Regardless of whether you think Timnit Gebru is a jerk or a very fine person, bias in real-world AI is a big problem.

To the extent algorithms are used to make significant, real-world decisions (whom to arrest, whom to extend credit, whom to grant parole, etc.) they had better be fair. And it seems that black-box algorithmic decisions are becoming more and more prevalent in the real world.

It's been clearly demonstrated that biased training sets give biased results. It's also hard to come up with un-biased training sets. This is a problem that's worth working on.