CAN REAL EYES RECOGNIZE AI LIES?
·Updated:
·

​Last week, we highlighted This Person Does Not Exist, which used a new Generative Adversarial Network (GAN) technique developed by Nvidia to produce disconcertingly realistic images of people. Now, two professors at the University of Washington have created Which Face Is Real, a web game where you try to to determine which of two images is real and which is generated by the neural net. 

If you take a moment to closely inspect each image, you'll be able to figure out a lot of the pairs, thanks to small imperfections, misplaced wrinkles, and Deep Dream-esque backgrounds in the AI-generated faces. And in some cases, when the AI really goofs, it's pretty simple — like this horror baby:

 

But it's concerning how often you come across a pair where choosing between the two is nearly impossible. Can you guess between these? 

 

 

 

From top to bottom, the right, right and left images are the real people. If you really focus on the guy on the right in the third image, there are some vague artifacts around his sideburn, but that's the only indication we get. The top two images have basically no indication of which is which, beyond guesswork. 

Go play around with the game for a bit. You'll leave it both impressed by the technological achievement behind the neural net and probably a bit distressed by what this portends for the future of "fake news."

[WhichFaceIsReal]

<p>Digg is what the internet is talking about, right now. It's also the website you are currently on.<br></p>

Want more stories like this?

Every day we send an email with the top stories from Digg.

Subscribe