Artificial intelligence goes beyond a simple neural network

(ORDO NEWS) — In an attempt to recreate the work of the human brain in an artificially created algorithm, scientists turned their eyes to astrocytes – specialized nerve cells that work together with neurons.

Most mechanisms and computer models controlled by artificial intelligence are based on the method of interaction of neurons – specialized brain cells between which a nerve impulse is transmitted. In machines, only part of the brain activity is reproduced and the influence of other cells of the nervous system on signal transmission is completely ignored.

For more than a hundred years, scientists believed that neurons are the only participants in a complex “mail” system that delivers messages from the brain to “recipients” – organs, tissues, cells – and vice versa. The true purpose of other nerve cells – astrocytes, which occupy twice as much space in the brain as neurons, has only recently been identified.

Astrocytes, reminiscent in their form of stars with rays, do not just fill the “empty” space, as they thought about them. The “star” cells, as sorting centers, control the direction with the method of transmitting “messages” exchanged between neurons. It is astrocytes that are responsible for training and creating patterns for the same rhythmic actions, such as walking, and also participate in sending their “postmen” to peace – helping to remove dead neurons from the brain.

Scientists from Rutgers University were the first to transfer the still poorly studied functions of astrocytes to artificial intelligence and fundamentally changed the structure, work and training of artificial neural networks. In an article that was published on the arXiv preprint site and will be presented at the ICONS 2020 conference in July, researchers showed what happens inside astrocytes when they interact with neurons, and also introduced the first robot controlled by the neural-astrocytic network.

Artificial intelligence goes beyond a simple neural network 2

It turned out that astrocytes, receiving information about the environment, interact with neurons and determine the nature of their interaction with each other. When scientists placed the neuromorphic central pattern generator (CPG), equipped with artificial astrocytes, in special neuromorphic chips controlling a six-legged robot, they saw a fundamentally different movement: smoother and more “lively”.

Konstantinos Michmizos, associate professor of computer science at Rutgers University and lead author of the study, believes that the control signals that artificial astrocytes transmit to artificial neurons will make robots less dependent on environmental factors and will allow better control of movement. The method of constructing artificial intelligence using the functions of astrocytes has already been included in the training programs of students at Rutgers University.


Contact us: [email protected]

Our Standards, Terms of Use: Standard Terms And Conditions.