Fake News Challenge

Classification is still very needed. I'm interested to see what comes out of this. Especially with regard to eliminating biases of a classification system.

Capitalism isn't going anywhere obviously and shouldn't. On the other end is state-run media, e.g. North Korea. We wouldn't want things to swing in that direction or even toward RT or many other examples.

I think we should be looking to replicate whatever magic is and has been in PBS, early BBC, ITN, and other earlier UK and French programming.

Yep very fair, but it would have to be assumed the AI could differentiate, which is not necessarily trivial.

All of this would be much easier if there was some "standard" for writing news, sectioning off opinions that derive 'non-obvious' conclusions, linking primary sources, etc. But there isn't, and you would have to censor those who did not conform if there was one, which is not a good solution.

OK, fake or real?

(background: Russia might have "kompromat" on President Trump...blah blah )

Headline: President Trump appoints "friend of Russia award" winner as Secretary of State.

Just put a "Alert" or "Fake news" button. A lot of political writing can be considered "fake news" depending on our biases.

Your arguments are strong, however have you considered an algorithm based on the laziness and cheapness of humans? Consider that a lot of propaganda is heavily coordinated and not creative in the sense of being centrally created and issued.

An algorithm could watch, say, 50 sources and see if they simultaneously start using identical phrases, given access to a thesaurus aren't those odds a little unusual? This happened a couple times in the last election. You'd have a much simpler task of quote analysis of course, to exclude directly quoted content.

Then factor in something to detect press releases and video news releases where the verbiage of large sections is identical, this strategy is almost too easy.

Another factor is "large known propaganda sources" simultaneously releasing the story but small propaganda sources don't operate at all because fundamentally there is a size cutoff. The Wash Post will be in every propaganda operation because they are huge, but the "nowheresville register" certainly will not. Real news would of course be covered by both, so it should be very easy to detect organic self organizing behavior vs small scale conspiratorial behavior.

Now algo-proof propaganda could emerge much like handwritten letters to your congressional representative, before being inevitably ignored, are considered more impressive than mere form letters or signing a list. But it would detect at least some fake news based on fake news being part of a coordinated centralized mass produced campaign rather than trying to detect fake news based on its inherent fake-ness itself. Sorta like swatting bugs that bite you or try to, rather than identifying each individual bug then swatting only the mosquitos. Its very difficult to detect if something is a Walmart product, but its easy to detect if its mass produced, if the quality is very low, if it was made in China, if it contains chemicals banned in the USA for safety reasons, and in the long run works just as well.

I don't think it has to do with language (unless by "language" you mean what I mean, in which case sorry for duplicating). I think it has to do with that a lot of what people call "truth" is a product of their opinions, feelings, hopes, desires, etc.

There are facts of course - like such and such event happened or not, such and such statistical data is showing this number of not. But it is never served raw, it is always garnished with opinion. E.g. the fact is that Trump signed certain executive order. But you can say it's "Muslim ban" or "temporary restriction on admittance of nationals from terror-ridden countries". Not the same.

And various "factchecking" sites are fully complicit in this and inject opinion as often as not - to the point exactly the same thing said by two different people can be deemed "mostly true" or "mostly false", moreover, exactly the same thing said by the same person can be "mostly true" and "false".

You can't hope to have robots make sense of it if we can't. You can, of course, build a neural network and train it, but it would just put "robot approved" stamp on biases of whoever trained it.