Will AI replace political science teachers? Parts of the job are moving, not the whole job. A language model can draft a lecture outline, summarize a new journal article, and score a multiple-choice quiz in seconds. It cannot run a seminar on contested ideas, read a room of tired sophomores, or defend a grade when a student appeals it. That split is what the scores above measure.
Why the seminar room keeps a person in it
Most of the week is talk and judgment. Leading classroom discussion on arguments that have no settled answer is the core of the work. A political science class is not a transfer of facts. Students test positions, get pushed back on, and change their minds in front of other people. Software can prompt. It does not carry the authority that makes a student revise a weak claim.
Grading is the second sticking point. Evaluating student essays and exams means weighing evidence, reasoning and voice, then writing feedback a person will act on. Machines produce feedback quickly, but the professor still owns the grade, the appeal and the academic integrity case behind it. The same goes for advising students on courses, graduate school and careers, where the useful part is knowing the student.
Pay and headcount sit alongside that. The Bureau of Labor Statistics counts about 16,970 of these positions in the United States, with median pay near $98,070 and projected employment change of 2.5% from 2025 to 2035 (BLS, 2025). That is slow growth, not a collapse. Pressure in higher education tends to show up as fewer new tenure-track lines and more adjunct sections, rather than courses taught by nobody.
What AI does, what it helps with, what stays with faculty
Start with the tasks a model can finish on its own. Compiling bibliographies and reading lists, and marking structured quizzes and short-answer items, are the clearest cases. On our task split that group accounts for about 10% of task time. Our coverage figure, which answers “Can AI do it?”, reads 40 on a 0 to 100 scale; the coverage method page explains how task time is weighted.
Then the assisted work. Preparing lecture materials and slide decks, and keeping up with new research in the field, both go faster with a model in the loop. The professor still chooses the argument, the readings and the framing. Tasks of this kind come to roughly 47% of task time.
What is left needs a person in the room or on the call. Facilitating debate, supervising student research and theses, and serving on department and university committees sit here, along with mentoring. That group is about 43% of task time. Nothing physical blocks any of it, which is unusual for a job this resistant: the holdouts are judgment and accountability, not hands.
What the evidence shows, and what it does not
No one has run a clean head-to-head test of AI against political science faculty. Our evidence grade for this occupation prints as D, and a D grade means the quality question has not been measured, so we publish no parity number. We will not guess one.
Two kinds of study would settle it. The first is a blind comparison of instructor-written and model-written feedback on the same student essays, scored by other faculty for accuracy and usefulness. The second is a controlled comparison of learning outcomes in sections taught with and without an AI tutor, holding the syllabus steady. Until that exists, treat any confident claim about machine teaching quality as untested. The quality parity method sets out what moves a grade up, and the full scoring method covers the rest.
When this could change
Most likely between 2034 and 2045 (8 in 10 of our scenarios). The replacement-year method explains what that window is and is not.
Two things could pull the date earlier. One is budget pressure: large introductory sections are expensive to staff, and automated marking plus model-led tutoring is cheap per seat compared with instructor time. The other is institutional buy-in, since no robots or new hardware are needed here. A policy decision and a license are enough to change how a 300-seat course runs.
Two things push the other way. Accreditation rules and faculty governance decide who may assess a student and sign off on credit, and those rules move slowly. And students and parents are paying for contact with a scholar, which is hard to sell as a model subscription. Reputations in this field rest on supervision and mentorship.
Good to know: the realistic near-term risk is fewer new teaching lines and larger sections, not courses without a professor.
How to stay needed in a political science department
Lean into the work that only holds up with a person attached to it. Three places to put your hours: running discussion and simulations that require live judgment, supervising undergraduate and graduate research end to end, and advising students through course choices, funding and careers. Those are the tasks that show up in the needs-a-human group above.
Two skills matter alongside that. First, assessment design that is hard to fake: oral defenses, in-class writing, staged drafts, and work tied to local data or current events. Second, practical AI literacy, so you can set clear course rules, mark model-assisted work fairly, and teach students to check sources instead of trusting output.
What to do: rewrite one assignment this term so a model cannot complete it alone, and say in the syllabus exactly how AI use will be judged.
Related work is worth a look if you are weighing a move. The closest pages are Economics Teachers, Postsecondary, Sociology Teachers, Postsecondary and Political Scientists, which shares the subject matter but trades teaching for research and analysis. You can also see the wider postsecondary teaching family, the education sector page, or put two of these side by side on the compare tool. Our list of jobs that mostly need a person gives the broader context.