Between lines and neurons: Can AI replace the human brain
The human brain remains one of the most complex and poorly understood systems in nature. Despite decades of research, scientists continue to rethink the fundamental concepts of how it processes information and makes decisions. One of these discoveries was recently presented by a group of researchers from the University of Illinois at Urbana-Champaign. How new data can affect the development of artificial intelligence, medicine and technologies of the future — in the material of Izvestia.
How the brain makes decisions
For a long time, scientists have imagined the brain as a complex but consistent information processing system. According to the classical model, the signal from the sensory organs first enters the primary sensory zones, then passes through several levels of analysis and only after that reaches the areas of the brain responsible for complex cognitive processes and decision-making. In this scheme, the sensory areas acted more as "receivers" and transmitters of information, and the final choice was considered a function of the higher parts of the cortex.
However, a new study by a group of scientists from the University of Illinois at Urbana-Champaign shows that this mechanism can be much more complicated. The work led by Professor Yuri Vlasov casts doubt on the idea that a decision is formed only at the last stages of information processing.
During the experiment, the researchers observed the activity of nerve cells in mice that were moving along a virtual corridor and had to choose the direction of movement depending on the signals they received. An analysis of the work of the animals' brains showed that signs of decision formation occur already in the primary somatosensory cortex (S1), an area that was previously considered primarily as the first stage of processing incoming sensations.
According to scientists, the brain does not work as a linear system of data transmission "from bottom to top", but as a dynamic network in which different levels constantly interact with each other. The higher parts of the brain do not just receive ready-made information from sensory areas, but actively influence which signals will be processed and how significant they will be.
Dmitry Repin, Chief Researcher at the A.A. Harkevich Institute of Information Transmission Problems of the Russian Academy of Sciences, Doctor of Medicine and Doctor of Sociological Sciences, notes that it is the change in ideas about the role of primary brain areas that is the main result of the study.
"Previously, it was believed that a solution was born only in the uppermost, most "smart" parts of the brain, and everything that happens at the level of primary perception of sensations is just the collection of raw data for subsequent transmission to the top," explains the expert.
According to him, the traditional model of the brain could be compared to a conveyor belt: information comes from the senses, then is sequentially transmitted between different structures until it reaches the area where the final analysis and selection takes place.
— The authors found that already at the very first stage of sensation processing — where the signal first arrives — there are signs of that very "solution". That is, it begins to take shape much earlier than expected," says Dmitry Repin.
However, the new study does not mean that the sensory areas of the brain independently make decisions in the usual sense of the word. We are talking about a more complex process in which different parts of the brain simultaneously participate in the formation of a reaction. The primary zones receive not only incoming signals from the outside world, but also feedback from higher brain regions.
— The brain does not work as a strict conveyor, but rather as a team, where everyone consults with each other at the same time, rather than passing the task strictly along the chain from one department to another, — the expert notes.
It is the presence of such feedback loops that may explain why the human brain is able to quickly adapt to environmental changes. Instead of completely processing all incoming information every time, he constantly corrects perception, highlighting the most important signals.
How knowledge about the brain affects AI
Today, most modern algorithms are based on the principle of sequential information processing: data is received at the input, passes through many computational layers, and only after that the answer is formed. This approach allows you to create powerful models, but requires significant resources, both computational and energy.
Alexander Kugaevskikh, PhD, associate professor at the Faculty of Software Engineering and Computer Technology at ITMO University, told Izvestia that one of the main problems of modern AI is that it is still far from the principles of the human brain.
— The human brain consumes only about 20 watts. The fact is that memory and computing are combined in the brain: synapses simultaneously store information and participate in signal transmission," the expert explains.
According to him, in computer systems, a significant part of the energy is spent not directly on computing, but on moving data between memory and the processor. In the brain, these processes are combined in one structure, which allows it to work much more efficiently.
Another important difference is related to how different parts of the system are activated. The human brain does not use all its resources at the same time: at a particular moment, only a part of the neurons necessary to perform a specific task are working. Computer models, by contrast, often perform billions of operations even in cases where some of the information is not essential.
— The biological neuron fires only when it is necessary. At the same time, modern AI models continue to perform a large number of mathematical operations, even with minor changes in input data," notes Alexander Kugaevskikh.
That is why the study of brain mechanisms can become the basis for creating more cost-effective algorithms. If modern neural networks can be compared with a system that goes through a long route every time, then future models can learn to act more flexibly — to use the information they have already accumulated, quickly assess the situation and connect additional computing resources only if necessary.
As Dmitry Repin emphasizes, an important conclusion of the study is the very idea of distributed decision-making.
"Most of the artificial intelligence systems we are used to are designed like this: a signal enters from one side, passes through many internal layers and exits from the other — and for each new answer, we have to run a huge amount of calculations through this entire chain again," says the expert.
According to him, a more biological architecture could allow artificial systems to evaluate some of the information directly at the early stages of processing.
— The brain, apparently, is designed much more ergonomically: a significant part of the preliminary assessment takes place immediately on the spot, where the signal first appears, and only what is really needed for a specific task is included in the work, — explains Dmitry Repin.
Neuromorphic computing, systems that attempt to replicate the principles of biological neurons, may become one of the possible directions for the development of such technologies. Unlike the usual artificial intelligence models based on the mathematical representation of a neuron, they use more complex signal transmission mechanisms.
— There are spike neural networks that are built not on a formal neuron, which is familiar to AI, but on one of the models of a biological neuron. A formal neuron has one discrete state, while a biological neuron has more than 30. This can potentially significantly reduce energy consumption," says Alexander Kugaevskikh.
However, according to the expert, so far such technologies remain mostly experimental. Despite decades of research, spike neural networks have not yet become a full-fledged alternative to modern artificial intelligence architectures.
— At the same time, research is already underway to add dynamic feedback to the architectures of transformers, on which modern artificial intelligence is built. In the meantime, many systems use chains of reasoning to simulate such feedback, but this mechanism does not exist inside the neural network itself, but above it," notes Alexander Kugaevskikh.
In the future, the transition to a more "biological" AI may change not only the algorithms themselves, but also the devices that people use every day. We are talking not only about large computing centers, but also about technologies that require autonomy: robots, medical devices, wearable electronics and industrial monitoring systems.
— It is possible to imagine devices that make decisions directly on the spot — on the principle of "here and now", without constant communication with large computing centers and at the same time with minimal energy consumption. This is especially important for autonomous robots, medical sensors, and industrial systems," says Dmitry Repin.
Such technologies can make digital devices more independent: for example, a medical device can analyze a patient's condition faster without sending data to a remote server, and a robot can quickly respond to environmental changes without delays associated with information transmission.
How new understanding of the brain can change medicine
One of the promising areas may be the treatment of conditions associated with impaired perception, movement, and sensory information processing. According to Dmitry Repin, this will change the approach to finding targets for medical treatment.
—If the decision is indeed partially formed already at the level of primary perception of sensations, and not only in the upper parts, this significantly complements the idea of which parts of the brain need to be paid attention to in case of chronic pain, certain motor disorders or problems with perception of one's own body," the expert notes.
For example, in chronic pain, the disorder may be related not only to the source of the damage, but also to how the brain processes and interprets incoming signals. A deeper understanding of these processes in the future may help to create more accurate therapies that affect the mechanisms of formation of a pathological reaction. At the same time, the expert emphasizes that so far we are not talking about ready-made medical solutions, but about a promising area of research.
Another area of application may be neural interfaces, technologies that allow direct exchange of information between the human nervous system and a computer. Today, such devices are already used in experimental medicine to help people with motor disorders, for example, to control prosthetics.
However, modern neural interfaces face the problem of finding the most accurate sources of information in the brain. Usually, researchers' attention is focused on complex departments related to the preparation of actions and conscious intentions. New data can expand the capabilities of such technologies. If important information does appear already in the early stages of signal processing, scientists will receive additional points for interaction with the brain.
— Such devices are most often configured to "read" a ready-made signal from the upper, most complex parts of the brain. If the necessary information is present at the very first stage of sensation processing, this provides additional, earlier points from where the signal can be taken, which means a potentially faster and more accurate device response," the expert explains.
In the future, this may lead to the creation of more advanced medical systems, such as prosthetics that respond more quickly to human intentions, or devices for restoring lost functions that work through two-way information exchange with the brain.
At the same time, the development of such technologies raises not only scientific, but also ethical issues. The more deeply scientists study decision-making mechanisms and gain the opportunity to interact directly with the brain, the more important it becomes to protect human autonomy.
"The closer such devices come to directly reading and affecting the very parts of the brain where the solution is born, the more acute the issues of human consent, protection of personal data about the work of his brain and reasonable boundaries of intervention become," Dmitry Repin notes.
According to the expert, the development of neurotechnologies requires not only scientific discoveries, but also clear principles of their application. That is why, in 2024-2026, the Institute of Information Transmission Problems of the Russian Academy of Sciences developed a draft code of ethics for the creation and use of devices for brain-computer interaction.
Where is the boundary between man and machine
In the future, an important focus will be not only the ability of machines to solve complex tasks, but also the ability to do it more efficiently with less energy and resources.
— One of the main brakes on the development and implementation of AI now is its excessive energy consumption. Improving energy efficiency will not only bring us closer to creating strong AI, but will also significantly accelerate the development of robotics and autonomous unmanned vehicles," notes Alexander Kugaevskikh.
He also emphasizes that more cost—effective systems can be one of the steps towards creating general-purpose artificial intelligence (AGI), technologies capable of adapting to new conditions and performing a wide range of tasks.
At the same time, it's not just about large digital platforms. Energy-efficient AI can transform everyday devices by making them more autonomous. Today, many intelligent functions depend on a constant connection to cloud services: data is sent to a remote server, where it is processed, after which the device receives a ready response.
Systems based on principles similar to the work of the brain can change this model: part of the calculations will take place directly inside the device, without accessing external data centers.
"Devices inspired by how economically the brain works are theoretically capable of reducing costs many times, or even orders of magnitude, and making possible the appearance of truly "smart" small devices that do not need to constantly communicate with a remote server," Dmitry Repin believes.
However, the development of more advanced artificial systems inevitably raises the question of the boundaries between humans and machines. The closer technology gets to the principles of how the brain works, the more important it becomes to understand what exactly distinguishes human thinking from the computational process.
Переведено сервисом «Яндекс Переводчик»
