Loading
Comparison · dataset August 2026

Data Scientist vs Research Scientist: which is more exposed to AI?

A 17-point gap separates these roles — Research Scientist is the more defensible seat.

Data Scientist sits at 65% time-weighted AI exposure against 48% for Research Scientist, a 17-point gap driven by the 54% of data scientist work time that current models can already substitute outright. Research Scientist holds a larger human-critical core — 52% of the role's time sits in work like "conduct laboratory experiments" that models score poorly on. The two sit in different families — Computer & Math and Science & Research — so any move between them is a career change, not a lateral step.

17PP GAP

Seven dimensions, side by side.

METRICDATA SCIENTISTRESEARCH SCIENTISTDELTA
AI exposure65%48%17pp gap
Resilience score64/10074/10010pt gap
Substitutable work time54%34%Fully automatable today
Human-critical work time24%52%Models score poorly here
Median salary$118k$124k$6k apart
10-year growth35%8%Data Scientist
US workforce202k148kBLS OEWS
Task level

What actually creates the gap.

Data Scientists spend 54% of their time-weighted week on tasks a current model can produce end-to-end, against 34% for Research Scientists. The single largest contributor is "clean and transform datasets", graded at 88% and worth 14% of the role's time. That one task accounts for more of the gap than any difference in seniority, tooling, or industry.

Data Scientist
MOST EXPOSED TASKS
  • Clean and transform datasets88% · 14% time
  • Write analysis code and notebooks84% · 16% time
  • Generate charts and exploratory summaries82% · 10% time
  • Build baseline predictive models78% · 14% time
HUMAN-CRITICAL CORE
  • Communicate uncertainty to stakeholders14% · 8% time
  • Frame business and research questions18% · 10% time
  • Decide deployment and governance trade-offs22% · 6% time
Research Scientist
MOST EXPOSED TASKS
  • Literature review and synthesis88% · 18% time
  • Data analysis and statistical modelling82% · 16% time
HUMAN-CRITICAL CORE
  • Conduct laboratory experiments12% · 14% time
  • Peer review and scientific debate14% · 8% time
  • Interpret novel findings and implications18% · 12% time
What transfers

Both roles lean on cognitive, procedural, judgement — that is the part of your experience that travels intact. The real divide is manual: Research Scientists score 52 there against 4 for Data Scientists, a 48-point spread. That is the gap you would actually have to close.

DIMENSIONDATA SCIENTISTRESEARCH SCIENTIST
Manual452
Creative5474
Switching between them
ModerateDIFFICULTY

Research Scientist appears in our dataset as a mapped adjacent career for Data Scientists: the move lowers exposure by 17 points, landing at 48%. Switch difficulty reads moderate — capability profiles are 22 points apart on average.

Score your own exposure in 8 questions →

Common questions.

Is Data Scientist or Research Scientist more at risk from AI?

Data Scientist. It scores 65% time-weighted AI exposure against 48% for Research Scientist — a 17-point gap. 54% of data scientist work time is already fully substitutable by current models, versus 34% for Research Scientists.

Which pays more, Data Scientist or Research Scientist?

Research Scientist, by roughly $6k at the median ($124k versus $118k). Note that the higher-paying role here is also the less AI-exposed one, which matters if you are weighing pay against durability.

Can a data scientist switch to being a research scientist?

Research Scientist appears in our dataset as a mapped adjacent career for Data Scientists: the move lowers exposure by 17 points, landing at 48%. Switch difficulty reads moderate — capability profiles are 22 points apart on average.

Which role is growing faster, Data Scientist or Research Scientist?

Data Scientist, at 35% projected ten-year growth versus 8% — a 27-point difference. Growth and AI exposure are separate signals: a role can grow in headcount while the content of the work is substantially rewritten.

Methodology
Scores are time-weighted across each role's canonical O*NET tasks, graded against current frontier-model capability. How we score.