How Humans and AI See Faces in Objects

This piece, by Onno Berkan, was published on 10/08/24. The original text, by Rachel Gordon, was published by CSAIL on 09/30/24.

Pareidolia is the “phenomenon of seeing faces [...] when they aren’t really there.” This is common in humans, but less so in AI systems. In fact, a recent CSAIL study showed that AI could not recognize faces in objects until being trained on animal data.

Researchers think that this may be revealing towards why humans recognize faces where there are none– it’s better to see the Virgin Mary in your sandwich than to miss the tiger in the bushes. 

The researchers further curated a dataset of 5,000 images of faces and found that there is a Goldilocks zone wherein the images are just complex enough for both humans and AI to recognize faces. In this zone, face detection seemed to peak.

This dataset “dwarfs those of previous studies”, and researchers believe that it can be used in anything from intelligent design to improving human-computer interaction. One notable use case is in self-driving cars, which could use the ability to tell live faces from those who aren’t.

Want to submit a piece? Or trying to write a piece and struggling? Check out the guides here!

Thank you for reading. Reminder: Byte Sized is open to everyone! Feel free to submit your piece. Please read the guides first though.

All submissions to berkan@usc.edu with the header “Byte Sized Submission” in Word Doc format please. Thank you!