The speaker explains that in order to improve the training set of molecules for machine learning, it is essential to collect huge amounts of data about the actual gases and to convince someone in a lab to measure them at outrageous temperatures. The ultimate goal is to do the full end-to-end process of gas collection, analysis, and applying machine learning to the data that comes from other planets.