3 ms·
The interesting bit to me is that the guy didn’t make the cut. ML based methods are way behind humans at this point, but how far off is the day that we could tr
by boulos 9y ago
The interesting bit to me is that the guy didn’t make the cut. ML based methods are way behind humans at this point, but how far off is the day that we could train something to pass? 5 years? 3? Less? Especially for the sort of short-form comments on news articles, it seems possible to me to at least insert high-confidence “sounds American, but false” today-ish.
- avs733 9y agoSo this is an entirely out of my ass theory but here goes... The reason something like ML is failing here is because what the russian trolls are trying to do is fundamentally change how discourse occurs, how people in the US receive and process information, and the tone of the conversation. ML is great are reproducing things that already exist. That's why they get accused of being racist...because they are reproducing patterns in data. The russians were trying to cause things to happen by changing the conversation. That's fundamentally different and I haven't seen a lot of ML systems where one can dictate a goal like that and have the system work out the response. From what I know, that seems like one of the differences in the 'older' schools of AI and the newer schools that let the field actually have such an explosion of success nowadays...they got way more comfortable with backward vs. forward looking learning.