Algorithms, Bias, and Beautiful Women

I've been keeping my eyes open for things to discuss with students this fall, especially in Advanced Placement Computer Science Principles. This week bias in algorithms kept crossing my path. It was even a topic for the #EthicalCS Twitter chat this week. It's a real problem if we really want software to meet the needs of everyone. And really to keep people safe.

I've got a couple of good examples to share. One is attempts to scientifically determine what beautiful women look like. Personally I think that is a silly goal as beautiful is in the eye of the beholder but it sells magazines I guess. Take a look at this story. The 10 Most Beautiful Women in the World, According to Science. All of the women are white. Do we really believe that beauty is limited to white women or is there perhaps a bias involved? I would suggest the latter.

While that is sort of trivial in the scheme of things some biases in algorithms have a lot more risk. Take this story A white mask worked better': why algorithms are not colour blind about the discovery that some facial recognition doesn't recognize Black faces. More information at this TED Talk Joy Buolamwini - How I'm Fighting Bias in Algorithms. Imagine the possibilities. Police and other authorities use this sort of software and this suggests the possibilities for miss identification are frightening.

Take a look at this story as well Samsung adds and swiftly removes sexist Bixby descriptor tags Not so much an algorithm bias is a software inclusion of biased opinions. How did they miss that? I wonder how many women were in on that decision?

Biases are pretty much unavoidable. As one professor Tweeted me "Most biases are inherent/unavoidable part of cognition. See books by D. Kahneman, R. Thaler, or D. Arielly." If anything this agues for more diversity on software teams. Different biases may, one hopes, help to balance things out in algorithms and software in general. I think though that as educators it is the job of computer science teachers to discuss this issue with students. They need to be aware of the issue if they are to have any chance to moderate the effects.