How Which Face Is Real uses AI
Which Face Is Real was launched in February 2019 by Jevin West of the University of Washington’s Information School and biologist Carl Bergstrom as part of their Calling Bullshit project, which teaches data reasoning skills for a digital world. The game is as simple as its name: two portraits appear side by side, and you click the one you think is a real photograph.
Every fake comes from StyleGAN, the face generator developed by NVIDIA researchers, by way of the site This Person Does Not Exist, and every real face comes from the FFHQ dataset of Creative Commons and public-domain photos. A generative adversarial network works like a game of cat and mouse: one network creates faces while another tries to catch the fakes, and each forces the other to improve.
The creators built the game to show how convincing synthetic faces had become and how easily they could be used for fake profiles and false identities. With practice, players learn to look for the small glitches that give generated faces away. It is free, takes seconds per round and needs no account.



