Machine learning is reshaping how researchers process data and model complex physical systems. Its intersection with the sciences offers useful perspective for automation teams assessing data-driven methods, including the importance of domain knowledge, suitable models, and application-specific challenges.