AI for the Geologist: New Discoveries in Old Maps
Mineral exploration is being quietly transformed. Many of the world's significant resource discoveries this decade have not come from new fieldwork but from re-interpreting decades-old aerial geophysics, drill cores and satellite imagery using deep learning. ML models trained on the geophysical signatures of known deposits can re-rank historical survey data and surface targets that human geologists, working with last-generation tools, missed.
For developing economies in Africa and Latin America with vast under-explored territory and underfunded geological surveys, this is a genuinely transformative opportunity — provided the data sovereignty piece is handled. Models are only as good as their training data, and the legacy field data is often held outside the host country, by foreign mining majors who funded historical exploration and have no commercial incentive to share.
The opening, then, is institutional rather than technological. State-led aggregation of geological survey archives — much of which was funded by colonial-era or post-independence government exploration anyway — into open or carefully-licensed datasets that domestic AI labs and joint ventures can train on. Angola's MIREMPET, Mozambique's INAMI, Namibia's Ministry of Mines and Energy, the geological surveys of the SADC region — these archives represent decades of public-funded survey data sitting on tape and in PDFs. Unlocking them as ML training corpora, with appropriate sovereign control over downstream commercialisation, would create discovery pipelines worth orders of magnitude more than the data licences themselves.
The technology side of this is essentially solved. The institutional side — data governance, sovereign IP, royalty allocation models, multi-stakeholder agreements between ministries, universities and operators — is where the real work sits. It is also where Visionologies' interest converges with national strategic interest in a way that few comparable opportunities do.
