AI chatbots are reshaping classroom debates, raising concerns over homogenised discussion
Yale students and academics say AI chatbots are making classroom discussions more polished but less original. A study warns that large language models can homogenise perspectives and discourage independent thought. Universities are responding by shifting to in-class, handwritten and oral assessments, while students acknowledge AI can reduce engagement and encourage intellectual laziness.
Generative AI chatbots are becoming embedded in university learning at Yale, students and academics told CNN, not only for essays and homework but also for real-time seminar participation. Students described classmates uploading readings and PDFs into chatbots before class, and even typing a professor’s question into AI during discussion to produce an immediate response to repeat aloud.
While this can make contributions sound more polished and prepared, some students said seminar conversations increasingly stall or feel flatter, with fewer personal interpretations and less exploratory debate. One student, ‘Amanda’, said she has noticed many classmates arriving with slick talking points but then offering near-identical arguments and phrasing, making discussions feel less distinctive than in earlier years.
Students gave several reasons for leaning on AI. ‘Jessica’, a senior, said she uses it daily, particularly in an economics seminar where the professor cold-calls students, both to digest readings quickly and to help her translate ideas into cohesive sentences when she struggles to phrase her comments.
‘Sophia’, a junior, said some students appear to use AI to draft ‘scripts’ for what to say in class, driven by insecurity about gaps in their understanding. She believes this weakens creativity and the ability to make original connections, replacing genuine engagement with impressive-sounding language.
A Yale spokesperson said the university is aware students are experimenting with AI in the classroom and noted a wider faculty trend towards limiting or banning laptops, using print-based materials, and prioritising direct engagement and original thinking.
The article links these observations to a March paper in ‘Trends in Cognitive Sciences’, which argues that large language models can systematically homogenise human expression and thought across language, perspective and reasoning. The paper’s authors say LLMs predict statistically likely next words based on training data that overrepresents dominant languages and ideas, potentially narrowing the ‘conceptual space’ for how people write and argue.
They warn that models tend to reproduce ‘WEIRD’ viewpoints, Western, educated, industrialised, rich and democratic, even when prompted otherwise, which may make those styles seem more credible and socially correct while marginalising other perspectives.
Researchers also describe a compounding feedback loop. As AI-generated outputs circulate in human discourse and eventually re-enter training data, sameness can intensify over time. Co-author Morteza Dehghani said offloading reasoning to AI risks intellectual laziness and could have broader social consequences, from weakened innovation to greater susceptibility to persuasion.
Educators quoted described both benefits and risks, and outlined practical responses. Thomas Chatterton Williams, a visiting professor and Bard College fellow, said AI can ‘raise the floor’ of discussion for difficult material but may suppress eccentric or truly original ideas, leaving students without a voice of their own or a sense of authorship.
Former teacher Daniel Buck called AI a ‘supercharged SparkNotes’ that can answer virtually any question, making it harder to detect shortcuts and easier for students to bypass the ‘boring minutiae’ where learning takes hold.
He worries that this also undermines relationships with professors and sustained cognitive work. Yale philosophy professor Sun-Joo Shin said model improvements forced her to redesign the assessment. Problem sets now earn completion credit and feedback, while in-class exams, oral tests and presentations carry more weight.
Williams said he has moved from writing to spontaneous, in-class, handwritten work and uses oral exit exams. Students who avoid AI argued that they are still affected by classmates’ reliance on it because it reduces the value and variety of seminar time, while others urged a middle path in which AI is treated as a collaborator, used to critique ideas rather than as a substitute for generating them or doing the reasoning.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!