A new study conducted by Charité — Universitätsmedizin Berlin has found that nerve cells in the human neocortex are wired differently than in mice. Human neurons communicate in one direction, while in mice, signals tend to flow in loops. This unique characteristic increases the efficiency and capacity of the human brain to process information, which could potentially aid in the development of artificial neural networks. The neocortex is a critical structure for human intelligence, with 20 billion neurons processing sensory perceptions, planning actions, and forming consciousness. Understanding how these neurons are wired is key to understanding information processing in the brain.
Previously, neural architecture in the neocortex was primarily based on findings from animal models such as mice. In these models, neighboring neurons communicate in a dialogue-like manner, with signals flowing in recurrent loops. However, the human neocortex is much thicker and more complex, leading researchers to question if the same connectivity principles applied. By examining rare tissue samples and using advanced technology, researchers found that human neurons do not engage in recurrent loops like in mice. Instead, information tends to flow in one direction, without returning to the starting point directly or via cycles.
To study the flow of signals between neighboring neurons in the human neocortex, researchers developed an improved version of the “multipatch” technique. This technique allowed them to listen in on the communications between multiple neurons simultaneously, mapping a network of nearly 1,170 neurons with about 7,200 possible connections. They found that only a small fraction of neurons engaged in reciprocal dialogue, with the majority of information flowing forward. This forward-directed signal flow was shown to have benefits in terms of processing data, as demonstrated through a computer simulation of speech recognition tasks.
The directed network architecture observed in humans is more powerful and resource-efficient compared to models based on mice. This architecture allows more independent neurons to handle different tasks simultaneously, resulting in the ability to store more information within the local network. While it is unclear if these findings extend to other cortical regions or explain uniquely human cognitive abilities, they offer insights into cost-efficient information processing in the human neocortex. These insights could provide inspiration for refining artificial neural networks and optimizing their performance for various tasks.
AI developers have often looked to biological models for inspiration in designing artificial neural networks. The findings from this study suggest that the forward-directed connectivity observed in the human brain is a cost-effective approach to information processing that can lead to better results for certain tasks. By understanding the principles underlying the human network architecture, developers can potentially enhance the performance and efficiency of AI networks. Overall, these insights into the structure and function of the human neocortex could open up new avenues for innovation in artificial intelligence.
