Researchers at the University of Illinois at Urbana-Champaign have unveiled groundbreaking evidence that could revolutionize our understanding of the brain and impact the field of artificial intelligence (AI). Their findings indicate that decision-making processes initiate significantly earlier in the brain than previously thought, paving the way for the development of high-performing and energy-efficient AI systems.
Led by Yury Vlasov, a professor of electrical and computer engineering at the Granger Institute of Technology, the study published in Proceedings of the National Academy of Sciences (PNAS), highlights an unexpected function of the brain’s early sensory regions in decision-making. This challenges the traditional belief that decisions only arise after a strict hierarchy of brain region involvement.
Rethinking Brain Decision-Making Processes
The human brain is often viewed as the most complex structure in the universe, yet its complete functioning remains a mystery. This is why reverse engineering the brain was designated by the National Academy of Engineering in 2008 as one of the 14 Great Engineering Challenges of the 21st Century.
For decades, many AI systems, including convolutional neural networks, have followed the traditional notion that information flows in a unidirectional manner. According to this model, sensory details move upwards through increasingly complex brain regions, culminating in decision-making in the frontal cortex.
Vlasov and his team are now re-evaluating this perspective.
They are investigating models based on natural intelligence refined over millions of years of evolution, where decision-making does not strictly depend on a progressive flow of information. This approach emphasizes interconnected feedback loops, enabling information to travel both ways between various brain regions.
Understanding this intricate architecture may be vital in creating future AI technologies, as biological systems undertake complex tasks with significantly less energy than current AI frameworks.
“We aspire to glean insights from billions of years of evolution,” Vlasov stated. “What architectural strategies does biological intelligence employ? Can these principles be mimicked to enhance AI efficiency, reduce power consumption, and elevate its intelligence? Our focus is on improving decision-making capabilities where current AI falls short.”
Early Brain Regions and Decision-Making
To explore these processes, the research team centered on the initial stages of sensation and perception within the brain.
Utilizing virtual reality environments, the researchers measured neural activity as mice navigated through hallways and made perceptual choices. They discovered decision-related activity in the primary somatosensory cortex (S1), one of the brain’s foundational sensory processing areas.
Contrary to merely forwarding information, S1 appears shaped by higher-order brain regions through feedback loops. This top-down regulation indicates that decision-making involves a continuous exchange of information across multiple brain regions rather than a straightforward upward flow.
The complexity of the brain’s neural code remains largely unexplored, as Vlasov notes. “However, this systems-level understanding holds promise for developing more efficient artificial neural networks and rethinking future AI generations. Analogies drawn from real brain functions could significantly enhance AI design,” he added.
Implications for the Future of AI
The researchers underscore that their study does not serve as a definitive guide for creating superior AI; instead, it offers valuable insights into the brain’s decision-making organization, potentially inspiring future AI frameworks.
Next steps for Vlasov and his team include examining the timing of these brain signals more comprehensively. They aim to devise novel methods for measuring neural activity to deepen their understanding of how feedback loops facilitate and synchronize different levels of brain processing.
“By analyzing the rapid temporal nuances of neural activity, we hope to elucidate how feedback loops participate in decision-making,” Vlasov expressed. “This approach may unlock mechanisms currently not understood, revealing how dynamically organized feedback loops shape various levels of processing and can be integrated into new AI architectures.”
Source: www.sciencedaily.com