Historian · The trained interrogator of the past — from Herodotus and Sima Qian to the digital archive, turning fragile documents into checkable accounts of what happened.
History looks exposed to language models from the outside — the work is reading and writing text, which is exactly what the machines do. The exposure is real for the middle of the workflow: transcription, translation, search and first-pass summarization are automating quickly, and a historian in 2030 will command document volumes no twentieth-century scholar could touch.
But the profession's core transaction runs in the opposite direction from a language model's. A model generates plausible text from patterns; a historian's job is to distrust plausible text — to establish which documents are genuine, what they can honestly support, and what the silences mean. Generative AI floods the world with convincing fakes and fluent, unsourced summary, which makes the verification craft scarcer and more valuable, not less.
A meaningful share of the workflow — transcription, translation, cataloging, document triage, literature summary — is already automating, and routine synthesis writing will follow. What resists is the judgment layer: authenticating sources, weighing conflicting evidence, framing the question, and standing personally behind an interpretation under peer review. The realistic outcome is fewer hours per discovery rather than fewer historians per discovery — with the real employment risk lying in academic budgets, not in the technology.
Scored from the tasks, not the job title. Lower is safer.
Jobs AI cannot take →Deciding whether a document, image or dataset is genuine — and what it can honestly prove — is precisely the skill a world of cheap synthetic documents demands more of. Models trained on text cannot referee the authenticity of text.
Choosing what a body of evidence means, which question it answers, and how it changes the standing debate is a judgment staked on a named scholar's reputation, defended in seminar and peer review.
The overwhelming majority of the world's archival holdings exist only on paper, parchment or tape in physical repositories. Whatever is not scanned is invisible to every model, and someone still has to go, handle, and choose.
Interviewing survivors, earning the trust of communities, and creating new sources where none were filed — the raw material of contemporary and postcolonial history — is human relational work end to end.
Courts, restitution panels and truth commissions accept testimony from historians who can be cross-examined and held responsible. A system that hallucinates citations cannot take the stand.
Handwritten-text recognition — Transkribus and its successors — now reads early modern secretary hands and chancery scripts at scale, collapsing what was months of paleographic labor into hours of correction.
Semantic search across digitized corpora surfaces relevant files from millions of pages in seconds, replacing much of the box-by-box skimming that once consumed the bulk of an archive trip.
Machine translation gives serviceable first drafts across most modern research languages and improving results for older ones, though specialist checking still decides what a charged term meant in its century.
Models summarize secondary literature, draft annotated bibliographies and organize notes competently — with a known failure mode, invented citations, that forces verification of every reference they touch.
Mass digitization plus machine transcription turns centuries of records into searchable text — the Venice Time Machine project set out to do this for a thousand years of Venetian state archives — letting historians ask population-scale questions that were physically impossible before.
Computational analysis of millions of pages — word frequencies, networks, geographies — becomes a standard complement to the traditional deep reading of single documents, and doctoral training is quietly adding code and statistics to Latin and paleography.
Generative models make convincing fake documents, photographs and audio nearly free to produce, at the same moment public trust in institutions is contested. Provenance, chain of custody and authentication — archival science's oldest tools — become front-line civic skills.
Podcasts, YouTube and trade publishing now reach audiences of millions without academic gatekeepers — the UK podcast The Rest Is History became one of the world's most downloaded shows — pulling career incentives, and standards battles, outside the seminar room.
Builds and maintains the corpora, databases and analysis pipelines historical research increasingly runs on — a hybrid scholar-engineer role now hired by universities, national libraries and archives.
Traces the ownership history of artworks and artifacts for museums, auction houses and restitution cases — a specialty that grew sharply after the 1998 Washington Principles on Nazi-confiscated art and remains chronically short of trained people.
Advises studios on period accuracy and story: Assassin's Creed's re-created cities and prestige-TV writers' rooms both retain historians, and the game industry's appetite for credible pasts keeps growing.
Applies source criticism to the present — authenticating images, tracing claims, reconstructing events from fragmentary records — for newsrooms, platforms and open-source investigators such as Bellingcat, where the method is historical even when the events are yesterday's.
The most likely future splits the job title. The archive-and-interpretation core — fewer, highly trained people authenticating sources and framing arguments — persists and gains leverage from the tools. Around it grows a wider belt of historically trained work in provenance, verification, heritage and media that did not exist a generation ago and does not require a professorship.
The genuine threat to the profession is economic rather than technological: academic hiring in history contracted sharply after 2008 across the Anglophone world and has not recovered, and no transcription engine caused that. Whether societies fund the verification of their own records is a political choice the technology only sharpens.
What AI does settle is the direction of value. When fluent text about the past costs nothing to generate, the scarce good is no longer narrative — it is the warrant behind it: the checked source, the named scholar, the footnote that survives being pulled. That has been the historian's product since Thucydides; the machines have simply made the difference visible.
The scholar paid to profess a subject in public — twenty-five centuries of teaching and discovery, from Plato's grove to the tenure track and the AI tutor.
AI-resistant 76 🏫The world's largest profession: from Sumerian tablet houses to AI-era classrooms, one adult tasked with turning thirty strangers into a room that learns.
AI-resistant 78 🧩The scientist of mind and behavior, from Wundt's 1879 Leipzig laboratory to the therapy room — trained to hear what a person cannot yet say themselves.
AI-resistant 82 🌐The profession that carries meaning between languages — from the Septuagint and Jerome's Vulgate to Toledo, Nuremberg's booths and the machine-translation age.
AI-resistant 28 📖The keeper and guide of recorded knowledge, from Nineveh's clay tablets to the e-lending fight — organizing what humanity knows so that anyone can find it.
AI-resistant 55