There is literally no support in the article for the contention that "artificial intelligence picks up bias from human creators" as opposed to making correct inferences from reality. All of the examples they provide, modulo blurry tank pix, are of the latter.