AI Research Atlas
Where the people who do AI research are, and what they are working on.
We index the AI research literature ourselves, from arXiv, OpenAlex, ORCID and DBLP, and every figure here is a count on that index. Pick a country to see its subjects, pick a subject to see where it lives. Each number says how many profiles it stands on, and the aggregates behind the page are free to download.
Where the doctoral pipeline is heading
People who started publishing in the last three years, by subject, against the three years before. The counts are counts. What follows from them is a projection: a doctorate started today defends in three to four years, so the subjects growing here are the ones with candidates in 2029.
3 to 6 years agolast 3 years
Doctorates in progress, by subject
Started, not yet defended. The stock that reaches the market next.
Papers indexed, by year
The corpus behind every count on this page.
How the page is built
The atlas is one HTML file. The data sits inside it as JSON, so there is no API call, no framework and no chart library. What you see is drawn from what already arrived with the page. The one external script is a cookieless page counter that sets no identifier.
No figure on it is typed by hand. A scheduled job rebuilds the index from the public sources, recounts everything, and regenerates this page and the published aggregates in the same pass. When a number moves, it moves because the sources moved. The method page says what each source contributes.
The aggregates are on /data as CSV and JSON under CC BY, with a manifest that carries the profile count behind every row, and on Hugging Face.
What it does not measure
Publication, not research. A person enters the index through a paper. Engineers who ship models without publishing, and fields that publish in closed venues without a preprint, are under-counted. Read the map as the visible literature rather than as everyone working on AI.
A country is an institution. It follows the affiliation on recent papers, so it lags a move by a few months to a year.
Subjects come from the OpenAlex taxonomy. Its buckets are uneven, and a wide label such as Topic Modeling holds a large share of the corpus because of how the classification is cut, not because of what people work on. Subjects overlap and never add up to the total.
Only counts are published here. Every figure is an aggregate over many profiles, and no individual record appears on this page. The index behind it does name people, from those same public sources. What it keeps and how someone gets out of it is on the privacy page.