At the source, you do need some form of verification, similar to "trusted certificates." For example, there could be a database of known mappings, such that apple.com and subdomains are authorized to make statements cited as press releases about Apple, Inc. A link would have to be intelligently interpreted as a citation, perhaps involving a search for similar stories (or it just results in 'unable to find link to primary source' even when it's legitimate, i.e. false positives). Subjectiveness is another area that's quite difficult, but would likely result in percentage of 'unique claims' that the AI was not able to derive from the source material.
It's more difficult than just blindly blocking any stories that aren't covered on 'major, trusted sources', and probably only a couple of organizations could implement anything close to this right now, but it's the best we can strive for. All the other 'solutions' I've heard involve censorship of unpopular opinions.
Edit: But yes, to the point, my envisioned implementation would not find any referenced sources in your article or through a search, and if it did, those sources would not be authorized to make the claims about the Whitehouse reptiles, so it would get the notation that indicates 'could not verify.'
At first, most real news stories would be unable to be verified, but over the time, there would be fewer false negatives. The important thing to recognize, though, is that the notation just means the reader has to make his own judgment; it does not state whether or not it's fake.
And "real stories" should also probably have a notation with a direct link to what the algorithm thinks is the base source.
Have to admit, something like this could be useful as a simple way to improve journalism in general. Yeah, it wouldn't exactly 'detect fake news', but it would help by:
1. Encouraging journalists to actually link to a source at all. Because at the moment, a depressing number of them don't even bother sourcing anything. So you end up just having to 'trust' the organisation that the story has been checked out beforehand.
It's an especially big problem with web versions of printed newspaper articles, and is a general problem with print newspapers as a whole (since due to space limits, they don't publically source/cite anything).
2. Getting journalists to link to the original source. Because at the moment, you'll see news sites sourcing other news sites which then source other news like a lengthy game of Chinese Whispers or Telephone. It's how a lot of 'fake news' makes its way in the media food chain.
So yeah, something like this would be useful just to get publications to cite their sources better. Or to make them look for the real source rather than one of a million middlemen.
Just encountered an article from AP, which purports to "fact check" some Trump statements. The whole article did not contain even a single link. Not even to the statements they are checking. I went back to look at another "fact checks" from AP - yep, not a single link to the source at any of them. Sigh. And these people probably think Wikipedia is unreliable and poorly sourced.
It's more difficult than just blindly blocking any stories that aren't covered on 'major, trusted sources', and probably only a couple of organizations could implement anything close to this right now, but it's the best we can strive for. All the other 'solutions' I've heard involve censorship of unpopular opinions.
Edit: But yes, to the point, my envisioned implementation would not find any referenced sources in your article or through a search, and if it did, those sources would not be authorized to make the claims about the Whitehouse reptiles, so it would get the notation that indicates 'could not verify.'
At first, most real news stories would be unable to be verified, but over the time, there would be fewer false negatives. The important thing to recognize, though, is that the notation just means the reader has to make his own judgment; it does not state whether or not it's fake.
And "real stories" should also probably have a notation with a direct link to what the algorithm thinks is the base source.