Curiously, a multitude of AI algorithms can figure which 'features' are relevant (ignoring training data requirement for a second*), to what extent, and how to

Curiously, a multitude of AI algorithms can figure which 'features' are relevant (ignoring training data requirement for a second*), to what extent, and how to use them as needed. On the other hand, humans (the authors in this example) often don't understand that, and can take any of the input features as the key determining factor. * When, why, and if, AI needs more training data than humans aside, even humans have received enough of it to have developed some common sense, but yet ...