The fast models of artificial intelligence, which actively entered our lives very far away. They lack the flexibility and energy efficiency of our brain. The creation of new neurological networks basically, close to the most real gray matter of our main body, is participating in scientists of Nizhny Novgorod State University named Ni Lobachevsky.
One of them is the head of the neurological computing center, a professor at the Faculty of Neurological Technology, Doctor of Physics and Mathematics, the presidential prize in the field of science and innovation to young scientists Susann Gordleeva talks about the future of artificial intelligence to bring the help of new mathematical models.
– Susanna, because you want to improve what seems to be perfect for many people?
– For most people, artificial intelligence is familiar with famous language models such as TATGPT and Deepseek. They deal with documents perfectly, perform translations from foreign languages and are actively developing. However, they are still built on the technologies of the previous century – mathematical algorithms created to analyze and process data based on long -term training. Although they are called nerve networks, they have nothing to do with the actual biological networks in our brain. When they are created, neurological physiologists and neurologists do not have enough data on how the brain activity, how the thinking process is done. Mathematicians, equal and large, taught their artificial nerve networks, based on ideas that are very conditional about the work of nerve cells.
– Let us know, what is the main difference between working and creating technologies?
– The architecture of traditional official nerve networks is static. It does not change while working. Based on a specific task, we choose a certain architecture, based on a large amount of carefully prepared data, we train or set the gravity (power) of the connections in this nerve network. The learning process takes a lot of time. After training, nerve networks often do not change their architecture. In the brain, the architecture of the nerve networks, on the contrary, is very flexible, it changes all the time during operation, from any experience achieved. Here we are talking right now – you have received a new information about your neural network and nerve network that has been studied, in real time.
– How did this happen?
– In your brain, regarding the new information obtained, the electrical impulses in the brain cells begins to create, this activity immediately changes the nerve network settings, the weight (force) of plastic contacts between the neurotransmitter – neurotransmitter changes. Well, we have monitoring this mechanism, describing mathematics in our models of processing and storing information for artificial brain, making it biological.
– Can they do similar things around the world? At what level is your scientific group?
– Indeed, this area is developing quite positively around the world and we are not inferior to the world developers in this field, and at some points we even exceed them. In the decade of science and technology announced by the President, we were quite actively developed and from basic research that had shifted to creating laboratories and prototypes. In addition to us, programmers, those who create hardware, meaning Iron Iron developers, work on the issue of creating new systems for AI.
– What are they creating?
– New microCircuits and micro -components help eventually run our programs on the computer. On the old plans, the new architecture of the Viking brain will not work effectively. There is another important task: both of us and micro -electronics experts understand that when creating new artificial nervous networks, we have to search for high energy efficiency, look for less cost architectural principles of computing systems. If we continue to spend such an energy on the work of supercomputers like now and increase consumption with their strength, the whole idea with artificial intelligence will soon go to a dead end – we will not have enough electricity.
– How can this problem be solved?
Once again, again – switching to help our brain … Why are traditional computing devices more thermal machines? Yes, all for them, along with data processing energy, in addition, also spend their energy for their storage and even release the remaining heat that needs to be processed, that is, the start of the unit cooling strongly. This is because traditional computers are done according to the architecture of Von Neumann, in which the information storage unit and its processing unit are divided into physical and between these blocks, the data exchange process continuously occurs. A large amount of time and electricity goes into distillation of these data sets! In our brain, everything occurs in one place: we store information and process it in one place – in nerve cells and connect between them. The nerve cells in the brain are very short and are rarely created by electrical signals, and according to such a rare pulse principle is the information processed in the brain and the knowledge of Islam. Our brain is a huge economy in terms of energy consumed compared to computers.