Artificial intelligence inspired by research on babies’ learning has been developed by Luis Piloto and his collaborators at Google-owned DeepMind in London. The new AI is dubbed PLATO, which stands for Physics Learning through Auto-encoding and Tracking Objects.
Developmental psychologists have discovered a way to test how babies understand the motion of objects. By showing babies a video of a bouncing ball that unexpectedly disappears, they can track their gaze. Their degree of surprise can be gauged by timing how long the infants stare in a particular direction.
Similarly, the team at DeepMind trained a neural network with animated videos of simple objects like cubes and balls. The system was fed about 30 hours of videos depicting simple actions like two balls bouncing off each other or a ball rolling down a slope.
“Luckily for us, developmental psychologists have spent decades studying what infants know about the physical world and cataloging the different ingredients or concepts that go into physical understanding. Extending their work, we built and open-sourced a physical concepts data set. This synthetic video data set takes inspiration from the original developmental experiments to assess physical concepts in our models,”
Piloto said.
Solidity, Continuity, And Persistence
PLATO was able to gain the ability to predict how objects would behave in different circumstances.
“In particular, it learned patterns such as continuity, in which an object follows an uninterrupted trajectory rather than magically teleporting from one place to another; solidity, which prevents two objects from penetrating each other; and persistence of the objects’ shape. At every movie step, it predicts” what will happen next,”
Piloto said.
According to Piloto, the AI was not intended to model the behavior of an infant, but it may serve as the first step to AI for cognitive scientists that can test various hypotheses about how human babies learn.
Reference: Hespos, S., Shivaram, A. Can a computer think like a baby?. Nat Hum Behav 11 July 2022
