Practice your fake-spotting skills in this DIY drill. It’s from a university course named ‘Calling Bullshit’ at the University of Washington.

Each face-off pairs pictures from two databases: Real images from Flickr, fakes from, a demonstration site that shows images generated by an AI model.

To better distinguish real from fake, professors Jevin West and Carl Bergstrom suggest checking for any of several possible anomalies:

  • Water splotches
  • Background problems
  • Eyeglasses
  • Other asymmetries
  • Hair
  • Fluorescent bleed
  • Teeth

Their course is ‘how to think critically about the data and models that constitute evidence.’

[In the example picture, the computer-generated image is on the left]


Which Face Is Real?
Carl Bergstrom & Kevin West | University of Washington | 2019