That's not exactly a surprising result. Didn't Microsoft have to shut down their Twitter-trained AI after it started emulating a typical terminally-online neo-Nazi? I might be a CS dropout, but I stuck around long enough for garbage in, garbage out.
Tay was basically a publicity stunt. It had a poorly documented feature where you could get it to directly repeat things you messaged it, I think via PM, making it seem far more advanced than it actually was. Which obviously lead to trolls spamming it with shit like "1488 RAHOWA GAS THE KIKES", to name one thing at least.
Definitely not a surprising result. Though I don't think the article is claiming it is a surprise.
Right, what did they expect? Lets feed it tons of data from a questionable source and we'll get a good outcome?
True, but unlike Tay, in this case it was a feature, not a bug.