Contrary to the popular belief, brain neurons are not the most important element needed for the brain to function, nor is it the only cell that can process and learn new information. Only recently it was discovered that another brain cell – astrocyte – located between the neurons has even more important functions in the brain. It plays a key role in learning and in the central pattern generator (CPG), responsible e.g. breathing.
Astrocytes, i.e. star-shaped gliale cells, which were considered in the past to be merely a support for neurons in the brain, still cause confusion in neuroscience. Their positive functions include maintaining blood flow in the brain and influencing synaptic conduction, with the negative ones possibly discovering a role in epilepsy and mental retardation1.
Artificial Intelligence (AI) is usually equipped by scientists with a human-like neural structure, and so far the potential inherent in astrocytes has not been noticed. Rutgers University scientists, aware of the shortcomings in this area, have started to work on creating algorithms shaped like the human brain, not only recognizing the superiority of neurons, but also appreciating the role of astrocytes. Their approach will be presented at the ICONS 2020 conference in July, and their work can be read tutaj. Researchers want to dispel the myth that the brain structure typically used in AI systems is not identical to the neuronal structure. Other cells play an equally important role. This approach assumes that neurons and astrocytes can work together. The brain is a complex organ in each cell must work together to function. Using only some cells for AI systems may not give such spectacular results as taking into account all known cells in the brain.
First, scientists focused on what happens inside the astrocytes when they communicate with neurons that receive and send neural impulses. Then they used this model of operation as a basis for building neural-astrocytic networks, which were then built into integrated circuits that can control robots. In their recent research they used neuromorphic chips developed by Intel – Loihi.
In this system, the functions of the robots emerge naturally from the interaction between artificial neurons and astrocytes. Therefore, the structure and functioning of these systems is very different from learning algorithms that focus exclusively on neurons and do not make full use of the current knowledge about brain function.
Typically, neuromorphic chips are used to make simple models of neurons that reproduce only part of the cell activity in the human brain. Rutgers scientists were among the first to successfully create neurons capable of more complex behaviors using a neuromorphic chip. They presented their model to control the walking of a six-legged robot. The system created by them achieved extraordinary results, allowing the robot to move smoothly at different speeds without being affected by external sensory disorders.
Konstantinos Michmizos, assistant professor in the department of computer science at Rutgers University and chief researcher of the project, stated that the brain is the only organ we don’t quite know. Intelligence of the brain and its diseases result from the interaction between neurons and astrocytes.
The innovative approach presented by researchers is likely to provide new opportunities for the use of the AI. By increasing understanding of how astrocytes work in the brain, new ways are found to harness the computational power of these non-neural cells in neuromorphic brain models and make robots behave more like humans.
[1] http://laboratoria.net/aktualnosci/25940.html
