Original research judgment
92Deciding which questions are worth years of work — before evidence exists to train on — is the profession's defining act, and generative systems built on the existing literature are structurally weakest exactly there.
The professor's most visible product — the lecture — is precisely the part of the job machines copy best. A recorded course scales to millions, an AI tutor answers at 3 a.m. without office hours, and automated graders already mark code and short answers at many universities. If the job were only content delivery, it would be in serious trouble.
But universities do not pay professors mainly to deliver content; they pay them to produce knowledge, to certify competence, and to reproduce the profession by training researchers. Those functions rest on judgment about what is true and what matters — exactly where current AI, which synthesizes the existing literature rather than extending it, remains a tool in the hands of the person accountable for the answer.
Low
机器可替代的工作占比
Roughly the content-delivery and first-pass-assessment layer of the job is automatable now or soon: recorded and AI-assisted instruction, routine grading, literature triage, administrative reporting. The core — choosing research questions, supervising apprentice researchers, refereeing what enters the canon, and answering for the certification of students — requires accountable expert judgment that no current system can carry. The realistic near-term future is fewer purely teaching roles and more machine leverage per professor, not the disappearance of the chair.
按任务计分，而非头衔。越低越安全。
AI难以取代的职业 →机器夺不走的
Mentorship and doctoral supervision
85Forming a researcher takes years of individually calibrated challenge, encouragement and honest correction inside a relationship of trust; it is apprenticeship, and apprenticeship has never been deliverable by tool.
Accountability for certification
78Degrees are trusted because named, credentialed humans answer for what they attest. Delegating that judgment wholesale to systems that can be gamed would spend the very trust universities sell.
The live Socratic room
68Reading a room, pressing the student who coasts, drawing out the one who hides — the improvised, social craft of live teaching resists automation long after content delivery falls.
Institutional stewardship of fields
60Editing journals, refereeing disputes, setting curricula and standards — the profession governs itself, and self-governance requires members whose judgment peers accept as legitimate.
它们已经在夺的
Lecture content delivery
75Recorded courses, MOOCs and AI tutors already carry introductory instruction at scale; once content is standardized, the marginal lecture is the most replaceable hour of a professor's week.
Literature search and synthesis
72AI systems now draft literature reviews, screen thousands of abstracts and flag relevant citations in minutes — work that consumed weeks of every research project and most of every doctoral student's first year.
First-pass grading and feedback
68Autograders have marked programming assignments for a decade; essay-scoring and feedback systems increasingly handle routine assessment, leaving instructors the contested and high-stakes cases.
Course administration and Q&A
62Georgia Tech's "Jill Watson" answered students' routine forum questions in 2016 without most noticing she was software; scheduling, syllabus queries and deadline logistics are exactly this kind of automatable traffic.
工作如何变化
The flipped classroom becomes default
With content delivery cheap — video, adaptive tutors, AI explainers — class time shifts toward what only presence provides: discussion, problem-solving, peer instruction. The lecture survives, but as one tool among several rather than the job's definition.
AI joins the teaching team
Following Georgia Tech's 2016 Jill Watson experiment, universities now deploy AI assistants for routine questions, drafting feedback and tutoring — while the professor's new task is quality control: auditing the machine's answers and teaching students to interrogate rather than trust generated text.
The role unbundles
Online mega-providers already split the professor's bundle — course design, delivery, assessment, mentoring — across specialists and software. The research professorship persists at one end while a growing family of teaching, design and verification roles emerges from the pieces.
Research itself accelerates
DeepMind's AlphaFold effectively solved protein-structure prediction in 2020, collapsing a decades-old research program and redirecting a whole field's professors overnight. As AI absorbs more routine analysis, the scarce faculty skill becomes posing the next question — and verifying machine-generated answers.
分岔出的新职业
Instructional designer
The specialist who turns subject expertise into structured online and blended courses — one of the fastest-growing roles in education, staffed heavily by PhDs who chose design over the tenure lottery.
Learning engineer / EdTech specialist
Building and tuning the adaptive platforms, AI tutors and analytics pipelines universities increasingly teach through; a hybrid of pedagogy and software engineering that barely existed in 2010.
Science communicator and educational creator
Academically trained explainers now reach audiences no lecture hall ever held — mathematics and science channels run by ex-academics count subscribers in the tens of millions — turning the professor's explaining craft into a standalone career.
Industry research scientist
Corporate AI and biotech labs recruit professors and their doctoral students at multiples of academic pay — Uber hired some forty researchers from Carnegie Mellon's robotics center in one 2015 raid — making industry research the profession's largest adjacent employer.
展望
The tasks most exposed are those that can be fully specified in advance: the standardized lecture, the routine grade, the literature summary. The tasks most protected are the ones that cannot — the unformulated research question, the struggling doctoral student, the judgment call about what a field should count as true.
The likelier disruption is economic rather than technological: automation gives universities new reasons to divide the professor's bundle into cheaper parts, accelerating a shift toward teaching-only and course-facilitator contracts that began decades before AI. The research professorship endures; the question is how many of the surrounding roles keep the title.
Every previous technology aimed at this profession — print, broadcast, the MOOC — was announced as its end and absorbed as its tool. AI is the most capable yet, and the first to reach into assessment and synthesis. But as long as societies want new knowledge produced and credentials worth trusting, someone accountable must profess — and be answerable for it.
相近职业
不限同一领域，六项评分最接近的职业。
🧮
数学家
从巴比伦文士到菲尔兹奖得主，再到AI辅助证明：这门职业把难题一个定理接一个定理地变成永恒的确定性。
抗AI 70 🔭天文学家
丈量宇宙的科学家，从巴比伦泥板到空间望远镜，始终要判断数据中的哪一次闪烁是一项发现。
抗AI 70 🎬电影导演
通过决定每一个镜头、每一次表演和每一处剪辑，把剧本变成一部完成的电影，再说服制片人、制片厂和观众相信这份投入值得那笔预算。
抗AI 73 ⚛️Physicist
Derives and tests the mathematical laws governing matter, energy, space and time, from a lone chalkboard to a 3,000-author particle-collider paper.
抗AI 65 ⚖️法官
社会所信赖的裁决者：从汉谟拉比石碑到AI评分的保释听证，最棘手的争端至今仍要交由一位担责的人来了结。
抗AI 90 🧬Biologist
The scientist who studies life itself, from Linnaeus naming species by hand to editing genomes with CRISPR, still testing every idea against a living organism.
抗AI 64继续探索
继续探索
教育·人文领域的其他职业
🏫
教师
世界最大职业：从苏美尔泥板学校到AI时代教室，一名成人负责把三十名陌生人变成一间会学习的房间。
抗AI 78 📜历史学家
受过训练的往昔审问者——从希罗多德、司马迁到数字化档案库，把脆弱的文献变成可供核实的历史记述。
抗AI 70 🧩心理学家
研究心智与行为的科学家：从1879年冯特在莱比锡的实验室，到今天的治疗室，受过训练去听懂一个人自己尚且说不出口的话。
抗AI 82 🌐笔译员
在语言之间搬运意义的职业——从《七十士译本》、哲罗姆的《武加大译本》到托莱多、纽伦堡的同传厢与机器翻译时代。
抗AI 28 📖图书馆员
记录知识的守护者与向导,从尼尼微的泥板到电子借阅之争——把人类知道的一切加以整理,使任何人都能找到。
抗AI 55