Exploration and target generation is where most day-one AI adoption happens in mining, because the data - drill core, geochemistry, geophysics surveys - already exists in the volumes machine learning needs. This is the instrument layer: automated mineralogy platforms like TIMA, MLA and QEMSCAN, hyperspectral core scanners for clay and alteration mapping, portable XRF and LIBS analyzers, and airborne gravity and magnetics surveys.
On top of that instrument layer sits a growing set of prediction models: rock strength from point-load testing, comminution parameters, and the density and geochemical proxies that feed a resource model long before mining starts. If you’re deciding where to start applying AI to an exploration program, this is usually the highest-leverage place to look first, both because the data already exists and because the vendors serving this space (Corescan, TESCAN, Zeiss, CSIRO) have already done the hard work of making their AI classifiers production-grade rather than research prototypes.