Gender is - or was - convenient. With binary genders you have a convenient model to categorize people into different groups, in the same way that you can categorize by age, skin color, language, homeland, or religion. Politically, categories are mostly used to drive up the "us vs. them" tension factor that unifies groups, and pushing for harder-edged definitions of gender and gender roles is a convenient way to politicize just about everything in a human life. Further, it naturally works out that your voting bloc floats at around 50% when you appeal to the gender binary, which helps sustain the division in a way that you can't get with things like "race" or "religion".

Of course, one of the truisms of data modelling is that it never captures 100% of the system, and it's guided by your expectations of what the system will look like. Our push to digitally model everything exaggerates all the points of breakage. In the same way that you end up with people whose legal name are flagged fake by Facebook, or reverse image searches that think black people are "gorillas", you end up with a lot of data points that are bad fits for a model of binary gender.

This is not a huge surprise when humans are equated with other mammals in all their variety. But we have the problem of being emotional, well-socialized, and biased first, and smart second. So we're probably just going to have an extremely long culture war to rethink this categorization, with specious arguments trotted out at every turn.