Inspired by research into how infants learn, computer scientists have created a program that can learn simple physical rules about the behavior of objects and express surprise when they appear to break those rules. The results were published July 11 in Nature Human Behaviour. Developmental psychologists test how children follow the movement of objects by following their gaze. When shown a video of, for example, a ball suddenly disappearing, 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 colleagues wanted to develop a similar test for artificial intelligence (AI). The team trained a neural network – a type of software system that learns by recognizing patterns in large amounts of data – using animated videos of simple objects such as cubes and spheres.The software model, called Physics Learning through Auto-encoding and Tracking Objects (PLATO), was fed raw images from the videos, but also 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 was trained on 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, it learned patterns such as continuity, in which an object follows an uninterrupted trajectory rather than magically teleporting from one location to another; a strength that prevents two objects from penetrating each other; and constancy of shape of objects. “At every turn of the film, he’s predicting what’s going to happen next,” says Piloto. “As the film progresses, the prediction becomes more accurate.”Surprise!When shown videos of “impossible” events, such as the sudden disappearance of an object, PLATO could measure the difference between the video and its own prediction to provide a measure of surprise.
Piloto says that PLATO is not designed to model infant behavior, but that it could be a first step toward artificial intelligence that can test hypotheses about how human babies learn. “We hope that this can eventually be used by cognitive scientists to seriously model infant behavior.”Comparing artificial intelligence to how human infants learn is “an important line of research,” says Jeff Clune, a computer scientist at the University of British Columbia in Vancouver. “This means that the paper manually suggests most of the prior knowledge that gives these AI models their edge.”Clune and other researchers are working on approaches in which the program develops its own algorithms to understand the physical world.
Source Journal Reference: Davide Castelvecchi, DeepMind AI learns simple physics like a baby, nature news (2022). https://www.nature.com/articles/d41586-022-01921-7
