LO
University lecturer — tasks, one by one
The unit of analysis is the task, not the job title. Each one below carries its direction, whether the judgement rests on evidence or on platform inference, the reasoning, and what it does not establish.
Every task on this page#
Lecturing and teaching
Being augmented✓ Evidence-backedLecturing, running seminars and helping students understand the material.
AI tutors now do part of this work in trials. In a Harvard physics course, students' median learning gains with an AI tutor were over double those in an in-class active learning session, across two lessons; Harvard's CS50 made AI tools available to about 500 on-campus students, with 88% of sampled curricular answers correct. US projections expect postsecondary teaching employment to grow 7 percent from 2025 to 2035.
Short trials in single courses, one with a ceiling effect the authors note; a projection counts jobs, not teaching hours.
Assessment design and integrity
New task✓ Evidence-backedDesigning exams and coursework that test what students can do, and upholding academic integrity.
AI has made much unsupervised assessment unreliable, which creates new work redesigning it. At the University of Reading, 94% of AI-written exam submissions went undetected and on average scored half a grade boundary higher than real students; the UK's quality agency says the outputs of such software cannot reliably be detected.
One university's test in one subject and an agency's advice; neither measures how assessment has changed across the sector.
Marking and feedback
Being augmented≈ Platform inferenceMarking scripts and coursework and giving students feedback.
AI can take part of the marking load with close agreement to human grades: in an ETH Zurich study of physics exams, AI grading reached a coefficient of determination of about 0.91 when handling half of the grading load, with the remaining scripts left to people.
One study of one exam type; it does not measure marking practice or time saved across universities.
Research and writing
Still human-led✓ Evidence-backedDoing research, writing papers and grant proposals, and answering for the results.
Journals keep authorship with people because authorship means answering for the work. The medical journal editors' recommendations say chatbots and other AI tools should not be listed as authors because they cannot be responsible for the accuracy, integrity and originality of the work.
Editorial standards, not law; they govern authorship and accountability, not how much writing AI now assists.
Peer review
Still human-led✓ Evidence-backedReviewing other researchers' papers and proposals.
AI already shapes the wording of reviews, while editors' rules limit what reviewers may hand to it. Between 6.5% and 16.9% of text in peer reviews at several AI conferences could have been substantially modified by language models; at ICLR 2025, 27% of reviewers who received AI feedback updated their reviews. The same recommendations say reviewers must request permission from the journal before using AI to facilitate their review.
Studies from AI conferences, which may not reflect other fields; an editors' recommendation, not a law.