A new study is inspired by the theory of developmental psychology to teach an artificial intelligence some fundamentals of physics.
Shared in the journal Nature, the study was conducted by a team of researchers from the company DeepMind. They created a program capable of learning simple physical rules about the behavior of objects.
Learn “common sense”
While AI has made astonishing progress in recent years, the most advanced systems still struggle with “common sense”. Common sense that guides prediction, inference and action in everyday scenarios. The DeepMind scientists admit, however, that this intuitive physics is “fundamental to embodied intelligence”, because “it is essential to all practical interaction”, they write.
To carry out its test, the research team led by Luis Piloto was inspired by the psychology of infant development. Developmental psychologists study how babies follow the movement of objects by following their gaze. When shown a video of a ball suddenly disappearing, for example, young children express surprise. Researchers thus measure their surprise by the duration of their gaze in a given direction.
This is precisely what the team of scientists at DeepMind set out to find out, by training an AI dubbed “PLATO” with videos of simple objects, like rolling and colliding balls.
Five Key Physical Concepts
To do this, they focused their attention on five physical concepts considered “central” in the literature on developmental psychology: continuity, persistence of objects, solidity, immutability and directional inertia. “Each video with a violation of these physics principals is paired with a corresponding video that provides a baseline that conforms to physics principles, maintaining precisely the same single frame statistics for all videos,” the researchers explain.
The dataset also includes a separate corpus of videos intended to serve as training data.
Using this process, the researchers wanted to build a model capable of learning what they call “intuitive physics” like an infant discovering the world. The method used to measure knowledge of a physical concept is called the “expectation violation paradigm,” the researchers say.
28 hours of videos
The researchers observed learning effects after only 28 hours of videos. A definite progress, even if they recognize that “the range of types of objects and events in our dataset remains narrow compared to those encountered in the real world”.
By ripple effect, the researchers also wondered about the potential implications of their work for developmental psychology itself. “This topic should be approached with some caution, as the model we presented is not intended to provide a direct model of the acquisition of physical concepts in children. Nevertheless, we believe that there are several ideas that can be useful to development science. First, our modeling work provides proof of concept demonstrating that at least some central concepts of intuitive physics can be learned through visual learning.