Formulating the question worth asking
88Choosing which of a thousand possible research directions is actually worth years of funding and effort is a judgment call current AI systems cannot reliably make on their own.
Physics has an unusual head start in the AI era: much of what a physicist actually produces — a novel hypothesis, a working piece of apparatus, personal responsibility for a claimed discovery — is exactly the kind of judgment call and accountable creativity a language model cannot yet reliably substitute for.
That does not make the job AI-proof. Literature triage, first-draft code and routine data processing are already faster with AI assistance, and the sections below separate what is already shifting from what has, so far, resisted automation for reasons closer to judgment and accountability than to raw technical difficulty.
Moderate
机器可替代的工作占比
A real share of physics' routine execution — literature triage, boilerplate simulation code, first-draft documentation and figure generation — is already faster with AI assistance. What resists automation is deciding which question is worth years of a career, designing an experiment that can actually answer it, and taking personal accountability for a claimed result a machine cannot yet be held responsible for.
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
Designing and troubleshooting real apparatus
84Getting a physical detector, laser or cryostat to actually work, and diagnosing why it doesn't, still depends on physical intuition built from years of hands-on lab experience.
Peer judgment of validity and significance
80Deciding whether a new result is real, important and correctly analyzed is a social and scientific judgment call, not a lookup, made by accountable human reviewers.
Personal accountability for published results
90A named physicist's reputation, and sometimes their career, is on the line for a published claim in a way no AI system currently answers for.
Building and sustaining large collaborations
72Coordinating thousands of scientists across dozens of countries on a shared experiment requires trust and diplomacy no current AI system manages.
它们已经在夺的
Literature search and summarization
70Triaging the daily flood of new preprints on arXiv and flagging relevant results is increasingly handled or accelerated by AI tools.
Boilerplate simulation and analysis code
66Generating standard data-processing pipelines and simulation scaffolding from a described method is a strong fit for current code-generation tools.
First-draft paper writing and figure generation
55Drafting an initial version of a paper's introduction or generating standard plots from processed data is already partly automated in many groups.
Routine data cleaning and calibration processing
74Filtering noise, correcting for known instrumental effects and flagging anomalous data points increasingly runs through automated or machine-learning pipelines rather than manual review.
工作如何变化
AI-assisted literature triage becomes standard
Physicists increasingly use AI tools to survey the daily arXiv firehose and flag relevant work, compressing a task that used to take hours of manual scanning into minutes.
Machine learning as a standard analysis tool
Neural networks now run inside particle-detector trigger systems and sift terabytes of astronomical survey data for anomalies, work no human team could realistically do by hand at that scale.
Fewer solo theorists, more large team science
An increasing share of new physics results, especially in particle and astrophysics, carry hundreds or thousands of co-authors, changing what counts as an individual physicist's contribution and credit.
Industry now absorbs more PhDs than academia can
With tenure-track positions far outnumbered by new PhDs every year, a majority of physics doctorates now build careers in finance, technology or industrial research rather than a university.
分岔出的新职业
Quantitative researcher (finance)
Applies physics-trained mathematical modeling to financial markets at hedge funds and trading firms, one of the highest-paid destinations for a physics PhD.
Machine learning research scientist
AI labs increasingly hire physics PhDs for their training in building and testing mathematical models against messy real-world data, even without a computer science background.
Quantum computing engineer
Builds and calibrates the physical qubits and control systems behind quantum computers, a fast-growing specialization drawing directly on condensed-matter and atomic physics training.
Data scientist
Applies statistical analysis and modeling skills built during a physics PhD to industry problems in technology, healthcare or logistics, a common landing spot outside pure research.
展望
Demand for people with physics training will likely stay strong through the next decade, driven less by any single AI trend than by continued growth in semiconductors, quantum computing and data-heavy industries that value the field's mathematical rigor even where the job title has nothing to do with physics.
The honest read is that AI is a secondary concern next to the field's older, structural problem: physics PhD programs graduate far more researchers each year than tenure-track academic positions exist to hire, and that mismatch, not automation, is what already pushes most physicists toward careers outside the university they trained for.
相近职业
不限同一领域，六项评分最接近的职业。
🧬
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 🏗️建筑师
设计必须屹立不倒、符合规范、让客户满意且造价合理的建筑——一旦出问题，还要承担法律责任。
抗AI 78 💱经济学家
研究稀缺性的学者——从亚当·斯密的制针厂到中央银行的决策室，至今仍被要求预测任何模型都无法完全捕捉的东西。
抗AI 62 🎬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 🧪化学家
亲手制造并测量物质本身的科学家——从塔普提在巴比伦蒸馏香水的年代到今天的机器人实验室，仍是那个判断谱图意味着什么的人。
抗AI 70 👔律师
为客户提供法律意见，起草具有约束力的文件，并在案件进入法庭时为其辩护——若建议有误，还须承担个人法律责任。
抗AI 58继续探索
继续探索
科学·研究领域的其他职业
🧬
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 🧪化学家
亲手制造并测量物质本身的科学家——从塔普提在巴比伦蒸馏香水的年代到今天的机器人实验室，仍是那个判断谱图意味着什么的人。
抗AI 70 📊数据科学家
从数据中寻找规律、构建预测模型——一个2008年才出现的职位名称，建立在三个世纪计数、检验与可视化证据的传统之上。
抗AI 38 🔭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 🧮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