Computer and Information Research Scientist
| Median wage (USD) | 140300 |
|---|---|
| Projected employment growth (%) | 22 |
| AI exposure (%) | 50 |
| Maximum exposure (%) | 77 |
| Source | eloundou |
| What AI changes in this occupation | Conducting logical analyses of business, scientific, engineering and other technical problems, formulating mathematical models of problems for solution by computers (65 out of 100), designing computers and the software that runs them (66), assigning or scheduling tasks to meet work priorities and goals (69), maintaining network hardware and software, directing network security measures and monitoring networks to ensure availability to system users (67), evaluating project plans and proposals to assess feasibility issues (62) and developing performance standards then evaluating work in light of them (54) are the O*NET profile's six most delegable tasks, and six tasks of modelling, software writing, scheduling and monitoring. Work activities confirm it: working with computers (99 out of 100), getting information (91), making decisions and solving problems (91), analysing data or information (89), updating and using relevant knowledge (89), documenting and recording information (84) and processing information (83). Among skills, systems analysis (69), programming (66), systems evaluation (66) and mathematics (60) come right after judgment, critical thinking and complex problem solving (75). BLS describes the gesture itself: "develop new computing languages, software systems, and other tools to improve how people work with computers", "design and conduct experiments to test the operation of software systems, frequently using techniques from data science and machine learning" and "analyze the results of their experiments". The mathematical models, the algorithmic options, the code, the test benches and the project documentation all move to the tool. BLS puts the median wage for computer and information research scientists at $140,300 in May 2025 — $67.45 an hour —, under $82,200 for the lowest 10% and over $230,630 for the highest 10%, against $109,470 for computer occupations as a group and $50,980 for all workers. The best-paying industries are software publishers ($211,270), computer systems design and related services ($162,760), research and development in the physical, engineering and life sciences ($141,830), the federal government excluding the postal service ($124,870) and state colleges, universities and professional schools ($85,460): the federal government employs 31% of the 38,600 jobholders, ahead of computer systems design (16%), research and development (14%), state universities (7%) and software publishing (3%). The projection is +22% between 2025 and 2035 — 38,600 jobs in 2025, 47,000 in 2035, +8,400 — against +6% for computer occupations as a group and +3% for all occupations, with about 2,900 openings a year, mostly replacing workers who leave. BLS ties that growth explicitly to three drivers: the expertise needed to create new technologies related to artificial intelligence, the rapid growth in data collection by businesses, which feeds data-mining services, and a growing emphasis on cybersecurity, these scientists studying innovative ways to prevent potential cyberattacks. Entry is gated by the degree — a master's in computer science or a related field at minimum, a Ph.D. preferred by some employers, a bachelor's sufficient for some federal government jobs —, with no related work experience and no on-the-job training required according to BLS; advancement leads to computer and information systems manager roles. O*NET places the profile in Job Zone Five — extensive preparation, the most demanding level on the scale —, flags it as Bright Outlook and dates it 2026, and 42% of its respondents report working more than 40 hours a week. Problem analysis, invention and relationships remain human: analysing problems to develop solutions involving computer hardware and software (81 out of 100), applying theoretical expertise and innovation to create or apply new technology, such as adapting principles for applying computers to new uses (72), meeting with managers, vendors and others to solicit cooperation and resolve problems (66), participating in multidisciplinary projects in areas such as virtual reality, human-computer interaction or robotics (59), consulting with users, management, vendors and technicians to determine computing needs and system requirements (59) and directing the daily operations of a department while coordinating project activities with other departments (58) are the tasks the Eloundou study leaves untouched, up to 77% of the occupation at most. The working pattern confirms it: 100% of O*NET respondents use e-mail daily and 64% the telephone, 94% work indoors in an environmentally controlled setting, 54% spend their time sitting continually or almost continually, 62% are in constant contact with others and 51% rate working in a group or team as extremely important, 54% have a lot of decision-making freedom and 48% a lot of freedom over their tasks, priorities and goals. The top-rated skills are complex problem solving, critical thinking and judgment (75), then active listening, reading comprehension and systems analysis (69). The flip side is accepted: 58% of respondents rate the job "moderately automated", 17% "slightly automated" and 25% "not at all automated" — an occupation whose very purpose is to automate, which explains a moderate 50% exposure rather than untouchability; 34% consider that a mistake not easily corrected would have very serious consequences and 48% carry high responsibility for the work results of other people. |
| Key skills | Core computer science: algorithms and complexity, data structures, graph theory, computability, compilers, formal languages, computer architecture, operating systems, networks, databases, distributed and parallel computing, quantum computing · mathematics: linear algebra, analysis, probability and statistics, optimisation, numerical analysis, information theory, mathematical modelling · artificial intelligence and machine learning: supervised and unsupervised learning, neural networks, deep learning, natural language processing, computer vision, reinforcement learning, model evaluation and bias, large language models · research: hypothesis formulation, experiment design, evaluation protocols, statistical methods, reproducibility, literature review, paper and conference writing, peer review, research ethics · software engineering: system design and architecture, programming (Python, C/C++, Java, Rust, SQL), test-driven development, version control, continuous integration and deployment, cloud computing, high-performance computing · data: exploration and cleaning, warehousing, pipelines, visualisation, experiment reproducibility, personal data protection · cybersecurity: threat models, cryptography, system and network security · research management: proposals and grant applications, cost and schedule estimating, leading projects, supervising students and interns, multidisciplinary collaboration · communication: explaining research to a non-technical audience, conference presentations, technical reports, working in distributed teams · generative AI tools for writing and reviewing code, generating tests, reviewing literature, exploring algorithms and drafting technical documentation |
| Courses to get ahead | Machine Learning Specialization · Coursera (Stanford Online · DeepLearning.AI) · ≈ 2 months at 10 hours a week (33 hours) · 90 · Conducting the logical analyses that formulate the mathematical models of a problem for solution by computer (65 out of 100) and designing the software that runs them (66) are two of the profile's six most delegable tasks, and they are exactly the two gestures this specialisation installs: regression, classification, neural networks, model evaluation and software development practices. BLS ties the occupation's +22% growth explicitly to the expertise needed "in the creation of new technologies related to artificial intelligence"; being able to judge a model instead of trusting it is what separates the researcher from the tool user, in an occupation where 34% of O*NET respondents consider that a mistake not easily corrected would have very serious consequences. Algorithms, Part I · Coursera (Princeton University) · ≈ 5 weeks at 10 hours a week · 88 · The logical analysis of problems (65) and the evaluation of project plans and proposals for feasibility (62) require knowing what an algorithm actually costs: the course covers amortised analysis, sorting, priority queues, search trees and graphs, together with the performance bounds that come with them. That is the material that lets a researcher settle a proposal or an architecture on measurable facts rather than intuition — and the profile's top-rated skills after judgment are precisely systems analysis and systems evaluation (69 each). Computational Thinking for Problem Solving · Coursera (University of Pennsylvania) · ≈ 2 weeks at 10 hours a week · 86 · Analysing problems to develop computing solutions (81 out of 100) and applying theoretical expertise and innovation to create or apply new technology (72) are the two tasks the profile leaves most clearly human. The course works on problem decomposition, pattern recognition, abstraction and algorithm design — the approach that comes before the tool and that the tool does not choose for you. It is also how BLS describes the occupation: "explore problems in computing and develop theories and models to address those problems". Deep Learning Specialization · Coursera (DeepLearning.AI) · ≈ 3 months at 10 hours a week · 88 Understanding Research Methods · Coursera (Queen Mary University of London) · ≈ 6 hours of workload · 84 Prompt Engineering for ChatGPT · Coursera (Vanderbilt University) · ≈ 2 weeks at 10 hours a week · 83 |
| Occupations to pivot toward | Computer and Information Systems Manager · 175140 · 16 · 56 Computer Hardware Engineer · 161740 · 9 · 32 Data Scientist · 120230 · 35 · 59 |
| Skills to acquire for the pivot | Eloundou measures the computer and information research scientist's exposure at 50% (human β) and up to 77% (human ζ): conducting the logical analyses that formulate the mathematical models of a problem for solution by computer (65 out of 100), designing computers and the software that runs them (66), assigning or scheduling tasks (69), maintaining network hardware and software and monitoring networks (67), evaluating project plans and proposals to assess feasibility (62) and developing performance standards then evaluating work in light of them (54) are the O*NET profile's most directly delegable tasks, while analysing problems to develop solutions involving computer hardware and software (81), applying theoretical expertise and innovation to create or apply new technology (72), meeting with managers and vendors to solicit cooperation and resolve problems (66), participating in multidisciplinary projects in virtual reality, human-computer interaction or robotics (59), consulting with users and management to determine computing needs (59) and directing the daily operations of a department (58) remain human. Within the repository, 50 puts the computer and information research scientist 37th of the one hundred and four occupations: thirty-six occupations have a higher exposure — from the translator (84) to the psychologist (51) —, four are tied with it (the financial analyst, the underwriter, the general office clerk and the administrative services manager, 50) and sixty-three have a lower one, the nearest being the lawyer (48). Its ζ of 77 places it 40th: thirty-nine occupations have a higher ζ — from the writer and author and the cybersecurity analyst (100) to the general office clerk (79) —, five are tied with it (the administrative assistant, the secretary, the compliance officer, the research scientist and the physicist, 77) and fifty-nine have a lower one, the nearest being the human resources manager (76). The SOC code used is 15-1221.00 "Computer and Information Research Scientists": it is confirmed by the occupation's O*NET profile — exact code and title, updated 2026, Bright Outlook flag, Job Zone Five —, by the BLS projections table, which carries the line "Computer and information research scientists" under code 15-1221 (38,600 jobs in 2025, 47,000 in 2035, +22%), and by the Eloundou dataset, which carries a line titled exactly "Computer and Information Research Scientists" under that code; O*NET files under it the reported titles "Computer Scientist", "Computer Specialist", "Control System Computer Scientist", "Research Scientist" and "Scientific Programmer Analyst", and keeps computer systems analysts (15-1211.00), information security analysts (15-1212.00) and software developers (15-1252.00) separate. BLS puts the median wage for computer and information research scientists at $140,300 in May 2025 — $67.45 an hour —, against $109,470 for computer occupations as a group and $50,980 for all workers; the lowest 10% earn less than $82,200 and the highest 10% more than $230,630. The projection is +22% between 2025 and 2035 — 38,600 jobs in 2025, 47,000 in 2035, +8,400 —, against +6% for computer occupations and +3% for all occupations, with about 2,900 openings a year, mostly replacing workers who leave. The best-paying industries are software publishing ($211,270), computer systems design ($162,760), research and development in the physical, engineering and life sciences ($141,830), the federal government ($124,870) and state universities ($85,460), the federal government alone employing 31% of the workforce, well ahead of computer systems design (16%) and research and development (14%). Entry is gated by the degree — a master's in computer science or a related field at minimum, a Ph.D. preferred by some employers, a bachelor's sufficient for some federal jobs —, with advancement leading to computer and information systems manager roles. The working setting is decidedly sedentary and collaborative: 100% of O*NET respondents use e-mail daily, 94% work indoors in an environmentally controlled setting, 54% spend their time sitting continually or almost continually, 62% are in constant contact with others, 51% rate teamwork as extremely important, 54% have a lot of decision-making freedom and 48% a lot of freedom over their tasks and priorities; 58% rate the job "moderately automated", 17% "slightly automated" and 25% "not at all automated". This card's three pivots appear in the similar-occupations table on the occupation's OOH page: the computer and information systems manager ($175,140, +16% between 2025 and 2035, human β 56), the computer hardware engineer ($161,740, +9%, human β 32) and the data scientist ($120,230, +35%, human β 59). The median salary shown is $140,300 a year, or $67.45 an hour. In the repository of one hundred and four occupations it ranks 6th: five occupations pay more, from the physician ($239,200) to the human resources manager ($149,280), and the computer and information research scientist is ahead of the data engineer ($139,500), the pharmacist ($137,480), the DevOps engineer and the software architect ($135,980) and the software developer ($133,080), while the repository's median salary is $76,820 — the computer and information research scientist sits 83% above it. The +22% growth is the repository's fourth highest: three occupations grow faster (the nurse practitioner, 40%; the data analyst, 35%; the data scientist, 34%), none at the same rate and one hundred more slowly, for a median of 4%. The gap with neighbouring occupations is a matter of the nature of the work rather than pay: the computer and information systems manager pays 25% more but is more exposed to management (β 56), the computer hardware engineer pays 15% more and grows only +9% while being markedly less exposed (β 32) because the work touches silicon and physical prototypes, while the data scientist pays 14% less and grows +35%, with a higher exposure (β 59) because the material is almost entirely text and data. Progression runs through the doctorate and publication, through specialisation (machine learning, distributed systems, cybersecurity, quantum computing, robotics) and through leading research projects, which AI-assisted code generation does not replace. Every value on this card comes from a named source; only the task shares, derived from O*NET importance scores, and the course scores and durations are editorial estimates, documented in _ponderation. |
| GPT-4 exposure — beta / zeta (%) | 63 · 88 |