Research Scientist
| Median wage (USD) | 103410 |
|---|---|
| Projected employment growth (%) | 13 |
| AI exposure (%) | 57 |
| Maximum exposure (%) | 77 |
| Source | eloundou |
| What AI changes in this occupation | The literature review, the synthesis of publications in the field, drafting a grant application, the first draft of an article with its references, the statistical analysis of the data, generating the figures and writing the protocol are now delegated in a few hours to an assistant: it is the chain that runs from the research question to the submitted manuscript — the one that filled a good part of a researcher's year — that goes first Peer reviewers and funders now receive manuscripts and applications already drafted with AI, which shifts assessment towards checking raw data, reproducibility and the validity of the study design; literature searching, keeping up with the field and the descriptive sections are self-served; and the advantage moves to whoever frames the right question and produces data nobody else has, rather than to whoever writes best Designing an experiment that actually answers the question, handling biological samples and hazardous materials under safety procedures, reading an ambiguous result, carrying responsibility before an ethics board for a protocol run on patients, and signing a conclusion that others will build on remain human — it is the commitment to the truth of the result, in a job where a fabricated data point destroys a career and an institution, that makes the researcher |
| Key skills | Biology and life sciences · experimental design and study protocols · cell culture, molecular biology and laboratory techniques · biostatistics and data analysis (R, Python) · research ethics and regulation (ethics boards, consent, good laboratory practice) · literature searching and critical appraisal · scientific writing and grant applications · leading a research project and team · laboratory safety · generative AI tools for literature review, code, figures and drafting |
| Courses to get ahead | Writing in the Sciences · Coursera (Stanford Online) · About 25 hours in total · 91 · The course that attacks the most exposed task on the card: writing an article and a grant application that others will read and assess. It comes from Stanford, where it is taught to doctoral students, and it is about clarity, structure and the honest use of data — precisely what an assistant now drafts in the researcher's place, and what the researcher must be able to correct rather than produce. Design and Interpretation of Clinical Trials · Coursera (Johns Hopkins University) · About 10 hours in total · 88 · The core gesture of the job as BLS describes it: forming a hypothesis, setting up a trial, analysing the data and concluding on a treatment's effectiveness. This is the part AI speeds up — protocol, power calculation, analysis — without being able to carry responsibility for it, and the course comes from the public health school that shaped the discipline. AI For Everyone · Coursera (DeepLearning.AI) · About 7 hours in total · 84 · For deciding what to delegate and what to keep: the course explains what generative AI can do, where it fails and how to frame a project that uses it. A researcher whose literature review, analysis code and first draft now take hours needs to know what these tools actually produce — and what they cannot commit to. Understanding Medical Research: Your Facebook Friend is Wrong · Coursera (Yale University) · About 17 hours in total · 86 Mathematical Biostatistics Boot Camp 1 · Coursera (Johns Hopkins University) · About 13 hours in total · 82 Biology Meets Programming: Bioinformatics for Beginners · Coursera (University of California San Diego) · About 19 hours in total · 80 |
| Occupations to pivot toward | Natural Sciences Manager · 167220 · 8 · 57 Epidemiologist · 87220 · 19 · 56 Biochemist and Biophysicist · 127410 · 12 · 44 |
| Skills to acquire for the pivot | Eloundou measures the research scientist's exposure at 57% (human β) and up to 77% (human ζ): writing and publishing articles, writing grant applications, the data-analysis and findings-presentation side of methodological research, studying the causes and transmission routes of diseases and evaluating the effects of substances at various levels are largely delegable, but planning and directing studies, preparing and analysing samples, following safety procedures with toxic materials, teaching and operating equipment remain human. Within the repository, 57 puts the researcher 24th of the ninety-five occupations: twenty-three occupations have a higher exposure, sixty-nine a lower one, and two are tied with it (the administrative assistant and the secretary, 57); its nearest neighbours above are the market research analyst (58) and the data scientist (59), below the credit analyst (56) and the actuary (54). Its ζ of 77 places it 40th, tied with the administrative assistant, the secretary and the compliance officer (77), behind the HR manager (80) and ahead of the human resources specialist (76). The SOC code used is 19-1042.00 "Medical Scientists, Except Epidemiologists": this is where O*NET files the titles "Research Scientist" and "Researcher" — the only research profile in the SOC reference that carries that label — and the GPT-4 self-rated series (β 52, ζ 86) is kept in expositionGpt4 and is not displayed, in line with the decision of 21 August 2026. BLS projects 13% growth in medical scientist employment from 2025 to 2035, much faster than the average for all occupations: the 181,000 jobs of 2025 would rise by 22,700, and about 10,500 openings a year are projected over the decade — BLS notes that many of those openings come from replacing people who transfer to other occupations or exit the labour force, such as to retire, rather than from newly created positions alone. O*NET carries the same figure and rates growth "much faster than average (7% or higher)", with the same 181,000 workers in 2025 and 10,500 openings over 2025-2035; its description of the job is that of a researcher who forms hypotheses, develops experiments and runs clinical trials with physicians, and who leads teams of technicians or students. For reference, BLS projects +19% for epidemiologists, +12% for biochemists and biophysicists and +8% for natural sciences managers — the three pivots on this card. The ζ of 77% says a little over three quarters of the job is at least partly transformable; what protects employment is not technical shelter, but the fact that research produces new data and commits to conclusions a model cannot assume. The occupation's median salary is $103,410 a year, the May 2025 figure from BLS published on the Occupational Outlook Handbook page "Medical Scientists" — $49.72 an hour, with 10% of workers below $64,800 and 10% above $177,780. The same release shows the employer effect: $125,700 in research and development in the physical, engineering and life sciences, $106,800 in pharmaceutical and medicine manufacturing, $98,620 in hospitals, $81,910 in medical and diagnostic laboratories and $74,970 in state colleges, universities and professional schools — academia pays the least, even though a large share of funded research happens there. For context, in the same May 2025 snapshot: all occupations stand at $50,980 and the life scientists group at $96,310; natural sciences managers are at $167,220 (+8%), biochemists and biophysicists at $127,410 (+12%) and epidemiologists at $87,220 (+19%) — the three pivots on this card, each taken from its own BLS page. The job usually requires a doctoral or professional degree; BLS states it is typically a Ph.D., usually in biology or a related life science, sometimes a medical degree instead of or in addition to it. O*NET places it in "Job Zone Five: Extensive Preparation Needed" and flags it "Bright Outlook", and its respondents report a doctoral degree required in 36% of cases, a master's degree in 25% and post-doctoral training in 25%. The pay ceiling comes from the pharmaceutical and biotechnology industry or from directing a laboratory, not from the number of publications. Every value on this card comes from a named source; only the task shares, derived from O*NET importance scores, and the course scores are editorial estimates, documented in _ponderation. |
| GPT-4 exposure — beta / zeta (%) | 52 · 86 |