Searching for materials with the right properties
Materials scientists face a similar search problem to drug discovery: given a desired set of properties, like a battery material that stores more energy, charges faster, or degrades more slowly, or a material with useful superconducting behavior, which combination of elements and atomic structure will actually deliver it. The space of possible chemical compositions and crystal structures is vast, and historically much of materials discovery has relied on a combination of theory, intuition, and a great deal of trial-and-error lab work.
AI models trained on databases of known materials and their measured or simulated properties can predict how a new, untested composition or structure is likely to behave, before anyone has to synthesize it. This lets researchers computationally screen far more candidate materials than would be feasible to make and test by hand, focusing lab time on the compositions the model rates as most promising.