From driverless vehicles to weapon systems, artificial intelligence (AI) models hold a huge responsibility these days, so you’d like to think that technology has an idea of what’s going on, and among them we find PLATO .
Fortunately, we can all rest easy now thanks to the wizards of DeepMind, who created the first ever AI with a knowledge of physics comparable to that of a human child, and writing in the magazine Nature Human Behaviorthe researchers explain how we develop “intuitive physics” within the first few months of life, quickly coming to understand some fundamental laws that govern the material world.
For example, children tend to understand the concepts of “permanence” – whereby objects do not simply vanish – plus “solidity” and “continuity”, referring to the inability of objects to cross each other or to suddenly alter their trajectory in time and space.
However, the authors go on to state that “current artificial intelligence systems pale in their understanding of intuitive physics, compared to even very young children.”
To help the robots catch up, the team turned to the field of developmental psychology to develop an AI that can learn in the same way a child does.
For example, at the age of about three months, human babies are able to show surprise when an object disobeys one of the three pillars of our intuitive physics. This ability is known as the violation of expectations (VoE) paradigm and provides the inspiration behind the new artificial intelligence.
How the PLATO program behaves and what it does
Called PLATO – an acronym for Physics Learning through Auto-encoding and Tracking Objects – the deep learning system was trained on a series of videos of balls moving in space and interacting with each other, and this video data set was specifically designed to represent the concepts of permanence, solidity and continuityas well as two additional concepts known as “immutability” and “inertia”.
These refer to the fact that objects do not suddenly alter their basic characteristics or disobey the laws governing speed and direction, and when videos of scenarios were later shown that contravene any of these five principles, the program successfully reacted with a VoE signal.
“After training PLATO on videos of simple physical interactions, we found that PLATO passed the tests in our Physical Concepts dataset”
explained theauthor of the study Luis Piloto in a declarationto which he later added:
“By varying the amount of training data used by PLATO, we found that PLATO could learn our physical concepts with as little as 28 hours of visual experience.”
Surprisingly, AI could even identify transgressions of the laws of physics looking at a separate video dataset with objects he had never seen before.
“PLATO passed, without any retraining, despite being tested on completely new stimuli”
Piloto states.
This change certainly bodes well for the future of AI, since, as Piloto points out,
“If we want to deploy safe and useful systems in the real world, we want these models to share our intuitive sense of physics.”
Of course, there is still a long way to go, but child-like intelligence is not a bad start.
If you are attracted to science or technology, keep following us, so you don’t miss the latest news and news from all over the world!
#PLATO #system #basics #physics