No official statistic shows AI replacing scientists. The U.S. Bureau of Labor Statistics (BLS) projects employment of medical scientists to grow 13 percent from 2025 to 2035, against 3.5 percent for all jobs, and an automated research system that passed a conference workshop review in 2025 worked only on computational experiments and still made citation errors. What the data support is narrower: AI tools take over parts of research tasks, which is exposure, and no source reviewed here records scientists losing jobs, which would be displacement.
Key takeaways
- BLS projects 2025 to 2035 growth of 13 percent for medical scientists, 12 percent for biochemists and biophysicists, and 7 percent for chemists and materials scientists (all jobs: 3.5 percent).
- BLS says its AI exposure categories are not forecasts of employment growth or decline.
- Statistics Canada counted 336,570 full-time equivalent research and development personnel in 2023, up 7,810 from the year before.
- A 2026 Nature study of 41.3 million papers found AI-using scientists published 3.02 times more papers, while the range of topics studied narrowed 4.63 percent.
- The AI Scientist system had one of three machine-written papers accepted at an ICLR 2025 workshop, then withdrew it by prior agreement.
What official projections say about scientist jobs
Government occupation data are the closest thing to a count of scientist jobs. The BLS Occupational Outlook Handbook pages for three science occupations were last modified on August 27, 2026 and cover 2025 to 2035. The table summarizes them.
| Occupation (BLS) | Projected growth, 2025 to 2035 | Jobs in 2025 | Median pay, May 2025 | Openings per year |
|---|---|---|---|---|
| Medical scientists | 13 percent | about 181,000 | USD 103,410 | about 10,500 |
| Biochemists and biophysicists | 12 percent | 35,200 | USD 127,410 | about 2,900 |
| Chemists and materials scientists | 7 percent | about 93,500 | USD 95,060 | about 6,500 |
For comparison, BLS projects total U.S. employment to rise 3.5 percent, from 170.3 million to 176.2 million. The medical scientists page credits demand to an aging population, chronic disease, and research needs such as cancer and antibiotic resistance, and it does not name AI. The biochemist page does not mention AI either, and the chemist page mentions it only as a branch of computer science some theoretical chemists use. Projections are expectations about demand, not guarantees for any person, and most openings come from workers who leave the occupation.
Exposure is not displacement
On August 27, 2026, BLS released a data product that sorts 831 detailed occupations into four relative levels of AI exposure: low, moderate, high, and very high. The BLS exposure page says the levels combine three theoretical measures of what language models could do with two measures of observed use, one based on Anthropic's Claude usage and one on Microsoft Copilot data. It states that an exposure category is not a forecast of employment growth or decline, and that a high rating does not necessarily mean employment will fall. It also notes that the categories do not separate automation from augmentation, and that most of the underlying sources focus on language modeling and reflect capabilities through mid-2023. Exposure answers a narrow question: how many of an occupation's tasks an AI system could do or assist. Displacement requires evidence of lost jobs or hours, and exposure alone cannot supply it. The same distinction runs through the Statistics Canada work on AI and occupations described in another article on this site.
Canada's research workforce
Statistics Canada reported on January 26, 2026 that research and development (R&D) personnel numbered 336,570 full-time equivalents in 2023, a rise of 7,810. A full-time equivalent measures the time actually devoted to R&D, so a person working half a year on it counts as 0.5. The agency's earlier release for 2021 counted 207,420 researchers, 69.4 percent of the 298,850 R&D personnel that year. Its definition of researchers includes graduate students and the managers who plan their work. These series describe size, not AI effects. No Canadian official statistic measuring AI-related job loss among researchers was found for this article.
What an automated research system has done
Sakana AI announced on March 26, 2026 that its AI Scientist system had been published in Nature. The system generates ideas, checks them against the literature, writes and runs experiments, analyzes the results, and drafts a manuscript, then scores it with an AI reviewer. In a test at the ICBINB workshop at ICLR 2025, three machine-generated papers entered blind human review. One earned scores of 6, 7, and 6, an average of 6.33, and the team withdrew it as agreed beforehand with the organizers. The developers list limits themselves: naive or underdeveloped ideas, trouble with methodological rigor, hallucinated or inaccurate citations, duplicated figures in an appendix, and work restricted to computational experiments. Humans built the system, chose the submissions, and withdrew the accepted paper. A workshop is a lower bar than a journal, and the result was not independently replicated in the sources reviewed.
What AI use does to working scientists
A study by Qianyue Hao, Fengli Xu, Yong Li, and James Evans, accepted by Nature and covered by Chemistry World on January 23, 2026, analyzed 41.3 million natural science papers. It identified AI-augmented research with a classifier validated at an F1 score of 0.875. Scientists using AI tools published 3.02 times more papers, received 4.84 times more citations, and became project leaders 1.37 years earlier. At the field level, the volume of topics studied shrank 4.63 percent and engagement among scientists fell 22.00 percent. The pattern is observational, citations are an imperfect measure of quality, and the data show careers changing, not scientists disappearing.
Which scientist tasks the evidence covers
BLS lists the duties of medical scientists in plain terms. The table matches each duty against what this article's sources actually document.
| Duty listed by BLS for medical scientists | What the sources reviewed show |
|---|---|
| Plan and carry out studies on disease | The AI Scientist was limited to computational experiments; laboratory and clinical studies were not tested |
| Analyze samples and data | AI tools are reported to help find patterns in large datasets and automate high-throughput experiments |
| Write grant proposals and seek funding | No evidence reviewed |
| Publish findings and present them | Machine-drafted manuscripts have passed one workshop review but contained citation errors |
| Follow laboratory safety procedures | No evidence reviewed |
FAQ
Does BLS predict that AI will eliminate scientist jobs?
No. The 2025 to 2035 outlook pages for medical scientists, biochemists, and chemists project growth, and they do not name AI as a demand factor. BLS separately says its exposure categories are not employment forecasts.
Can AI write a paper that passes peer review?
One machine-generated paper earned an average score of 6.33 from blind reviewers at a 2025 workshop. That is a single case at a workshop, and its developers describe remaining weaknesses.
Is high AI exposure the same as losing the job?
No. Exposure describes tasks that AI could do or assist. BLS states that a high or very high rating does not necessarily mean employment will decline.
Are there Canadian figures on AI and research jobs?
Statistics Canada publishes research personnel counts, which reached 336,570 full-time equivalents in 2023. No official Canadian figure on AI-related job loss among researchers was found.
Related reading
- AI in science: research and discovery
- Will AI replace most jobs?
- AI, human workers, and job loss
- The case for human intelligence beside AI
More in AI and Society.
Last verified: October 2026
Sources
- Medical Scientists, Occupational Outlook Handbook, U.S. Bureau of Labor Statistics, modified August 27, 2026
- Biochemists and Biophysicists, Occupational Outlook Handbook, U.S. Bureau of Labor Statistics, modified August 27, 2026
- Chemists and Materials Scientists, Occupational Outlook Handbook, U.S. Bureau of Labor Statistics, modified August 27, 2026
- Employment Projections: 2025-2035 Summary, U.S. Bureau of Labor Statistics, August 27, 2026
- Artificial Intelligence (AI) exposure categories, U.S. Bureau of Labor Statistics, 2026
- Research and development personnel, 2023, Statistics Canada, The Daily, January 26, 2026
- Research and development personnel, 2021, Statistics Canada, The Daily, March 12, 2024
- The AI Scientist: Towards Fully Automated AI Research, Now Published in Nature, Sakana AI, March 26, 2026
- Hao Q, Xu F, Li Y, Evans J. Artificial Intelligence Tools Expand Scientists' Impact but Contract Science's Focus, arXiv 2412.07727 (accepted by Nature)
- AI research tools benefit individual scientists but may slow scientific progress, Chemistry World, January 23, 2026