The American company IBM is developing an artificial intelligence system based on the model of the human brain. At the moment, the new neural network has already been taught to think logically, to understand complex relationships between objects, and in the future they plan to improve its ability to pay attention and produce and retain memories.
Today, artificial intelligence technologies are capable of demonstrating superficially human traits. For example, some are able to perform activities usually associated only with a person - writing songs, teaching, or, for example, creating works of visual art.
However, as technology advances, companies and developers are rethinking the basis of artificial intelligence, better understanding our own minds and how we can efficiently model it (using machinery and software). IBM is one such company, as it has already embarked on an ambitious goal of teaching AI to work like the human brain, according to Futurism.
Many of the existing machine learning systems rely on blocks of data (whatever work they do). However, this support has limitations - unlike the human brain.
We learn progressively and, in addition, we use logic to solve problems - modern AI is built on a different principle. However, DeepMind has reportedly developed a neural network that uses rational reasoning to complete tasks.
Timothy Lilicrap, a computer scientist at DeepMind, noted that scientists gave AI a special task and many objects to operate, thereby stimulating the neural network to find existing relationships. So, for example, the system was asked: "Does the object opposite to the blue object have the same shape as the tiny light blue object to the right of the gray metal ball?" In such tests, the neural network determined the required subject in 96% of cases (traditional machine learning models usually succeed in 42-77% of cases).
Scientists from IBM are going to improve the new neural network, says researcher Irina Rish (Irina Rish). Improve the ability of the algorithm to pay attention, as well as to produce and retain memories; developers want to create an architecture that would allow neural networks to develop on their own (just like a person, by trial and error).