What does it take to create AI that truly mimics human perception? Today’s most advanced models rely on massive computational power and vast datasets, but they often operate as black boxes—complex systems whose inner workings remain mysterious. For neuroscientists seeking to understand how real brains process visual information, this approach falls short. It replaces one intricate biological system with another equally opaque artificial one. To bridge this gap, Cold Spring Harbor Laboratory Assistant Professor Benjamin Cowley is taking a different path: thinking small. In a groundbreaking study published in Nature, Cowley, along with colleagues from Carnegie Mellon University and Princeton University, developed a remarkably compact AI model that not only matches but surpasses the ability of large AI systems to predict how monkey brains respond to visual stimuli. The key insight? Simplicity can reveal deeper truths. The team presented macaque monkeys with carefully selected natural images while recording activity in their visual cortex. Using this data, they first trained large AI models to predict neural responses—achieving performance that exceeded existing models by over 30%. Then came the breakthrough: they applied compression techniques to shrink the model to just 1/1,000 of its original size. The resulting system is so small it could fit in an email attachment. This miniature model is more than a technical feat—it’s a window into brain function. By analyzing how the model processes images, the researchers discovered that its internal "neurons" break down visual input into fundamental features like edges, colors, and textures. These components are then combined in unique ways, giving rise to specialized responses. One striking finding: certain neurons in the V4 region of the visual cortex—critical for object recognition—show a strong preference for dots. Why does this matter? Because dots are central to one of the most important visual cues in human life: the eyes. As Cowley notes, “In the monkey’s brain—and in our brains, too, most likely—there's a group of V4 neurons that love dots.” This specialization likely evolved to help detect faces and interpret social signals, such as eye contact, which are vital for communication and survival. Beyond understanding normal vision, the model opens new doors for studying brain disorders. Cowley envisions using such systems to model conditions like Alzheimer’s disease, where synaptic connections deteriorate. “If we know the images that drive neurons to communicate,” he explains, “we might one day reconstruct lost neural pathways.” In the future, this approach could lead to novel therapies—perhaps even using targeted visual stimuli to stimulate or restore brain function. While still speculative, the idea that looking at specific images might help combat neurodegenerative disease is no longer science fiction. Thanks to the power of small, interpretable AI, we’re getting closer to decoding the brain’s deepest secrets—one dot at a time.