For centuries, the development of glass relied on costly, slow trial-and-error experiments guided by chemical intuition and accumulated industrial know-how. This paradigm is now changing because modern high-throughput computation
and machine learning (ML) algorithms can uncover hidden potential in a short amount of time.

A stimulus for this transition was Zanotto and Coutinho’s 2004 paper, “How many non-crystalline solids can be made from all the elements of the periodic table?”¹ That work highlighted the astronomical number of possible glass compositions (more than 10⁵²) compared to the infinitesimal fraction already synthesized and experimentally characterized (about 10⁶). The implication was that conventional experimental approaches alone cannot efficiently explore the enormous “glass genome,” thus motivating the use of data-driven discovery strategies.