Fake news has nothing to do with “sounding journalistic”. This is how one ends up confusing propaganda (fake news) with truth, which is at the heart of a lot of journalism today.
The only thing such software will accomplish is further reinforce and convince the gullible that if someone sounds like a journalist then what they have to say must be true.
It’s akin to other (terrible) ideas where people think they can solve (really hard) societal problems with (insert hot technical solution), e.g., using algorithms to “certify” that certain algorithms are fair according to either naive or just plain wrong understanding of what fairness is (like, assigning money bail amounts based on an algorithm that is “fair” according to some specification: for example, that it’ll produce the same result whether or not race is an input, ignoring: A. the entire money bail system is terrible, doesn’t work, and shouldn’t be given further credibility and B. algorithms can easily recreate bias working around whatever minor corrections are attempted)
I wonder how this would do on mainstream, professionally-written news articles from 2002 talking about the Bush administration’s view that Iraq had WMDs. Or on articles from 1894 alleging that Dreyfuss was surely a contemptable spy for the Germans. Would it categorize “J’accuse” as fake news?
News, by definition, is about new and recent things. If I say, “There’s a new forest fire near Napa right now,” there’s no way for a computer to fact check that using training data. At best it can correlate with other sources, but that depends on finding trustworthy sources in the first place, which obviates the need for the detector.
That’s a better option, honestly. Having an AI correlate multiple data sources together to find the truth between them would be just as easy as training a machine learning algorithm to see how “newsy” the information sounds.
I think it’d be valuable to have an automatic correlation between news sources where the site red-lines ideas that one journalistic outfit mentions but isn’t backed up by the others. That’d root out fake news pretty quick (or at the very least find the consensus viewpoint).
If you found this article interesting I suggest checking out http://www.fakenewschallenge.org, There they thought long and hard what a fake news dataset should look like and some models trained against it.
Fake news has nothing to do with “sounding journalistic”. This is how one ends up confusing propaganda (fake news) with truth, which is at the heart of a lot of journalism today.
The only thing such software will accomplish is further reinforce and convince the gullible that if someone sounds like a journalist then what they have to say must be true.
Yeah, this is an incredibly terrible idea.
It’s akin to other (terrible) ideas where people think they can solve (really hard) societal problems with (insert hot technical solution), e.g., using algorithms to “certify” that certain algorithms are fair according to either naive or just plain wrong understanding of what fairness is (like, assigning money bail amounts based on an algorithm that is “fair” according to some specification: for example, that it’ll produce the same result whether or not race is an input, ignoring: A. the entire money bail system is terrible, doesn’t work, and shouldn’t be given further credibility and B. algorithms can easily recreate bias working around whatever minor corrections are attempted)
I wonder how this would do on mainstream, professionally-written news articles from 2002 talking about the Bush administration’s view that Iraq had WMDs. Or on articles from 1894 alleging that Dreyfuss was surely a contemptable spy for the Germans. Would it categorize “J’accuse” as fake news?
This seems misguided and probably impossible.
News, by definition, is about new and recent things. If I say, “There’s a new forest fire near Napa right now,” there’s no way for a computer to fact check that using training data. At best it can correlate with other sources, but that depends on finding trustworthy sources in the first place, which obviates the need for the detector.
That’s a better option, honestly. Having an AI correlate multiple data sources together to find the truth between them would be just as easy as training a machine learning algorithm to see how “newsy” the information sounds.
I think it’d be valuable to have an automatic correlation between news sources where the site red-lines ideas that one journalistic outfit mentions but isn’t backed up by the others. That’d root out fake news pretty quick (or at the very least find the consensus viewpoint).
If you found this article interesting I suggest checking out http://www.fakenewschallenge.org, There they thought long and hard what a fake news dataset should look like and some models trained against it.
This article promotes proprietary software, it does not really say anything else.