EXPOSED
Math is the subject where current AI is strongest: problem-set generation, worked solutions, lecture-note drafting, and even much of homework grading are already handled at usable quality, and online courseware plus AI tutors compete directly with intro calculus and college algebra sections that make up the bulk of the teaching load. What persists is live classroom presence, diagnosing why a specific student is stuck, defending grades and academic-integrity calls, and curriculum ownership — plus, for the tenure-track slice, original research and doctoral advising. Employment pressure here comes less from AI replacing the professor than from institutions using AI-assisted courseware to consolidate large service sections into fewer, larger, more adjunct-staffed offerings.
Mixed — a routine tier and a judgment tier. Generating problem sets, worked solutions, and even proof-checking is where LLMs perform best, and multi-section calculus and college algebra courses run on standardized syllabi and item banks that transfer directly to courseware — what holds this at 10 rather than 5 is the live board work, reading a room of 200 students who have gone quiet, and the office-hours diagnosis of a student who can differentiate but cannot set up a word problem.
Some physical or field component. Chalk-and-board lecturing to a scheduled room, proctoring in-person exams, and walking rows during recitation are physical and location-bound, which is why this is not a 2 — but the environment is a controlled campus classroom and the pandemic proved most of these sections can run over Zoom, so it does not reach the teens.
No licence, no signature requirement. No state licence gates postsecondary math instruction; the credential is a PhD or master's plus departmental hiring, and an institution can and does staff sections with adjuncts, ABD grad students, or vendor courseware without any statutory barrier — the 4 reflects accreditation faculty-qualification rules (SACSCOC, HLC) that require a credentialed human of record, not personal liability.
The human relationship is the product. Math is the subject students arrive at already convinced they cannot do, and whether a struggling student returns after failing the first midterm turns on a specific instructor's patience; add letters of recommendation, undergraduate research supervision, and doctoral advising where the advisor's name follows the student for a career — that relationship is the product, held just under the top band because intro service sections are assigned by registrar, not chosen.
Meaningful discretion. Setting the curve, deciding whether a student who bombed the final but nailed everything else passes, and adjudicating academic-integrity referrals for Chegg or ChatGPT-copied solutions are consequential calls you defend to a dean — but rubrics, departmental common finals, and appeals procedures constrain most of it, which puts this at 11 rather than the high teens.
Has AI actually changed your work?