Thread · 2 tweets · 28 Oct 2020

Right. I’m trying to think of a demo cluster that would get detected with some stability across models (like @coolharsh55 was pointing at) and would somehow get discriminated against structurally, and I’m finding it hard to imagine.

I don’t think it’s the case for all algos, but it’s certainly possible for all (or most) ML, I would think. It’s easily solved using the “Don’t Fuck With People’s Lives Using Tools You Can’t Understand” fairness/justice model.