■ MODERATE RISK ■ Education
No, but the job's content shifts under the professor's feet. Content delivery is the automatable part, and it was never the part that justified tenure — supervision, research direction, and institutional judgment are.
“AI can lecture, but mentoring, inspiring, and grading on a curve still needs a human touch.”
Our AI replacement risk score — how we score jobs
A professor's role bundles three careers: teaching, research, and service. Teaching means designing courses, delivering lectures, setting and marking assessments, and holding office hours. Research means writing grant applications, running a lab or research group, supervising doctoral students, reviewing papers, and publishing. Service means committees, admissions, curriculum design, and administrative work that expands to fill whatever time remains.
AI has already destabilized parts of all three. Lecture content and explanation is something models do tirelessly and personally, at any hour, in any language — a fact students discovered before institutions did. Assessment design has been thrown into crisis, since take-home essays are trivially generatable; universities are shifting toward oral exams, supervised work, and process-based evaluation. On the research side, literature review, first-draft grant text, statistical analysis, and code for experiments all compress dramatically. Administrative drafting does too. The net is that a professor's hours are being redistributed rather than freed.
What holds our score at 32 is that the valuable core is relational and evaluative. Doctoral supervision means shaping a person's research taste over years, knowing when to push and when they need a break, and vouching for them afterwards. Deciding what research question is worth pursuing — where the field's frontier actually is — remains a judgment call informed by community context that models don't participate in. Peer review, tenure decisions, curriculum standards, and academic credentialing require accountable humans within an institution whose entire product is trusted certification. Add the mentorship and inspiration effect that determines whether a student stays in a field at all. The pressures on academia are real, but they're budgetary and structural more than they are about being outperformed by a chatbot.
Automatability: our editorial assessment of current and near-term AI capability
Transformation is happening now in assessment and course delivery, with institutions scrambling to redesign evaluation around AI availability. By roughly 2040 the professorial role looks meaningfully different — less lecturing, more supervision, seminar work, and research direction. Job numbers face pressure from university finances and enrolment demographics more than from AI directly, though online AI-assisted instruction adds competitive strain to teaching-heavy institutions.
Not the role, though it directly competes with one part of it. A model can explain almost any standard topic better and more patiently than a rushed lecture. What it can't do is supervise a doctoral thesis over four years, decide what research question matters, sit on a tenure committee, or provide the credential that makes a degree worth anything. Those are institutional and relational.
Mainly by breaking assessment. Take-home essays and problem sets stopped being reliable evidence of learning, pushing institutions toward oral exams, in-person work, and evaluation of process rather than product. Lectures are also under question, since a student can get the explanation on demand — which raises the value of seminars, labs, and discussion formats that require being in the room.
The risks facing academia are mostly not AI. Funding pressure, enrolment shifts, and the long-standing oversupply of PhDs relative to permanent posts do far more damage than automation. AI adds pressure at the teaching-heavy end and reduces the labour cost of research support. Our score of 32 reflects a role being reshaped rather than one being replaced.
Openly and carefully. It's genuinely strong for literature review, code for analysis, drafting grant boilerplate, and stress-testing arguments. It's unreliable for citations and can produce plausible nonsense with confidence, so verification is non-negotiable. The gain is in reclaimed time — the researchers benefiting most are spending it on problem selection and supervision, not on more drafts.