I wonder if there is a simple way to explain how ML training works to non-techy folk, so that they realize it's mostly just tweaking equations until they reproduce the corpus they are fed, and they're going to do their best to reproduce that corpus, biases and defects included?