Research, Technology

Created artificial intelligence capable of thinking like a baby

Artificial intelligence systems are capable ofa lot - with some tasks they cope even better than people. Relatively recently, I talked about neural networks that can draw by words. People just need to enter a textual description of the future picture and the computer quickly produces a unique result - even the most brilliant artists of the world are not capable of such speed and imagination. Or here's another example: in 2020, artificial intelligence played against a person inside the flight simulator FlightGear and turned out to be faster and more accurate, from which we can conclude that the computer understands military tactics better than humans. However, each artificial intelligence is only good in a certain area - some draw well, others write music, and so on. And so, the computer mind today has not even reached the level of a child. But work on training neural networks is underway, and artificial intelligence has recently been introduced, which is developed in almost the same way as a human baby.

An artificial intelligence has been created with the mind of a child, what can he do?

What can newborn babies do?

What can a child who has justwas born? In the first days of his life, he can only breathe and cry - he needs help to survive. But after a few months, the baby begins to instinctively understand our world. For example, it becomes clear to him that people are not capable of simply disappearing out of the blue. While playing with a child, adults suddenly hide or simply cover their faces with their hands - this causes surprise in children and subsequent laughter.

Newborn babies know little, but are still smarter than artificial intelligence

You will be interested: Can babies cry while in the womb?

AI training

For artificial intelligence, such phenomenaincomprehensible, so scientists decided to create a system that is at least a little aware of the rules of the real world. It was called PLATO, which can be deciphered as "Teaching Physics through Automatic Object Coding and Tracking." To train the system, videos were used that demonstrate five key concepts that children learn in the first weeks of their lives:

  • constancy - objects cannot suddenly disappear;
  • hardness - solid objects cannot pass through each other;
  • continuity — objects consistently move in space and time;
  • immutability - objects cannot just change properties such as shape;
  • directional inertia - objects move according to the principles of inertia.

It is worth emphasizing that some objects are stillable to disappear and change their shape - for example, soap bubbles and so on. But babies do not understand this, because this is the next stage of their development. All of the above concepts were explained to the PLATO artificial intelligence using videos of balls falling to the floor, bouncing off each other, and so on.

Artificial intelligence PLATO knows the basics of physics

After the learning process, the researchers tested,whether artificial intelligence has mastered all the lessons correctly. It turned out that yes - when he was shown videos with impossible phenomena like solid objects passing through each other, he immediately noticed it. The system also recognized violations of the laws of physics in videos that showed objects that were different from balls. Thus, the authors of the scientific work were convinced that the artificial intelligence they created is able to learn and expand its basic educational knowledge. However, as the scientists answered, there is still a lot of work - the system they developed has many shortcomings. For example, he is not "surprised" when he is shown videos in which there is not a single object.

What will happen when Artificial Intelligence reaches its peak? The answer is here.

Why do scientists create artificial intelligence?

But why do researchers even want to createartificial intelligence with the mind of a baby? First, it's just a scientific interest - scientists have long wanted to create a computer endowed with its own intelligence. Secondly, such studies can reveal a lot about the development of intelligence in humans. For example, the authors of the scientific paper noted that their work demonstrated how the basic concepts of physics can be learned through visual learning. It also became clear that even innate knowledge about the physics of objects can be distorted if children are shown videos that violate the laws of physics.

Artificial intelligence can tell a lot about people

The creation of a full-fledged artificial intelligence is still far away. But neural networks are already capable of a lot, here are three examples for you:

  • Artificial intelligence has learned to flirt and confess your love
  • Artificial intelligence will build a dam in China
  • Artificial intelligence wrote an article for the site

What do you think about the results of this scientific work? Share your opinion in the comments or in our Telegram chat.