Your Brain Starts Making Decisions Before Your Thinking Brain Gets Involved
A study published in the Proceedings of the National Academy of Sciences is changing how scientists think the brain handles decisions, and by extension, how AI systems are designed. The research, led by electrical engineering professor Yurii Vlasov at the University of Illinois Urbana-Champaign, found that decision-making signals appear deep in the brain's first sensory processing stages, not just in the frontal cortex where textbook accounts say choices are made.
What neuroscience thought it knew
For decades, both neuroscience and artificial intelligence were built on the same assumption. Sensory information flows upward through increasingly complex brain regions, gets processed layer by layer, and eventually reaches the frontal cortex where a decision emerges. Every major neural network architecture is based on this bottom-up, feed-forward picture.
That picture now looks incomplete.
What mice in virtual reality revealed
The Illinois team designed a tactile virtual reality experiment for mice stripped down to almost nothing: they trimmed every whisker except one matched pair, the C2 whiskers, and set the animals loose in a corridor they had to navigate by touch alone. Mice learned to turn correctly about 80% of the time within three or four sessions, without any reward training or shaping protocols.
What the researchers found in the neural recordings was not what they expected. In the primary somatosensory cortex (S1), the very first stage where touch information enters the brain, neural activity did not simply relay data upward. During the evidence-accumulation phase before each turn, activity across hundreds of neurons collapsed from a scattered, high-dimensional pattern down to a single coordinated variable. That variable then ramped steadily upward until the mouse made its choice.
That ramp pattern is what theorists have long associated with decision-making. But it was supposed to appear much later in the brain, in premotor and frontal regions. Finding it in a sensory area, before information had even reached those downstream regions, means the brain's decision process starts far earlier and involves more of the brain than anyone had mapped.
The feedback loops the textbooks left out
Further analysis showed that S1 is not acting alone. Higher-level brain regions send information back down to it through rapid feedback loops, dynamically shaping how sensory areas process incoming data in real time. Decision-making is not a relay race. It is a continuous conversation between regions, running in both directions at once.
"The neural code of the brain is still mostly an unknown language," Vlasov said. "But this systems-level understanding can be viewed as a potential impact on how more efficient artificial neural networks can be built."
Why this matters for artificial intelligence
Current AI systems, including the convolutional neural networks that power image recognition and large language models, are built on feed-forward architectures. Data flows in one direction. The efficiency gap between these systems and biological brains is enormous: humans make complex decisions using around 20 watts of power; large AI models require data centers burning megawatts.
If the brain's efficiency comes partly from these bidirectional feedback loops, building them into AI from the start could reduce the energy cost of reasoning significantly. Vlasov is now investigating how fast temporal dynamics in neural activity engage those feedback loops during decision-making, with the goal of giving engineers a concrete architectural roadmap.
The study was published in PNAS (Vol. 123, No. 18; DOI: 10.1073/pnas.2514107123). The paper was led by Alex G. Armstrong and Yurii Vlasov at the Grainger College of Engineering.