Medieval teaching styles are enjoying an unexpected revival in higher education, and not only because instructors are returning to oral examinations as an alternative to AI-generated essays. Even those incorporating AI into research and writing assignments seem to be taking a medieval turn … and not in a good way.
The most obvious feature of Large Language Models (LLMs) is their dialogical structure. Students can enter prompts in a natural flow of conversation and receive surprisingly nuanced answers based on predictive algorithms. While these exchanges promise unprecedented personalization, they also present a practical problem. Educational dialogue is highly contextual, which is why “Socratic” remains perhaps the most overused and underspecified adjective in teaching. It is also why LLMs struggle to move between educational registers. For any given lesson, one might ask whether the LLM should be providing answers or asking questions? Affirming users or challenging them? Acting as the driving agent or the responsive object of human learning? These are not simply technical matters, to be fixed with fine-tuning. When optimizing LLMs for classroom use, one inevitably confronts the question of educational ends: Automated dialogue for what?
At the moment, there seem to be three answers to that question. When incorporated into classroom instruction, LLMs can operate as sources of information, as guides for structured thinking, or as adversarial opponents. It is no coincidence that these categories track closely with the dialogical methods of European scholasticism between the twelfth and sixteenth centuries—respectively, catechisms for entry-level students, disputation for advanced scholars, and open debate for professors. In both eras we find educators using dialogue to solidify authoritative knowledge, which at the same time they narrow through mediocre language, mechanical logic, and abstraction from human concerns. In trying to systematize relationships between words and humans, both medieval scholasticism and today’s automated dialogue sterilize the sources of human vitality.
Consider catechisms, which began as succinct statements of faith for students to memorize but evolved into a more interactive question-and-answer format with the expansion of print technology and the religious upheavals of the 1500s. Here is a typical exchange about teaching from Erasmus’s A Plain and Godly Exposition or Declaration of the Common Creed (1533):
Master: For so it hath pleased god and hath liked him to give his benefits and gifts to one man by another man.
Disciple: What hath it liked him to do so?
Master: First, to the intent that all pride and arrogancy might be excluded, which that spirit the lover of meek and mild minds doth hate and abhor, and secondarily that through doing benefits and good deeds, each to the other, charity and love might be purchased, increased, and nourished among Christian folk.
The purpose of catechisms was to ensure a smooth transfer of information, organizing important terms and phrases into structured paragraphs, even if the result was what the poet Friedrich Hölderlin would later describe as a “sleepy, idle conversation,” in which questions were merely pretext for predetermined answers.
Today, rudimentary uses of LLMs follow a similar script, positioning the user as the student and the algorithm as the teacher while communicating authoritative (or at least plausible) knowledge in concise, easy-to-understand snippets. As Bodong Chen notes, lessons generated through LLMs prioritize “structured repetition of key facts and concepts” and follow a predictable sequence: “an introduction presenting a real-world scenario, followed by direct instruction where teachers define concepts, provide examples, and demonstrate problem-solving methods, concluding with a summary reinforcing key ideas.” While effective for some tasks, this kind of teaching has obvious shortcomings. Even proponents complain that when LLMs “predominantly adhere to a ‘Question-Answering’ paradigm,” they fail to assess student comprehension and do little to promote independent thinking.
Designers have proposed “Socratic” LLMs as a solution. Using augmented-reasoning functions, multilayered prompts, and other forms of priming, programmers guide algorithms to guide students through a particular course of thinking. As one group of researchers puts it, a “Socratic tutor encourages students to explore various perspectives and engage in self-reflection by posing structured, thought-provoking questions.” Perhaps because it is being implemented by engineers, this approach leans heavily on structure. For example, the computer scientist Edward Cheng describes Socratic inquiry as a series of logical pathways–“definition, hypothesis elimination, elenchus, dialectic, maieutics, generalization, and induction”–and praises Socratic LLMs for their “absence of emotions and sarcasm,” which can “alleviate many of the problems associated with human interaction.”
I am not sure if Socrates himself would recognize a dialogue that lacked emotion or sarcasm, but Cheng’s model accords well with the primary teaching technique of medieval universities: the disputatio ordinaria. After recovering Aristotle’s writings on rhetoric and logic from Arabic sources, twelfth-century universities began to systematize formal exchanges on philosophical, theological, and legal doctrines. Students would take opposing sides of a question and retrace the path from canonical authorities to prescribed ends. The goal was to buttress orthodox beliefs through precision and syllogism, extending students’ mastery of deductive logic. The historian Alex Novikoff quotes one professor, Peter the Chanter, who argued that “calm deliberation should be the order of the day, not theatrical debates” (p. 349). The result was an academic idiom that expressed ideas in dry, standardized Latin to avoid ambiguity. While Novikoff disagrees with those who found disputation “pedantic at best, pointless at worst,” the point stands: as a pedagogical exercise, disputation prioritized form and facticity, rising no farther than technical debate.
The least predictable form of scholastic dialogue was the quodlibet or “ask me anything” event, in which the audience could pose questions to disputants (usually more seasoned professors) and have them answered on the spot. Questions tended toward the metaphysical and abstruse– such as “Whether God can bring it about that matter exists without form?”–and one suspects that those posing them might have been less motivated by the impartial pursuit of truth than the opportunity to wrong-foot rivals or superiors. In this sense, the quodlibet positioned the audience outside the traditional teacher-student relationship, encouraging the irony or exteriority necessary to craft difficult questions and evaluate the answers.
We see a similar approach in advanced, adversarial LLM assignments, which encourage students to learn by testing, critiquing, or playing with the interface. An antagonistic stance follows tech industry norms (which often use “adversarial attacks” to test functionality) as well as the impulses of the average adolescent (who, upon signing onto an LLM interface, wastes no time in trying to elicit profanity and social taboos). It also aligns with the “gamification” of education, which equates the learning process with quantitative targets, competition, and bounded creativity. While adversarial dialogue cultivates a certain kind of knowingness, by making the program the object of inquiry, it binds questioners to the very shortcomings that they strive to identify, losing sight of wisdom or character as the ends of learning.
The point here is not that LLMs are useless in the classroom. Rather, it is that specifying their various uses tends to narrow the concept of personalization, prioritizing structure and quantifiable outcomes over students themselves. Technological affordances inevitably condition users’ goals and expectations, and they often substitute diminished educational ends for those that are truly worthy.
Whether in the sixteenth century or today, the ultimate purpose of personalized education is personhood–that is to say, humanism–achieved through the cultivation of virtue, ethics, and aesthetics, which are by nature intersubjective and unquantifiable, emerging through human interaction rather than engineered dialogue. As Goethe wrote, “A teacher who can arouse a feeling for one single good action, for one single good poem, accomplishes more than he who fills our memory with rows on rows of natural objects, classified with name and form. For what is the result of all these, except … that the human figure pre-eminently and peculiarly is made in the image and likeness of God?”
Image via the British Museum.
