AI materials discovery is the use of AI models to predict which mixtures of ingredients will have the properties wanted, with robotic labs that test the best mixtures round the clock. Discovering a new material has always taken years of trial and error in a lab. Chemists made a mixture, tested it, and changed it, one experiment at a time.
The trial and error is slow because the number of possible mixtures is enormous. Even a battery liquid made from a few dozen ingredients, in different proportions, has more combinations than a lab can test in a lifetime. A chemist uses experience to pick which to try. AI is meant to pick better, and faster.
The new method has three steps. First, a model is trained on data about known materials and their properties. This is similar in spirit to the way foundation models are trained on text, although the models for materials are built for a different job. Second, the model predicts the properties of mixtures nobody has made, and ranks them. Third, a robotic lab makes and tests the top mixtures automatically. The results go back to the model, which improves. Such a loop is also called a self-driving lab.
Imagine a lab at a battery-cell maker that is looking for a better liquid for its cells. The liquid must conduct well, stay safe when hot and last through thousands of charges. The model ranks thousands of possible mixtures and selects the 100 most promising. The robotic lab tests all 100 in a day, while a human chemist might test a handful. The 10 best results are sent back, and the model suggests the next 100. After some weeks, the lab has a short list of mixtures to make in larger amounts.
A short list is not a product. A mixture that works in a few grams must still be made in tonnes, at a low cost, by a plant that does not exist yet. This scale-up often takes longer than the discovery. Models can also be wrong where they have little data, so every prediction is confirmed in the lab. Who owns a patent for a material found by a model is also a question that companies are working out. The field is young, so claims should be read carefully.
Companies are using this method first for batteries, catalysts and chips. Few materials found this way have reached production yet. A new battery cell material is one example. The results feed speciality chemicals makers, who need new formulas, and advanced materials makers, who need new compositions. Chemistry research tools of the same kind are described in AI for chemistry and materials R&D. In India, the buyers are battery and chemical companies and research institutes, and several start-ups and university groups are building such models. The wider area is materials science.