Faces of People Who Don't Exist, and Other Nice Tries

You probably read, as I did, the New York Times article about the website created by graphic chip company Nvidia to showcase how Generative Adversarial Networks (GAN) can create fairly convincing images of faces of people who don't exist. The technology is still rife with problems, as you can see in some of the artifact-strewn images -- particularly noticeable in how the algorithm botches human hairstyles with inexplicable blurs. It's a noble effort, but it still has a long way to go.

The race is on.

Just like video technologists pioneered a way to simulate people on "deep fake" video saying things they never said, we have photos of people who never existed, so how do you verify authenticity? More philosophically, does authenticity matter? Clearly, it still does, because one way scientists are trying to stay ahead of fake people photos is by detecting embedded geographic positioning system (GPS) metadata to tie a photograph to a physical location. But GPS metadata can be faked, too.

Like the Security and Exchange Commission's chase to monitor and police the financial industry's endless development of newfangled financial shenanigans, journalists and scientists are going to need to increasingly work side by side to determine the veracity of every electronic communication. Fact checking, as a practice, is well and good, but source checking may soon become more critical, unless, as Jordan Peele says, we want to become "some kind of fucked up dystopia."

Note: the Nvidia hyperlink is to an article in The New York Times about their role in monitoring and processing terabytes of facial recognition data collected by the People's Liberation Army in Urumqi as part of their massive "re-education campaign" of the Uighur population.