Skip to content

Vision

This page explains how fAIr is positioned and the direction it is heading. It is written for partners, contributors, and anyone trying to understand where the work fits.

Positioning

Geospatial data is increasingly generated by AI. Its usefulness depends on the quality of the underlying data and on human validation, which matters most in humanitarian and development contexts, where the most vulnerable places are also the most poorly mapped.

HOT has built open mapping tools and mapping networks since 2010, and develops fAIr together with those communities. In development since 2022, fAIr has taken shape as a platform where geo-AI models are discovered, trained, and validated by the communities who use them.

What fAIr provides

Alongside the models themselves, fAIr provides:

  • Validation and a feedback loop. Predictions are checked by people, and their corrections improve the models.
  • The ability to add any model. Open-source geo-AI models built elsewhere can be brought into fAIr, so mappers can use newer models as they appear.
  • A distributed community. Training and validation are carried out by mappers across the Global South, close to the areas being mapped.
  • Open data. Training and validation datasets are published openly, for example on Hugging Face.

Where fAIr is today

  • 3 organisations have added base models, and 59 of fAIr's 775 users have published 139 fine-tuned models.
  • fAIr has supported 3 activations for flood and earthquake response.
  • 249 users have reviewed fAIr predictions in fAIr, and over 9,300 MapSwipe contributors, counted per project, have validated them across 47 projects.
  • fAIr's training datasets have been downloaded more than 23,000 times on Hugging Face, and datasets with fAIr outputs more than 1,800 times on HDX.
  • Model, dataset, and prediction metadata is cataloged in STAC, so the collection stays open and discoverable.

Figures as of 13 October 2026; the dated breakdown is on Milestones.

Current initiatives

  • Community training data. fAIrSwipe and the Open Mapping Gurus program create open training data with community mappers; see Field projects and Datasets.
  • Crisis response. Prediction outputs are published as open data for flood and earthquake response; see Datasets.
  • Integrating open-source models. An open call brings existing open-source geo-AI models into fAIr, so the platform serves as a shared home for many models.

The structured plan is on the Roadmap.

The direction

fAIr is evolving into an open marketplace of vetted geo-AI models, with the pipelines to keep improving them. It is part of HOT's open, community-driven mapping work, where the same effort that serves humanitarian response also produces open training data that others can reuse.