I don't know whether these automated news aggregators rely on machine learning per se, but this kind of reminds me of recent ML fails like Microsoft's Tay bot t

I don't know whether these automated news aggregators rely on machine learning per se, but this kind of reminds me of recent ML fails like Microsoft's Tay bot turning into a Nazi, or Google Photos labeling black people as gorillas. Andrew Ng interviewed Ian Goodfellow in the first course of his deep learning specialization on Coursera. Ian said that he's really interested in machine learning security, in part guarding against untrusted or adversarial inputs. Whether they're poisoning your network in the first place, or your network is misclassifying them because they're too weird, it seems like you need to treat random inputs as potentially adversarial.