There seems to be a fundamental disconnect between the author's understanding of bias in AI and the actuality of it. Google's "Gorilla" incident was a result o

There seems to be a fundamental disconnect between the author's understanding of bias in AI and the actuality of it. Google's "Gorilla" incident was a result of bad training data. Recidivism scoring is the result of multiple unfortunate truths that the data concerning crime will lead you to, no racism or bias required. Nikon cameras detecting Asian faces as blinking happens for obvious reasons; claiming that there's racism or bias is especially absurd in this case. Microsoft's chat bot is the result of intentionally targeted training data to get a certain (unsavory) result. Google's language algorithm reflects the bias of language and its usage, not the algorithm. I will be glad to accept any example of a bias (and not the underlying data) presenting itself in an algorithm. Until then, the underlying argument behind these diversity pushes needs work to be truly compelling.