I don’t disagree that all these buckets are generalizations, with plenty of exceptions to the rule.
But, that’s why stats is a discipline that uses math, but is different than math. We don’t ignore those exceptions, we determine logical rules by which to normalize them. We figure out how to manage outliers.
The goal is to learn something about populations in the process. And although no single person in a sample might be 100% representative of the model you create…….that doesn’t mean the model can’t teach you something.