Comment by explorer83

My spouse is a polymer chemist for a large company. They spend tons of time trying things that don't make it to final products. Even if they produce a product which has the desired properties in a stable creation and application environment, it often fails in variations of the original creation or application environment. Essentially chemistry machine learning requires a complex physical feedback system feeding it tons of data to work. The large amounts of intricate physical data compiled in uniform data layouts required to feed machine learning models probably doesn't exist.