While we’re talking fraud, this NYT piece on people who sell YouTube views is fascinating and enlightening. YouTube counts how many people watch videos, puts the number next to them, and uses this to account for advertising money and decide which videos to promote. It’s therefore worthwhile to program a computer to click on your vid over and over again. Obviously YouTube tries to detect this. It’s part of the wider phenomenon of click-fraud, but it also has political consequences as you can use it to make YT promote your propaganda.
I love this detail:
Devumi’s customers included an employee of RT, a media organization funded by the Russian government, and an employee of Al Jazeera English, another state-backed company. Other buyers were a filmmaker working for Americans for Prosperity, a conservative political advocacy group, and the head of video at The New York Post.
There’s a kind of 1920s Chicago gangland flavour here. The Russians, the Kochs (via some relative of David Horowitz, or at least a Horowitz who is published on David Horowitz’s website, who espouses the same politics in the same style, and who is financed by the same patrons), and the Murdoch Post all use the same bootlegger. What was that about greed clarifying and cutting through again? And…Al Jazeera? I thought you were clean, bro.
Also, in 2013 the volume of fraudulent traffic nearly exceeded real traffic, which threatened to retrain YT’s machine-learning models so they would detect the good stuff as fraud. They pushed a fix, but I kind of wonder if that hasn’t actually happened and that’s why everything is completely crazy. This reminds me of an important point in Dan Davies’ book where he discusses the pervasive fraud in the US healthcare system in the 1990s.
Because claims for repayment were routinely denied, their denial was just an administrative process. Therefore, when a fraudulent claim was spotted, it was treated as an error rather than an attack. As a result, fraudsters would try lots of different options until they found one that was paid, and then hammer on that one until someone did something about it. In essence, the controls trained the fraudsters to get better – or in some sense, the fraudsters trained the controls.
But the heart of the matter is quite simply that there’s a view count, it has consequences, and so it is manipulated:
“The only way YouTube could eliminate this is if they removed the view counter altogether,” said Mr. Vassilev, the fake-view seller. “But that would defeat the purpose of YouTube.”
This is of course what I call algorithmic kitsch. It does seem to mesh with the emergent culture of shitty pol-entertainers described here. I do wonder if “metrics in everything” might not be much more worrying than “markets in everything”.
Compare, if you will, this essay on Wikipedia. It’s good, but if you want my take on it I would say that Wikipedia survives, and thrives, for two crucial reasons.
First, it is completely unashamed about expecting effort from its users. If you want to contribute, expect to have a job of work to do. If you just want to read it, you’ll need to…read it. You can’t just hit a button for instant outrage or Kardashians or whatever. Second, it doesn’t have a positive-feedback mechanism you can game like Twitter RTs or YT view counts, so it doesn’t drive that behaviour whether from bots or humans.