What the work looks like day to day
The U.S. Bureau of Labor Statistics describes the broader occupation, computer and information research scientists, as people who explore computing problems, build theories and models, design and run experiments, analyze the results, and write papers and present them at conferences. In AI, that loop is the job. You read recent papers, form a hypothesis, write the training or evaluation code, run it, and decide whether the result is real or noise.
A Meta posting for an AI Research Scientist in Paris lists three duties: publish fundamental AI research, release reproducible open-source code, and mentor interns, PhD students and postdocs. The same posting asks for skill in coding, running and evaluating large-scale experiments in PyTorch. Large experiments cost real money in GPU time, so a big part of senior research work is choosing which runs are worth the compute and stopping the ones that are not paying off.
How it differs from applied science and ML engineering
A research scientist is judged mostly on new knowledge: a method, a result, a paper. A Meta research posting frames the goal as long-term research aims broken into intermediate milestones, and asks for experience presenting research to peers. Publication record is often part of the bar.
Applied scientists sit closer to the product. An Amazon applied science listing asked for years of applied research experience and experience building machine learning models for business use. Machine learning engineers lean further toward shipping: training pipelines, serving, monitoring and reliability. The skills overlap a lot. The difference is what you are measured on: papers and new methods, models that move a business metric, or systems that run well in production.
How often is a PhD expected?
Often, at industry research labs. A Salesforce Research entry-level research scientist listing set a PhD in computer science, AI, machine learning or a related field as a minimum qualification. A Meta Paris research scientist posting required a PhD in math, statistics, physics or computer science, plus papers at venues such as NeurIPS, ICML, ICLR and JMLR. Another Meta research posting accepted a PhD or equivalent practical experience, so the rule is not absolute.
BLS gives a broader picture for the whole occupation. It says these workers typically need at least a master's degree, some employers prefer a PhD, and the federal government may hire some candidates with a bachelor's. If you have no doctorate, applied scientist, research engineer and ML engineer roles are the more common way in. A strong public research record helps with any of them.
Where the jobs are
BLS counted about 38,600 computer and information research scientist jobs in 2025. The federal government, excluding the postal service, employed 31 percent of them, ahead of computer systems design firms. That count covers all computing research, not AI alone. BLS projects 22 percent employment growth from 2025 to 2035, with about 2,900 openings a year, and names AI as one source of that demand. Industry AI labs hire far fewer people than product engineering teams do, so expect stiff competition for research titles.
Pay
USD a year, before tax. US figures from BLS Occupational Employment and Wage Statistics, May 2025, moved to 2026 with the Employment Cost Index. These are the nearest BLS series, not AI-specific: each covers the whole occupation, including non-AI researchers and many federal jobs, and leaves out stock and bonus. Refer to the AI Data Scientists salary guide for figures from live adverts.
What you need
- Research degreeMany industry lab postings list a PhD in computer science, machine learning, statistics, math or physics as a minimum. Some accept equivalent practical experience instead.
- Publication recordLab postings often ask for papers at venues such as NeurIPS, ICML, ICLR or JMLR, and for experience presenting research to peers.
- Experiment engineeringYou need to code, run and evaluate large experiments yourself. One Meta posting names PyTorch for this.
- Reproducible codeReleasing clean open-source code and reproducible results is listed as a core duty in some lab roles, and public repositories count as a plus.
- Research judgmentSetting long-term goals, breaking them into milestones, and killing weak ideas early matters as much as raw technical skill.
How to get there
- Years 1-4Bachelor's in a quantitative fieldComputer science, math, statistics or physics. Get research experience with a faculty lab if you can.
- Years 5-9Graduate researchA master's meets the typical BLS entry level. Many industry labs expect a PhD, during which you publish first-author papers.
- During PhDResearch internshipsIndustry labs run research scientist internships for PhD students, for example at Meta and Salesforce Research.
- First roleResearch scientistRun your own experiments, publish, release code, and start mentoring interns.
- Mid-careerSenior or lead researcherSet research direction for a team, settle technical disagreements, and decide where compute goes.
Open now on AutomationRoles.ai
Questions people ask
Is an AI research scientist the same as a machine learning engineer?
No. A research scientist is measured mainly on new methods and published results. A machine learning engineer is measured on systems that train, serve and run reliably in production. The skills overlap, and people move between the two.
Can I become an AI research scientist without a PhD?
Sometimes. Some Meta research postings accept equivalent practical experience in place of a PhD. Others, and Salesforce Research's entry-level listing, make a PhD a minimum. BLS says the broader occupation typically needs at least a master's degree, and the federal government may hire some people with a bachelor's.
What is an applied scientist?
An applied scientist uses research methods to build models for a product or business problem. An Amazon listing asked for applied research experience plus experience building machine learning models for business use. The role sits between pure research and ML engineering.
How much do AI research scientists earn in the US?
The nearest BLS series, computer and information research scientists, has a 2026-adjusted median of about $144,700. The middle half falls between $106,800 and $194,600. The series covers all computing research, not AI alone, and leaves out stock and bonus.
Is there a required certification?
No. No certification or regulator covers AI research work. Hiring rests on degrees, your publication record and how well you do in technical interviews.
Sources
- U.S. Bureau of Labor Statistics: Computer and Information Research Scientists, Occupational Outlook Handbook
- EURAXESS: Meta AI Research Scientist, Core Learning (Paris)
- EURAXESS: Meta AI Research Scientist, CodeGen (PhD)
- EURAXESS: Meta research scientist internship (PhD)
- MIT CSAIL Alliances: Research Scientist, Salesforce Research (Entry Level)
- CMU Parallel Data Lab: Amazon industry job listing
