It used to be easy to spot these accounts, but Face Generating AI developed by NVIDIA may make it impossible to separate the real from the fake.
I assume most people here are aware the amount of fake accounts used to astroturf across social media. Bots, shills, and sockpuppet accounts are being used daily to create the illusion of popular opinon. For those that are unaware, heres a great piece from a few years ago about how Sally Albright accidentally revealed her huge network of fake accounts.
It used to be easy to figure out which of these accounts were fake. A big giveaway was the lack of profile pictures. Facebook shills were the easiest to spot, a non human profile picture, and when you clicked on the account there were no actual photos, and a timeline made up of shared political articles from the same handful of sources, which have little to no likes or engagement. As technology advances, it becomes increasingly more difficult to spot these accounts.
In early 2019 NVIDIA developed a hyper realistic face generating AI capable of creating images of fake people (in any pose) so realistic that its almost impossible to prove if they actually exist. These pictures are so real that you would look crazy to question them online.
As of now there is no substantial evidence that any of these shill/troll farms are actually using this in their armies of fake accounts, but there are enough coincidences to make a strong case that this is in fact happening. First lets start off by asking why wouldn’t these people use this tool? It helps add a level of authenticity to these accounts that makes it harder to prove they’re fake. People can no longer call out accounts whose profile pictures are a dog, tree, meme, or my favorite…the super edgy anti-Trump “45 made to look like a swastika” meme. Next lets ask what happened to all of the avatar/non human profile pictures? They’ve seemed to have disappeared. While most people will suggest this means social media companies have cracked down on these accounts this is not the case. They’re still there, just with a new flashy human face to show off. Next time you see an account that you suspect to be an astroturfing shill, click on it. I can tell you 9 times out of 10 what you’ll see is 1 (maybe 2) profile pictures of a human face, followed by a bunch of avatar/meme based profile pictures dating back to 2015. Along with that you’ll see the same pattern of a timeline made up of political posts shared from the same 4 sources with little to no likes.
It is also important to note that this activity tends to spike in election seasons. Politicians dump money into these networks to try to make their candidates appear more popular. In 2016 we had Correct The Record, and Share Blue. Today we have “Defeat Disinfo” which is AI previously used to track terrorists on social media, but now repurposed to combat Covid Conspiracies/target pro Trump accounts. With the addition of face generating technology, its getting more and more difficult to tell whats real anymore.
If anyone has any more info to add, especially some sort of evidence of this technology being used by troll farms, please do.