University lecturer

United States. The statistics bureau projects that overall employment of postsecondary teachers will grow 7 percent from 2025 to 2035, much faster than the average, from 1,378,200 jobs, with about 103,300 openings a year. The page does not mention AI. It counts jobs for one country, not tasks.

A named person with standing publicly predicted something, on a date, in an attributable statement. It is recorded so that who said what, and when, stays checkable — and it never moves a task's assessment, because a prediction is not an observation. Its value arrives later: the record sits on the same page as the evidence about that occupation, so anyone reading the forecast reads the record of what happened next beside it. That is the reckoning; this site publishes no verdict on whether a forecast came true.

Medical journals that follow the ICMJE recommendations. They say chatbots such as ChatGPT should not be listed as authors because they cannot be responsible for the accuracy, integrity and originality of the work, responsibilities required for authorship, and that reviewers must request permission from the journal prior to using AI technology to facilitate their review. It is an editorial standard for medical journals, not law.

Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.

United States, one undergraduate physics course. In a randomised crossover design over two lessons, the median learning gains relative to the pre-test baseline were over double for students using a purpose-built AI tutor compared with an in-class active learning session; the authors note a ceiling effect in post-test scores. It is two lessons in one course; the authors declared no competing interests.

A demo, benchmark or paper shows the task can be done. Updates what the technology can do — not what employers will do.

An AI conference's review process in 2025. In a randomised study, an LLM-based agent gave reviewers feedback on their reviews; the paper reports that 27% of reviewers who received feedback updated their reviews, and that over 12,000 feedback suggestions were incorporated. The authors built the system they evaluate, and it ran at one conference.

Small-scale trial in a real setting. Tells us the deployment conditions are being tested, not that they hold — so one pilot is never enough on its own; two independent ones are.

Switzerland, physics exams at one university. The study uses psychometric thresholds to decide which answers AI grades and which go to people, and reports that AI can achieve a coefficient of determination of about 0.91 against human grades when handling half of the grading load, and about 0.96 for one-fifth of the load. It is one exploratory study of one exam type.

A demo, benchmark or paper shows the task can be done. Updates what the technology can do — not what employers will do.

United Kingdom, one university's psychology examinations. The researchers submitted AI-written answers to real take-home exams without markers knowing, and report that 94% of the AI submissions were undetected, 97% under a stricter criterion, and that the grades awarded were on average half a grade boundary higher than those of real students. It is one department and one set of exams, with 2023-era models; the authors declared no competing interests.

A demo, benchmark or paper shows the task can be done. Updates what the technology can do — not what employers will do.

United States, Harvard's introductory computer science course. The course staff describe deploying AI-based tools — a tutor-like assistant and code explanation — to approximately 500 on-campus students, and report that 22 out of 25 (88%) curricular answers they checked were correct. The staff are evaluating their own tools, and the paper thanks technology companies including OpenAI and Microsoft for their support.

An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.

Peer reviews at AI conferences (ICLR, NeurIPS, CoRL, EMNLP). Using a population-level estimator, the study reports that between 6.5% and 16.9% of text submitted as peer reviews to these conferences could have been substantially modified by LLMs, beyond spell-checking or minor writing updates. It covers AI conferences only and estimates shares of text, not individual reviewers.

Measured, large-scale use of a tool for real work, where the decision to use it was the worker's rather than an employer's. It is more than a capability record — the work is real, not a demo — and less than a deployment record, because no employer put it into production, required it, or built a process around it. Weighted `cautious`: `automating` means the machine can do the task AND there are adoption signs, and this is an adoption sign — but usage can be experimental, and much of the measurement comes from a party with a stake, so one record is never enough and two independent ones are. Note who is counting. Vendor telemetry sees this directly and sells the tool, so such a record names that stake in its scope; a statistics agency asking firms whether their workers use AI in tasks sees the same channel with no stake at all, and that is the better source where it exists.

United Kingdom, higher education. The agency's advice to its members says the outputs of generative AI software, despite its limitations, cannot reliably be detected, and sets out how providers can reconsider assessment accordingly. It is advice, not a mandate, and it does not measure how assessment has changed.

Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.

If you are starting out, expect AI tutors and grading tools to take part of the routine teaching load, and expect assessment design to become a larger and harder part of the job. Research authorship and review judgement stay with you under journal rules.

Expect to redesign assessment for a world where written work can be generated, and to supervise AI tools in teaching and marking. Your accountability as an author and reviewer is what publishers' rules keep with you.

Stay and strengthen

Stay, and lead assessment redesign

AI-written answers go undetected and cannot reliably be detected; someone has to design assessment that still measures learning.

Reshape the role

Move towards teaching with AI tutors

Trials show AI tutors can raise learning gains, and courses are putting AI tools in front of hundreds of students.

Adjacent move

Move into research integrity or editorial roles

Publishers and journals are writing rules on AI in authorship and review, and need people who can apply them.

Not on present evidence. AI tutors and grading tools do parts of teaching and marking well in trials, and AI-written answers slip through exams. But journals keep authorship and review judgement with people, and US projections expect postsecondary teaching employment to grow 7 percent from 2025 to 2035. The job is shifting towards assessment design and supervising AI.

We do not answer that with a number of years. Watch whether universities start awarding degrees on AI-taught and AI-marked courses without academics answering for them, and whether journals accept AI as authors or reviewers. Today the rules recorded here say no to both.

Partly. In an ETH Zurich study, AI grading of physics exams reached a coefficient of determination of about 0.91 when it handled half of the load, with the rest left to people. Detecting AI-written answers is harder: at the University of Reading, 94% of AI submissions went undetected.

US projections expect postsecondary teaching employment to grow 7 percent from 2025 to 2035, with about 103,300 openings a year. The parts of the job that rules and journals keep with people — authorship, review and answering for assessment — are growing in importance.