From a comment on the Stat blog:
“I don’t see any reason why getting assistance from an AI should be treated differently than getting assistance from a book or a teacher—it is still the human that incorporates the knowledge and decides how to proceed … Simple prohibition on AI use seems wrong to me. Reading a book can also provide answers that stand in the way of a student figuring something out for themselves, but I don’t think anyone is suggesting we ban reading”.
This is a useful sentiment for clearing up some of the confusion about the use of LLMs in education, because it reminds us that banning resources which help students complete the tasks they are set is in fact entirely ordinary, so that LLMs, in this regard, do not pose an entirely new problem.
Banning resources which help students complete the tasks they are set is in fact entirely ordinary.
An example—this year, I am teaching Plato’s Phaedrus in one of my Greek classes. Every week, the students are set a portion of the dialogue to translate. In class, they read out loud their translations, and we discuss how those translations are to be corrected or refined.
I make it very clear to students that, on the one hand, they are to prepare their translations with reference to dictionaries and grammars but, on the other, they are not to do so by using any previous translations of the text. There are of course scores of translations of the Phaedrus, one of Plato’s greatest dialogues, yet no student has ever objected that the restriction on consulting these is unfair or nonsensical.
Indeed even the consultation of grammars and dictionaries is something that, in the long run, we hope that students will be able largely to forego. Thus I quiz them weekly on grammar and vocabulary, with the aim of reducing their dependence on such reference works.
All of this is entirely routine in the teaching of Latin and ancient Greek, as it is in much other language teaching. Moreover there is a still wider area within which schools and universities routinely debar students from using resources that would be enormously useful to them: exams. Students had long routinely sat exams where they were prohibited from using any or most of the texts and tools that they had needed to study in the relevant course. When the internet came along, it was immediately obvious that this ban would be extended to consulting it during exams.
Of course, all such bans impose on students quite an artificial way of working, and, from the perspective of the product alone, generally make their work worse. When I ask students in an exam to translate a paragraph of Latin or Greek which they have never seen before, I know that most will produce translations far worse than those they could create with access to grammars and dictionaries. By the same token, when I ask students to translate the Phaedrus without using previous translations, or indeed without feeding the Greek into an LLM, I know their translations will be far less accurate than if they were not thus restricted. So why set these restrictions?
It is desirable that all students, personally, know enough Greek to make some sense of Plato themselves.
The immediate answer, of course, is that I am not asking students to translate the Phaedrus because I believe that the world needs a new English translation of the Phaedrus. Even if I did believe this, I would not assign the task to undergraduates with only a few years of Greek behind them. Rather I assign these translations because I believe—as do the students, since they have signed up for the course—that it is desirable that each of them, personally, know enough Greek to make some sense of Plato themselves, rather than relying entirely on earlier translations and interpretations. In other words, the result we are seeking is not really the translation of the Phaedrus at all, but a deepening of each student’s knowledge of Greek and of Plato.
This raises in turn the more fundamental question of why we consider that latter result desirable, a question that can then be extended to any area where we ask students to work without using resources that would improve their performance on the tasks we have set them. This question was in place before LLMs, so that the most obvious novelty, at least in the humanities, brought in by LLMs is to extend the problem to the assigning of essays.
Unlike exams or translations, essays were not an area where students were restricted in their use of resources. On the contrary, one of the tasks expected and rewarded in essays was the appropriate use of relevant primary and secondary material. This is a task that LLMs can now carry out as well as at least a mediocre student. But that merely brings into focus that we did not set that task because we were looking to students—or, more precisely, to students carrying out work below any level we would call “research”—for syntheses that would be of general use. Rather, just as with exams and translations, the work of synthesis was meant primarily to benefit the student herself. Unless we no longer believe that such work is beneficial, there is no reason to allow students to delegate it to an LLM, just because the LLM can do it fairly well.
At this point, some will argue that all the forms of work described above were assigned to students as various forms of cognitive training for real-world tasks that LLMs can now carry out just as well as humans. This is surely to overvalue the capacities of LLMs, and to neglect the related question of their (non-)profitability. Yet there is no denying that, at least for now, there is little mileage left in the argument that the best reason to learn to translate the Phaedrus is so that one can one day earn a living putting together elaborate corporate PowerPoint presentations. Then again, one need hardly take a purely contemplative view of the aims of education to see that this was always an awful way to justify studying the Phaedrus.
LLMs cannot think, whereas students puzzling their way through Plato’s Greek are doing just that, and thinking about something worthwhile. If the world beyond school and university can find no use for young people who have learnt to do this, then the fault lies with the world, not with the LLMs or with Plato.
