Even small babies are aware of the basic physics of everyday objects. Credit: Getty
Inspired by research on how babies learn, computer scientists have created a program that can learn simple physical rules about the behavior of objects and express surprise when they seem to break those rules. The results were published on July 11th Human behavior of nature1.
Developmental psychologists assess how babies track the movement of objects by tracking their gaze. When a video is shown, for example, of a ball disappearing suddenly, the children express surprise, which the researchers measure by how long they look in a certain direction.
Luis Piloto, a computer scientist at Google-owned DeepMind in London, and his collaborators wanted to develop a similar test for artificial intelligence (AI). The team trained a neural network, a type of software system that learns by detecting patterns in large amounts of data, with animated videos of simple objects such as cubes and balls.
The software model, called Physics Learning through Auto-encoding and Tracking Objects (PLATO), was fed with the raw images from the videos, but also with versions that highlighted each object in the scene. PLATO was designed to develop an internal representation of the physical properties of objects, such as their positions and velocities.
The system trained with about 30 hours of videos showing simple mechanisms such as a ball rolling down a slope or two balls bouncing off each other, and developed the ability to predict how these objects would behave in different situations. In particular, he 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 shape of objects. “Every step of a movie, he makes a prediction” about what will happen next, Piloto says. “As the film progresses, the prediction becomes more accurate.”
Surprise!
When showing videos with “impossible” events, such as an object that suddenly disappears, PLATO could measure the difference between the video and its own prediction, providing a measure of surprise.
Piloto says PLATO is not designed as a model of child behavior, but could be a first step toward AI that can test hypotheses about how human babies learn. “Hopefully this can eventually be used by cognitive scientists to seriously model babies’ behavior.”
Comparing AI with the way human babies learn is “an important direction of research,” says Jeff Clune, a computer scientist at the University of British Columbia in Vancouver. “That said, the paper hand-draws much of the prior knowledge that gives these AI models an advantage.”
Clune and other researchers are working on approaches in which the program develops its own algorithms for understanding the physical world.