So, Twitter's open-source algorithm contained a marker for Republican vs. Democrat accounts.

Elsewhere in the source code, there was a comment explaining that it didn't affect ranking, it was only meant as a metric to ensure their algorithm wasn't biasing to one side or the other. So, all good right?

Well... no, unfortunately, because it introduced false balance. It's no secret that Republicans are comparatively more likely to use outrage and hate as a publicity strategy. Not that Democrats are remotely innocent here, but comparatively it's less common.

So what happens when you try to adjust the algorithm to derank hate and outrage? The ranking for Republicans disproportionately drops, something that is not allowed by this metric. It's seen as a bug.

Because of this attempt at false balance, this attempt at "fairness for both sides", it would have become impossible to discourage posting hate, just because it would change the supposed fairness metric.

This is a great example of how technology isn't neutral, how building it "fairly for everybody" has very bad consequences, and why you always need to make a choice of what specific behaviour and ideology you want to encourage or discourage.