Using these methods for scientific problems: predicting protein structure, searching for materials, modelling weather, sifting experimental data. It is the area with the clearest demonstrated value, and it works because the problems have abundant data and objective checks on whether an answer is right.
Technical view
Successes cluster where structure is well defined and verification is possible: structure prediction, materials candidate screening, and weather modelling where learned models now match or exceed numerical simulation at a fraction of the cost while requiring physical validation. The recurring pattern is generation of candidates followed by physical or computational verification, since the model proposes and the experiment disposes. Data scarcity and distribution shift limit transfer to less well-resourced fields.
Related
Part of AI Terminology Index, section: Modalities and Applications.