Assessing the application of artificial intelligence to the discovery of new mineral species

Artificial intelligence is having a revolutionary effect on diverse areas of research, such as proteins, drugs, and materials (in July of 2025, Google Scholar listed around eighty publications with “artificial intelligence” and “materials” in the title and dated from the current year), including prediction of new entities.

The discovery of new mineral species constitutes a more demanding challenge as these predicted new mineral species must have natural occurrences resulting from geological processes. There are some initial results, however, that are not especially impressive, as we will discuss.
We assess diverse instances of available generative artificial intelligence tools (Aria, ChatGPT, Claude, Copilot, Gemini, Grok, M365 Copilot, Meta AI, Perplexity, and YouChat) in relation to their usefulness in predicting new, undiscovered mineral species along the following main lines: the current state of the art in relation to confirmed predictions, and proposed methodologies of artificial intelligence (including potential limitations) for this goal. Special attention will be given to the issue of natural occurrence. Accordingly, we promote an evaluation of artificial intelligence potential by artificial intelligence tools.

Results are widely variable, with some generic answers, some problems with references, and some promising suggestions regarding the conditions under which the new mineral species could be found.