Scientists create AI that can think like a baby

Creating an AI system capable of thinking like a human has been one of the most difficult tasks in computer science.

Now researchers claim to have created an AI that can think like a baby by teaching it the basic rules of the physical world.

Their deep learning system can learn “intuitive physics” — common sense rules about how physical objects interact.

During the experiments, the scientists trained a new system called PLATO with a set of animated slides with moving balls.

After being trained with a small set of visual animations, PLATO could demonstrate learning and even be “surprised” if the ball moved in an impossible way.

Researchers claim to have created an AI that can think like a baby by teaching the basic rules of the physical world. In the experiments, the scientists trained a deep learning system called PLATO with a set of animated ball motion slides (pictured).

INTUITIVE PHYSICS

The researchers explain that even very young children know about “intuitive physics” – the common sense rules of how the world works.

Intuitive physics is the common sense knowledge we use to understand how objects behave and interact.

Those who are familiar with intuitive physics have an idea of ​​how two objects can interact.

Whether we were born with intuitive physics or learned it quickly is a matter of scientific debate.

The new study was conducted by researchers at Princeton University in New Jersey. University College London and Google-owned company DeepMind published in Nature Human Behavior.

According to them, their findings are important for creating AI models that have the same physical understanding as adults.

“Understanding the physical world is a critical skill that most people use effortlessly,” said study author Dr. Luis S. Piloto of DeepMind.

“However, this still presents a challenge for artificial intelligence — if we want to deploy useful systems in the real world, we want these models to share our intuitive understanding of physics.”

In 1950, the legendary British computer scientist Alan Turing proposed to train an AI to give it the intelligence of a child, and then provide the appropriate experience to bring its intelligence to that of an adult.

“Instead of trying to create a program that mimics the adult mind, why not try to create a program that mimics the mind of a child?” Turing wrote in Computing and intelligencehis seminal research work.

In 1950, the legendary British computer scientist Alan Turing (pictured) proposed an AI learning theory to give it the intelligence of a child and then provide the appropriate experience to raise its intelligence to that of an adult.

In 1950, the legendary British computer scientist Alan Turing (pictured) proposed an AI learning theory to give it the intelligence of a child and then provide the appropriate experience to raise its intelligence to that of an adult.

WHAT IS THE GOOGLE DEEPMIND PROJECT?

DeepMind was founded in London in 2010 and acquired by Google in 2014.

It now has additional research centers in Edmonton and Montreal, Canada, and a DeepMind Applied team in Mountain View, California.

DeepMind aims to push the boundaries of AI by developing programs that can solve any complex problem without the need for training.

The company made headlines for a number of its creations, including software it created that taught itself how to play and win 49 completely different Atari games using only raw pixels as input.

He is also known for creating artificial intelligence that defeated professional go player Lee Sedol, world champion, in a five-game match in 2016.

“If this was then subjected to an appropriate course of study, an adult brain could be obtained.”

The authors of this new study explain that even very young children are aware of “intuitive physics” – the common sense rules of how the world works.

For example, if someone were to wave their keys in the air and announce that they were going to let them go, everyone would know that unsupported objects do not float in the air.

They would also know that two objects—like the keys and the table underneath—do not pass through each other. Hence, people expect the keys to fall until they hit the table.

This knowledge is not unique to adults—even three-month-old babies have such expectations, and they react if they encounter a “magic” situation that seems to violate those expectations.

For example, infants at five months of age are surprised when they are shown a situation involving a physically impossible event, such as the sudden disappearance of a toy.

In their study, the scientists wondered if AI models could learn a diverse set of physics concepts, in particular those that infants understand, such as density (two objects don’t pass through each other) and continuity (objects don’t blink or flicker). outside of existence.

They built the PLATO artificial intelligence system so that it can represent visual input as a set of objects and reason about the interactions between objects.

The authors taught PLATO by showing him videos of many simple scenes such as the ball falling to the ground, the ball rolling behind other objects and reappearing, and the balls bouncing off each other.

The researchers wondered if AI models could learn a diverse set of physics concepts, in particular those that babies understand, such as density (two objects don't go through each other) and continuity (objects don't blink or disappear).

The researchers wondered if AI models could learn a diverse set of physics concepts, in particular those that babies understand, such as density (two objects don’t go through each other) and continuity (objects don’t blink or disappear).

After training, PLATO was tested by showing him videos that sometimes contained impossible scenes, such as the disappearance and appearance of the balls on the other side of the frame.

Like a small child, PLATO showed “surprise” when shown something that made no sense, such as objects moving through each other without interacting with each other.

“One interpretation of the definition of surprise is the expectation of seeing something and getting a different result,” Dr. Piloto said.

PLATO makes predictions about the configuration of the objects it will next observe. When the video is played, it then observes the actual configuration of the objects.

“The surprise is the difference between the predicted configuration and the actual configuration in the next video frame.”

These learning effects were seen after watching just 28 hours of video.

The authors conclude that PLATO can offer a powerful tool for investigating how people learn intuitive physics.

The results also show that deep learning systems modeled after an infant outperform more traditional “learning from scratch” systems.

“The findings in this article suggest that Turing may have been right,” Susan Hespos and Apoorva Shivaram say in a cover letter. News and views a piece.

“Common sense physics is a situation in which development refines and refines knowledge without fundamentally changing it.

“This means that learning about object knowledge in infancy could provide insight into adult object knowledge and potentially guide us on how to build better computer models that mimic the human mind.”

“THE GAME IS OVER!” GOOGLE DEEPMIND SAYS IT’S CLOSE TO HUMAN-LEVEL AI BUT IT STILL NEED TO BE EXPANDED

DeepMind, a British company owned Googlemay be on the verge of achieving human-level artificial intelligence (AI).

Nando de Freitas, DeepMind Scientist and Professor of Machine Learning at Oxford Universitysaid it was “game over” in regards to tackling the toughest challenges in the race to build artificial general intelligence (AGI).

AGI refers to a machine or program capable of understanding or learning any intellectual task that a human can perform and do so without learning.

According to De Freitas, scientists are now looking to expand AI programs, for example with more data and processing power, to create AGI.

De Freitas’ comments appeared in response to an author’s article published on Next network that people living today will never achieve AGI.

In May 2022, DeepMind introduced a new AI “agent” named Gato, which can perform 604 different tasks “in a wide variety of environments”.

Read more