Creating an artificial intelligence system that can think like a human has been one of the biggest challenges in computer science.
Now, researchers claim to have created an artificial intelligence that can think like a child, by teaching it the basic rules of the physical world.
And a special deep-learning system can learn "intuitive physics" - common sense rules for how physical objects interact, according to an RT report.
In experiments, the academics trained the new system, called PLATO, with a set of moving balls.
Trained with a small set of visual animations, PLATO was able to demonstrate learning and even "surprise" if the ball moved in an impossible way.
The new study was conducted by experts at Princeton University in New Jersey, University College London and Google's DeepMind, and was published in Nature Human Behavior.
They say their findings are important in the quest to build artificial intelligence models that have the same physical understanding of adult humans.
In 1950, the legendary British computer scientist Alan Turing suggested training an artificial intelligence to give it that of a child, and then providing the right experiences to build the intelligence of an adult. Instead of trying to produce a program that simulates the adult mind, why not try to produce a program that simulates a child?
And the authors of this new study explain that even very young children are familiar with "intuitive physics" - the common-sense rules for how the world works.
For their study, the researchers wondered whether AI models could learn a variety of physical concepts — specifically those that young children understand, such as rigidity (that two objects do not pass through one another) and continuity.
And they built a PLATO system, which could represent the visual input as a set of objects and reason around the interactions between the objects.
The authors trained PLATO by showing videos of several simple scenes, such as balls falling to the ground, balls rolling behind other objects and reappearing, and balls bouncing off each other.
After training, PLATO was tested by showing videos that sometimes contained impossible scenes, such as balls disappearing and reappearing on the other side of the frame.
And just like a little kid, PLATO showed 'surprise' when anything appeared to him that made no sense, like objects moving through each other without interacting.
PLATO makes predictions about the composition of the objects it will then monitor. As the video plays, it then notes the actual composition of the objects. The surprise is the difference between the composition he predicted and the actual composition in the next frame of the video.
These learning effects were seen after watching no more than 28 hours of videos.
The researchers concluded that PLATO could provide a powerful tool for researching how humans learn intuitive physics.
The results also show that infant-style deep learning systems outperform more traditional "learning from scratch" systems.