Scientists at the University of Illinois at Urbana-Champaign have discovered evidence that could change the way researchers think about both the brain and artificial intelligence. Their findings suggest that decision-making begins much earlier in the brain than traditional theory suggests, and provide new ideas for designing future AI systems that are higher-performing and much more energy-efficient.
The study, led by Yury Vlasov, professor of electrical and computer engineering at Granger Institute of Technology, Proceedings of the National Academy of Sciences (PNAS). The study points to an unexpected role for the brain’s early sensory regions in decision-making and challenges the long-held view that decisions emerge only after information has passed through a strict hierarchy of brain regions.
Rethinking how the brain makes decisions
The human brain is widely considered to be the most complex structure in the known universe. Scientists still don’t fully understand how reverse engineering the brain works. That’s why reverse engineering the brain was recognized by the National Academy of Engineering in 2008 as one of the 14 Great Engineering Challenges of the 21st Century.
For decades, many artificial intelligence systems, including convolutional neural networks, have been inspired by the idea that the brain processes information unidirectionally. According to this traditional model, sensory information travels upward through increasingly complex brain regions until it reaches the frontal cortex, where decisions are made.
Vlasov and other researchers are increasingly questioning whether the picture is complete.
Instead, they are exploring models based on natural intelligence that has been refined over hundreds of millions of years of evolution. In this framework, the brain does not rely solely on a gradual flow of information. Decision making also relies on interconnected feedback loops that allow information to move in both directions between brain regions.
Understanding this architecture could help in the development of future artificial intelligence, as biological intelligence performs highly complex tasks with far less energy than today’s AI systems.
“We want to learn from a billion years of evolution,” Vlasov said. “How is that biological intelligence organized architecturally? Can we learn from and emulate the architectural aspects of the brain to make AI more efficient, less power consuming, and more intelligent than it is today? At the decision-making level, that’s where current AI is lacking.”
Early brain regions show decision-making activity
To investigate how these processes work, the research team focused on the early stages of sensation and perception in the brain.
The researchers recorded the neural activity of mice as they moved through virtual reality hallways and made perceptual decisions. They found evidence of decision-related activity in the primary somatosensory cortex (S1), one of the brain’s earliest sensory processing regions.
Rather than simply passing information forward, S1 appears to be influenced by higher-order areas of the brain through feedback loops. This top-down regulation suggests that decision-making requires continuous communication across multiple brain regions, rather than a simple unidirectional flow of information.
“The brain’s neural code is still a largely unknown language,” Vlasov says. “But this system-level understanding can be seen as having potential implications for how we can build more efficient artificial neural networks, and thus how we can think through the next generation of AI. Perhaps we can use these analogies that we learn from real brains to further improve AI.”
What the findings mean for the future of AI
The researchers stress that their study does not provide a blueprint for building better artificial intelligence. Instead, it provides new insights into how the brain organizes decision-making and could ultimately provide inspiration for future AI architectures.
Next, Vlasov and his team plan to investigate the timing of these brain signals in more detail. They also plan to develop new techniques to measure neural activity to better understand how feedback loops occur and coordinate different levels of brain processing.
“By observing the fast temporal dynamics of neural activity, we may be able to better understand how these feedback loops are involved in decision-making,” Vlasov said. “Perhaps it’s an approach that could reveal mechanisms that are currently unknown, how feedback loops are dynamically organized and shape and shape different levels of processing. Maybe it could be implemented in new architectures for AI.”
