The points she makes otherwise might indeed be worthy of discussion. However I think some of what she describes is simply misclassified data - black people get misclassified, but so do white people (but of course it doesn't make the news). Unless she can prove that white people are misclassified less frequently than black ones, she has no point.
The Tay incident was not due to bias but to trolls purposely feeding the IA racist information. Despite what she vaguely claim later in the article it wasn't encoded, not even subconsciously, in the bot by the developers.
I never saw that claim in the article. She did claim that white, straight men dominate fields of technology and robots, but that doesn't seem controversial to me (though I would happy to see evidence against it).
I'm not sure that either of your points diminish the argument of the article. I didn't get the impression that the author thought that developers are purposefully creating racist robots. To me, she was saying that those who suffer bigotry the least will also be the least likely to account for it in the systems they design because they see the world as less bigoted than it is and has been. Sure, in hindsight, the two examples you mentioned can be explained as poor sources of information. But if we're going to avoid bigoted tech & robots, we'll need to catch those issues beforehand, and I think her point is that more diversity would lead to better foresight on such things.