The images shown are randomly selected from two collections: a collection of real people and the other a collection of images created by an AI.

The AI images were generated using the current state of the art generative adversarial network StyleGAN (the code for which was recently released by researchers at Nvidia here). The real images are from the image set used by the research team to train the AI.

There are ~10,000 real images and ~10,000 fake images. Simply guessing 'Real' on every image should get you to ~50% accuracy.

tl;dr: Do your best to pick out the fake faces from the real ones.

Humanity is counting on you :-)

@tomhoag