AI System Sorts News Articles by Whether or Not They Contain Actual Information

motherboard.vice.com

41 points·zeristor··7 comments

This is still in its infancy but you have to start somewhere. I wonder how long it will take to get to a system that can scan a whole series of articles on a related subject and be able to discern information which is more trustworthy from that which is suspect. Maybe it's not possible.

Good luck with that.

The issue here is that it reaches right back to the core ideas around epistemology, and those, whether we are culturally mature enough to accept it or not, get murky.

> Maybe it's not possible.

Well said.

This is probably the only possible answer to fake news. With human censors one can always cry “bias!” but a publicly published open source AI would be irrefutable.

It wouldn’t really be irrefutable, it just shifts the claims of bias to what was included as training data.

My guess would be that fake articles use different words and structure than real reporting based on facts.

It's entirely possible that an AI could pick up on even subtle differences.

If there's a strong incentive (financial or otherwise) for generating "fake news" content, then this looks like the makings of an arms-race to me.

The less willing the classifier is to issue false-positive "fake" ratings, the more wiggle-room the news-fakers have to play.

sounds like a job for [adversarial networks](https://arxiv.org/pdf/1406.2661.pdf)!

Reply on news.ycombinator.com