More generally, if one wants to learn about the distribution of opinions in a group X, one needs to make some kind of a poll, instead of listening to the self-proclaimed speakers for the group.

Otherwise the result may be more strongly influenced by "what makes people become public speakers for a group" than by merely "belonging to the group X".

Maybe we should always remind ourselves about the forces of self-selection. Looking at a Mensa member, instead of just "a highly intelligent person" we should also think "a person who prefers to publicly associate with groups defined by innate traits (as opposed to behavior or achievements)". Looking at a professional feminist, instead of "a woman", we should also think "a person who built their career on hating men". Looking at a men's rights activist, instead of "a man", we should also think "a person who got burned by a divorce". Etc.

It is also important to notice how much easier is this to do for the groups one doesn't like (where it feels like an obvious step that doesn't even require an explanation), than for the groups one does like (where it feels like an unfair generalization).

But this reminder itself is not sufficient to find out the opinions of the silent majority. (Reverting stupidity is not intelligence.) Recognizing that we have noisy data doesn't automatically un-noise them. Unfortunately, even the public online poll would suffer from "people who prefer to express their opinions in online polls" selection bias.