>We are working to fix the issue that allowed this to happen This is not some issue that you can fix. This is not some color being off, or some chat message appearing out of order. This is about AI in its continued infancy that’s now being actively and maliciously targeted. It’s also about a huge responsibility of the worlds two greatest media outlets that both think they can have machines do a humans job and miserab…
The real issue is that any system which attempts to organize or curate untrusted information is subject to adversarial attacks.
AI up until this point, has been mostly trained on labeled, human curated data sets. But as it starts to get trained on real user input, it'll be subject to attacks by people trying to game it. AI needs to address this robustly, but it's in it's infancy as you say.
I mean, we're still working on bias issues like machines learning that "doctor" is "male" because you trained on a corpus of English text, and therefore, most of the input text referenced a doctor as male, and that's the AI getting poisoned by cultural biases, which wasn't even an organized attack.