Great read. Even in STEM research as a grad student I’m very tired of every saying “let’s try machine learning on this problem” to get something that works marginally better than some conventional models but requiring huge amounts of computation and data.