Related Wired and NY Times articles linked further down the Twitter thread: https://www.wired.com/story/best-algorithms-struggle-recogni... and https://www.nytimes.com/2020/06/24/technology/facial-recogni...

I thought he was being cheeky to be honest.

It’s more than a simple bug. It’s a design flaw that results in mistreatment for people of certain skin colors.

You are likely correct that the intent was not malicious. This is no excuse. Throughout history, everyone believes they are doing the right thing or are properly justified, even when heinous acts are being committed. We are master story tellers, and we tell the most twisted stories to ourselves.

At the end of the day, the impact that our actions have matters a lot more than our best intentions.

The cackling villain style of racism is mostly dying with Lee Atwater's generation. The human systems they installed perpetuate it with such subtlety that people can train an algorithm nearly 100% on white faces and not even realize it or think about why those are the most available faces to them. And when they realize it, it's so hard to fix.

This is why the academic definition of racism focuses on systems and the ways people unwittingly perpetuate it rather than trying to assign individual blame. That definition clashes with the definition most people think of where there's someone doing it on purpose.

Racism doesn't require malicious intent. It can arise from simple ignorance.

If the algorithm was modeled to include many different races I doubt this would be a problem.

So in your opinion nobody is racist here? This case does not demonstrate racial bias?

Assuming this is not a once off, the situation certainly implies that either the people who developed the algorithm or the people who made the decision to use it in their product either didn't test that it worked for darker skin colours or worse, they knew the faults but didn't care enough not to use it. That's racism to me.

You've not included any content in your argument, while the people you're calling stupid have. It's possible to argue about racism or lack thereof in a rational way, just like any other topic. It doesn't have to be hostile right off the bat every single time.

Your algorithm is trained on your racially disparate status quo and outputs racially disparate results. If your intent was race-neutral outputs, then you screwed up and the algorithm is implicitly racist. If your intent was outputs racially-biased towards fair-skinned people, then you're explicitly racist and your algorithm is still implicitly racist. Take your pick.

The alternative is not to do it. Companies that produce these algorithms have said not to deploy them against people.

You don't need to have something better in mind to criticize something. Criticism alone can be useful if the endeavor didn't need to be attempted at all.

We can do so much better than this

Racist algorithms, and faces that we miss