Re: Why use the word racist.
"Light is obviously not racist, you bleeding fool."
Yes, it's called reductio ad absurdam.
Of course light isn't racist. But light has differing effects with regards white and black people, and in particular makes it much harder to make out black faces in dark conditions. If humans (who are very good at pattern recognition) have more trouble in low-light conditions with black faces than white ones, why it is clear that a computer should be able to treat both equally well in higher-light conditions?
The point is that racism is meant to be where decisions taken by people have a deleterious consequence for one or more racial group. After many, many examples of algorithms having difficulty with black faces, perhaps it's time to start thinking that, maybe, algorithms have a harder time distinguishing black faces because it is an objectively harder problem?
If that is the case, then suddenly it looks unseemly to throw words like 'racist' around to describe the perhaps completely blameless programmer (because, as you have stated, non-people cannot be racist, so if the algorithm is racist that means the people who programmed it must be).
Repeating the 'these computer programmers are racist' mantra/libel is starting to become tedious, when the evidence is stacking up that it is just more difficult. Of course, evidence is not a proof, but it should be fairly easy to check. Use a million-strong dataset of white faces and a million-strong dataset of black faces. Then evaluate the performance of the algorithm on a new selection. I've not heard of such an experiment being conducted, but it would go a long way to settling this argument.
And, here's the interesting question: what happens if it turns out that these algorithms will always be worse at distinguishing black faces? You know, like humans are.