Learning like a baby during the first months of life is the type of artificial intelligence () than developmental psychologists from have developed. They were inspired by the way babies follow le items. When shown a disappearing ball, babies express surprise. The researchers then measure the of that amazement when they look in a particular direction. As with an infant, the neural network was enriched with models including around 30 videos of moving objects, such as rolling or bouncing balls and cubes.
A surprise AI!
pour Physics Learning through Auto-encoding and Tracking Objectswas designed so that the AI could appropriate the dimensions of objects, their properties and their position and in the space. The AI eventually learned to predict how these objects would behave in different situations. Simple teachings, on the solidity which prevents two hard objects from penetrating each other, or the trajectories and their continuity. By dint of teaching, the predictions became more and more precise. And when an event was considered impossible, for example, in the case where an object suddenly disappears, Plato then showed his reaction of displaying the difference between the video and its prediction. The idea of this project is not so much to create an artificial intelligence on the model of human learning, but precisely to model the learning of infants.