Why this work stays close to people
Anthropology and archeology begin in places a model cannot reach by itself. Someone has to stand in the trench, record what an artifact was lying next to, and notice the layer of soil that changes the date of everything above it. Someone has to sit with a family for months before the useful questions even occur to them. Excavation and site recording, and long-term ethnographic fieldwork, are the spine of the job. Both are physical, social and local.
The paperwork around them is a different story. Transcribing interview audio, translating a document, cataloging finds, matching pottery sherds against a reference set, tidying a literature review, formatting a compliance report: software now does a lot of that quickly. That is the honest shape of the change here. Tasks erode from the middle of the workflow, and the tasks that erode first are often the ones a junior field assistant or graduate student used to be paid to do.
Scale matters too. This is a small occupation. BLS counted about 8,990 anthropologists and archeologists in the US, with median pay of $70,770 (BLS, 2025), and projects employment to grow around 6% from 2025 to 2035. Competition for a handful of openings shapes careers here more than any tool does. You can see how that compares with neighboring research jobs on the full job rankings.
What AI does, what it helps with, and what people keep
Start with the tasks AI can take end to end. Cleaning and coding transcripts, running a first pass on image classification of finds, and converting field notes into a structured database are the clearest examples. Those tasks make up 4% of the share we assign to work AI already handles without a person checking each step.
Next, the assisted middle. Literature searching, statistical analysis of survey data, lidar and satellite scanning for possible sites, and drafting the descriptive sections of a report all go faster with a model in the loop, but a specialist still sets the question and signs off on the answer. That assisted group accounts for 23%. Across all tasks, the Can AI do it? score for this job is 24 out of 100.
Then the part that stays with a person: 73% of scored task time. Excavating and documenting a site, building the trust that makes ethnographic interviews honest, negotiating permissions with communities and landowners, and advising clients or agencies on cultural resource decisions all sit in that group. Those tasks carry consent, liability and relationships, and none of them can be handed to a tool that was not present.
What has actually been tested
Very little, which is the key point. On our Is it better than a person? scale, the evidence grade for this job is D. That grade means there is no direct, published head-to-head test of AI systems against qualified anthropologists or archeologists on their own work, so we publish no quality number for it. Plenty has been written about AI in archaeology and about machine learning in site detection; none of it measures a model against a working professional under the same conditions.
A few studies would settle it. Blind comparisons of AI coding versus expert coding on the same ethnographic transcripts. Artifact classification scored against a curated museum catalog with known provenance. Site reports written from identical field data and graded by reviewers who do not know the author. Reproducible field survey results, where a model’s predicted sites are tested by digging. Until work like that exists, treat confident claims in either direction with care, and read how we score jobs before using the number.
When the picture could shift
Most likely between 2038 and 2058 (8 in 10 of our scenarios). Two things could pull that earlier. Multimodal models keep improving at reading field photographs, drone imagery and scanned archives, which is most of the analysis side of the job. And running those tools costs far less per year than the staff time the page’s cost comparison shows, so budget pressure pushes adoption fast once quality is proven.
Two things hold it back. Roughly a third of this job’s task time has a physical component, and the robot capability it implies is a dexterous humanoid working on uneven, fragile ground. Nothing like that is deployable on a dig today. Second, consent, permits, repatriation law and community agreements are human obligations; an agency or a tribal nation signs with people, not with software. The field is also small and often grant-funded, so new tooling spreads slowly. For how that window is built, see the replacement-year method.
How to stay needed in this field
Lean into the tasks that only hold up with a person present. Run the fieldwork yourself and keep your field documentation good enough that someone else could reproduce your conclusions. Own the community and stakeholder side: permissions, consent, repatriation conversations, and the long relationships that make access possible at all. Take the advisory role on cultural resource compliance, where a client needs a named professional to defend a judgment.
Two skills raise your floor. First, research design: deciding what question is worth asking, what counts as evidence, and where a model’s output would be misleading. Second, practical data stewardship, including supervising AI-assisted transcription and classification, spotting where it quietly fails, and documenting what was machine-generated. Both make you the person who can be trusted with the automated parts rather than displaced by them.
What to do: pick one current project and write down which steps you would let a model draft and which you would never let it decide.
Neighboring work uses much of the same training. Historians share the archival and interpretive core. Sociologists apply similar qualitative and survey methods to living populations. Geographers overlap heavily on spatial data and GIS, which is where archeology’s technical growth sits. You can also read the wider social scientists job family, the education sector page where many of these roles sit, put two of them side by side on the compare tool, or browse the list of jobs that mostly need a person.