Everyone talks about the jobs AI is taking. A project published by Brown University on 26 October talks about the jobs AI is made of — and the numbers are the point. Cunyan Ma, a Brown undergraduate, has built a public map of the AI data-labelling supply chain: 13 data-labelling companies, 141 of their corporate customers, and 31 countries containing data-worker sites. It is the first at-a-glance picture of who actually does the work under every chatbot, and where they sit on the map.
🔍 THE BOTTOM LINE: The single largest new job category of the AI boom — data annotation, 282,000 US positions by LinkedIn’s count — is also its most invisible and most contested. Ma’s map makes the geography legible: labelling firms chase “inexpensive labor,” much of it in countries like Kenya, Indonesia and Colombia, while their customers are mostly Western corporations. Read two ways, it is both a labour-rights indictment and a reminder that the AI economy’s biggest employer is human effort — work that did not exist as a job title five years ago.
What the map shows
The project draws on the public DataWork Landscape database of labelling firms and their customers, plus Ma’s own original database of labelling firms’ delivery centres — used as a proxy for where the workers actually are. The lines connect the nodes: a retailer or tech firm outsources annotation to a specialist; the specialist routes it to delivery centres where wages are lowest.
“Companies outsource that work to the Global South — Kenya, Indonesia, Colombia and elsewhere,” Ma told Brown’s news office, describing annotation ranges that include screening AI-produced harmful content to train safety filters. “How that manifests in reality is… folks around the world who have to read paragraph after paragraph of AI-produced child pornography, bestiality, violent content or whatever else, just to try to train AI to be safe.”
Her advisor, Brown computer science professor Suresh Venkatasubramanian, put the framing that matters for jobs coverage: “It’s all humans underneath… The problem with a lot of current discourse around AI is that it seeks to ignore that.”
Why this is a jobs story, not just an ethics story
For career-compass readers the map lands on a number already covered on this site: LinkedIn estimates cited by The Wall Street Journal counted more than 750,000 AI-linked US jobs created since 2023, with 282,000 data annotators as the biggest single category — ahead of engineers. Annotation is entry-friendly by design, which is exactly why it appears in such volume, and exactly why researchers flag its pay and conditions offshore.
Three things follow for anyone thinking about this labour market:
- It is real employment at real scale. The AI boom’s biggest employer so far is not the labs; it is the distributed human effort of teaching models what they are looking at. That is job creation — the category barely existed as organised remote work in 2021.
- The conditions debate is the price-discovery mechanism. Ma’s “AI colonialism” framing is contested — label it as her characterisation, not settled fact — but the wages question it raises is functioning like any labour-market signal: pressure that historically pushes work toward better-documented contracts, higher rates and, in time, specialisation. Annotators who train safety data are doing quality-control-grade judgement work; the market has been slow to price it that way, which is precisely why the pressure exists.
- Visibility is the lever. Ma’s stated hope is that a distributed workforce that “doesn’t even know about their fellow co-workers” can use the map to find each other. Whatever one thinks of unionisation, workers with a public map of their own industry have a better starting position in every negotiation about it than workers without one.
The NZ read
New Zealand has no large-scale annotation industry on Ma’s map, but it has something the 31 countries largely do not: the English-language, education-dense workforce that higher-value AI data work gravitates toward. Global firms increasingly hire expert annotators — teachers marking pedagogy data, lawyers grading legal reasoning, clinicians reviewing medical labels — and pay well above the entry tier. NZ’s recent AI-adjacent hiring signals — agent-orchestration skills up 1,721 percent on Wall Street platforms — point the same direction: the premium sits with people who can judge model output, not merely label it.
The honest caution: none of NZ’s expert-annotation activity is yet visible at industry scale in official labour statistics, and Ma’s map is explicitly illustrative, not comprehensive. Read it as the first structured look at a workforce that is real, growing, and worth watching — the jobs underneath AI are jobs, and the question the map sharpens is not whether they exist but who holds them and on what terms.
❓ FAQ
What does the Brown AI labour map show? Links between 13 data-labelling companies, 141 of their corporate customers and 31 countries containing data-worker sites, combining the public DataWork Landscape database with Ma’s own database of labelling firms’ delivery centres.
How many data annotators has AI work created? LinkedIn estimates cited by The Wall Street Journal count roughly 282,000 US data-annotator positions created between 2024 and 2026 — the largest single category among 750,000-plus AI-linked jobs since 2023.
What is ‘AI colonialism’? A term used by critics, including Ma in this project, to describe the flow of low-paid data-labelling work from Global South countries to serve mostly Western AI companies. It is the author’s characterisation of the pattern, not a neutral industry term.
Can you get a job in AI data annotation without a degree? Yes — annotation is entry-friendly by design, and it is the largest new AI job category. Higher-paid specialist tiers (expert annotators with domain knowledge) are growing alongside it.
📰 Sources
- Brown University — Brown student maps the hidden global labor powering AI (26 October 2026)
- DataWork Landscape — public database of data-labelling firms and customers
- The Wall Street Journal / LinkedIn — AI-linked job creation estimates (2026), as analysed in our 5 October coverage
- Hacker News — community discussion of the Brown project (26 October 2026)
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
