Thread by @ScienceVet2

You cannot assume your science is free from bias simply because it is science and you didn't put bias in on purpose. Turns out self-driving cars can't recognize people with darker skin as well.

Most likely reason? Designers probably used themselves.

arxiv.org/pdf/1902.11097…

This is a problem common in everything from robotics to psych. It's way easier to use your team for the seeds of a run of machine learning or as the models for your image recognition. But if your team doesn't represent everyone out there... your robot will be biased

An article that isn't the actual paper showing the bias, for those who don't want to drag themselves through it. vox.com/future-perfect…

We've learned some interesting things about hiring bias this way, too, actually. It's really hard to train a computer to sort resumes because getting uncorrupted training data is next to impossible.

reuters.com/article/us-ama…

And it's not just in new tech, there was a similar problem with color film. It started to change because of chocolate and wood grain, and a better color film wasn't released until '95

petapixel.com/2015/09/19/her…