In a groundbreaking development at the intersection of artificial intelligence and neuroscience, researchers have unveiled an AI model that can effectively decode the neural circuits of mice, silencing them to observe resulting behavior with unprecedented precision. The findings, detailed in the article “AI trained on the brain reveals how animal behavior changes when specific neurons are silenced,” published by Tech Xplore in February 2026, open new avenues for understanding how distinct neuron populations contribute to complex behaviors in animals.
The research, conducted by a team from the Flatiron Institute’s Center for Computational Neuroscience and the University of California, San Francisco, represents a significant advance in functional brain mapping. At the core of the work is a machine learning system trained on recordings from hundreds of neurons in the visual cortex of mice. Rather than just analyzing neural data, the AI model is designed to simulate what happens when specific neurons are selectively silenced—effectively offering researchers a virtual experiment to predict real-world behavioral outcomes.
By mimicking the effects of inhibitory interventions in silico, the team was able to highlight the contributions of particular neurons in behavioral tasks such as visual orientation and motion detection. The model’s predictions closely matched the results of actual neural inactivation experiments conducted using optogenetics, a technique that enables precise control of neurons with light.
What distinguishes this approach is the model’s interpretability. Built on a relatively simple architecture compared to other deep learning systems, the AI can distinguish between different functional groups of neurons, classifying them based on their influence on behavior. This allows neuroscientists to probe how specific neural subpopulations interact and influence perception—providing a mechanistic understanding that has historically been elusive.
The research not only deepens scientific understanding of the brain’s internal code but could also have broader implications. Better models of neural activity could inform improved brain-computer interfaces, open up new treatments for neurological disorders, and contribute to the development of more sophisticated artificial intelligence systems modeled on biological brains.
Despite these promising results, the researchers acknowledge that more work is needed. The current model is based on a well-characterized region of the brain in a specific animal under tightly controlled conditions. Generalizing the method to other brain areas, more complex behaviors, or different species presents notable scientific challenges.
Nonetheless, as highlighted in the Tech Xplore article, this study marks a critical step toward a future in which artificial intelligence can seamlessly complement neuroscience, offering tools that are not merely descriptive but predictive and explanatory. In doing so, the work paves the way for a deeper, more nuanced understanding of the brain’s inner workings.
