AI in the Data Lab's work
AI has been part of the Data Lab's toolkit since 2020. During the COVID-19 pandemic, the Data Lab used natural language processing to collect daily food prices from online sources. Since then, machine learning, text analytics and, more recently, large language models have become part of products across FAO's statistical work.
Some examples:
| Work | AI method | What it does |
|---|---|---|
| Augmented analytics (LUMINA-AI) | Conversational AI assistant with literature retrieval, on top of Bayesian spatio-temporal models | Maps where and when children are most at risk of acute malnutrition, and explains why |
| AI-assisted reporting (SCOPE Report Generator) | Large language models; retrieval-augmented generation | Drafts the evidence base of agrifood systems country diagnostics for UN country programming, for analysts to finalise |
| Technology and innovation mapping | Fine-tuned large language models; unsupervised clustering | Builds a two-dimensional taxonomy of agrifood patents from more than 3.5 million applications, and classifies FAO's innovation projects |
| Classifying policies and legislation | Machine learning; text mining | Classifies new documents in FAOLEX, which holds texts from more than 200 countries in over 40 languages |
| Food Loss and Waste Database | Natural language processing | Extracts food products, value-chain stages, places and loss quantities from scientific articles and reports |
| Topics Explorer | Topics and sentiment analysis | Tracks the popularity and tone of press coverage of topics linked to the Sustainable Development Goals |
| Subnational vulnerability estimates | AI methods linking satellite and socio-economic data | Identifies vulnerable populations at subnational level where survey data are missing or outdated; piloted in Kenya and Viet Nam |
From data to decisions: where AI helps
Our work serves statistics, policy analysis and programme support. Across all of it, AI helps at four stages, from gathering data to explaining results.
| Stage | What AI does | Our work |
|---|---|---|
| Gather | Finds and extracts data from sources that were never built for analysis | Daily food prices collected online; food loss data extracted from scientific articles and reports |
| Structure | Classifies text and combines sources into comparable data | FAOLEX documents, agrifood patents and FAO projects classified; survey and Earth observation data combined for LUMINA-AI |
| Analyse | Finds patterns that would be hard to see by hand | Patent groups that reveal technology niches; press tone turned into indicators in Topics Explorer; vulnerable populations identified from satellite and socio-economic data |
| Explain and report | Drafts outputs and answers questions in plain language, with experts in charge | Report drafts from the SCOPE Report Generator; the LUMINA-AI assistant explaining model results |
Sharing what we learn
The Data Lab helps colleagues across FAO use AI well. In 2025, with the FAO AI Community of Practice, it co-organised AI Fundamentals, a webinar series for non-technical staff. It covered machine learning, generative AI for text and images, geospatial AI, and the governance of responsible AI adoption.
See Responsible AI for how we develop and apply these methods.
Work with us
The Data Lab works with other FAO divisions and country offices, international organizations and research partners. Get in touch if you have a question that AI could help answer, data that is hard to use in its current form, or an idea for adapting one of our tools to a new country or topic. Contact the Data Lab.