supposed threat from AI should be treated as an opportunity to recalibrate US higher education away from the research-centred Teutonic model and back to the human-centred Oxbridge model — and away from producing research and back toward training in thinking. The British prime minister Harold Macmillan recounted the educational philosophy of his ancient philosophy tutor at Balliol, J. A. Smith, before the First World War. Smith said that “a few — I hope a very few — will become teachers and dons.”

For the rest, what they would learn at Balliol would be pointless except for one thing — “you should be able to detect when a man is talking rot, and that, in my view, is the main, if not the sole, purpose of education.” There is no better technique for teaching us to recognize the talking of rot than the tutorial system. And there is no time in history, given the proliferation of charlatan politicians, shady intellectuals and dubious management gurus, all empowered by AI bots, when the ability to spot “rot” has been more important.