(ORDO NEWS) — Artificial intelligence (AI) systems are already far ahead of us in certain areas – like playing Go or processing huge amounts of data – but in other areas, AI is still far behind humans, even just a few months after we were born.
For example, even small children instinctively know that one object passing after another for a short time should not disappear and reappear in another place. Seeing such a magical action, the kids are surprised.
But such a simple continuity rule, along with other basic physical laws, was not so intuitive for AI. Now a new study is introducing an AI called PLATO that was inspired by research into how children think and learn.
PLATO stands for Teaching Physics by Automatic Coding and Tracking of Objects and has been taught through a series of coded videos designed to represent the same basic knowledge that children acquire in the first few months of life.
“Fortunately for us, psychologists have spent decades studying what babies know about the physical world,” says neuroscientist Luis Piloto of the DeepMind artificial intelligence research lab in the UK.
“By extending their work, we have created and opened up a dataset of physics concepts.
This synthetic video dataset draws inspiration from original developmental experiments to evaluate physics concepts in our models.”
There are three key concepts that we all understand from a very young age:
- Persistence (objects do not disappear suddenly)
- Strength (solid objects cannot pass through each other)
- Continuity (objects consistently move in space and time)
The dataset the researchers created covered these three concepts, plus two additional ones:
- Immutability (properties of an object, such as shape, do not change)
- Directional inertia (objects move according to the principles of inertia)
These concepts were laid out with cutscenes of balls falling to the ground, bouncing off each other, disappearing behind other objects and then reappearing, and so on. Having trained PLATO with this video, the next step was to test him.
When the AI was shown videos of “impossible” scenarios that contradicted the physics it had learned, PLATO expressed surprise: it was smart enough to admit that something strange had happened that violated the laws of physics.
This happened after relatively short training periods, in some cases as little as 28 hours. From a technical standpoint, as in the infant studies, the researchers looked for evidence of expectation-defying (VoE) signals showing that the AI understands the concepts it has been taught.
“Our object-oriented model demonstrated robust VoE effects in all five concepts we studied, despite the fact that we trained on video data in which specific sensing events did not occur,” the researchers write.
The team ran further tests, this time using different objects from those in the training data. Again, PLATO has shown a clear understanding of what should and should not happen, demonstrating that he can learn and expand his basic knowledge.
However, PLATO has not grown to the level of a three-month-old baby yet. The AI was less surprised when it was shown scenarios in which there were no objects, or when the test and training models were similar.
What’s more, the videos PLATO was trained on included additional data to help it recognize objects and their movement in three dimensions.
It appears that some additional knowledge is still required to get the full picture. Research can give us a better understanding of the human mind and also help us build a better understanding of it in AI.
“Our simulation work provides a proof of concept, demonstrating that at least some of the core concepts in intuitive physics can be acquired through visual learning,” the researchers write.
“While studies in some presocial [advanced-born] species indicate that certain basic physical concepts may be present from birth, in humans, evidence suggests that intuitive physical knowledge emerges early in life but can be influenced by visual experience.”
Contact us: [email protected]