Perhaps you mean that, in characteristics where humans are known to vary, one should suspend judgment / assume the default probability distribution, rather than assuming the person is known to be average?
Yes. I put notions like “humans are generally vulnerable to Death by Hemlock” in a different class than notions like “Girls don’t like science”. For one thing, the stakes are a lot higher in the former case: you don’t harm a female by not assuming she doesn’t like science, but you might kill a human by feeding them hemlock under the assumption that you “need more data”. There’s plenty of empirical data on the effects of hemlock poisoning in entities you’d likely classify as “human” (for the purpose of this exercise), after all, and it seems pretty clear that hemlock ingestion is much more hazardous than not being subjected to the assumption that you hate science because you have a uterus.
Again, I think it’s import to see the kind of categorization you dislike as ‘inept categorization’, including attempts to infer from the category things that have already been observed and hence ought properly to be screened off; rather than ‘forbidden categorization’.
No argument from me there.