Though not normally desired as houseguests, rats and mice play an important role in scientific discovery, particularly in the realm of medicine. According to a report published by Missouri Medicine, humans and rodents are much more similar than most would believe. The three species share 95% of their genes. Coupled with rats’ lower gestation period, sexual maturity development, lifespan, and required living space, they make an almost perfect subject for testing biomedical advancements that could eventually find application in humans. But, what if instead of using them as an analog for humans, laboratories employed them as a stretch goal for AI? Courtesy of iStock.com/mikhasik. While not typically known for their eyesight, Davide Zoccolan of the Visual Neuroscience Lab of the Scuola Internazionale Superiore di Studi Avanzati (SISSA) in Italy recognized that rats have highly sophisticated ocular abilities and can recognize the most important parts of a scene as it shifts. Therefore, he and a team of researchers attempted to replicate rats’ ability to recognize objects under various conditions using a convolutional neural network (CNN). CNNs are advanced AI tools for image recognition and are inspired, at least in part, by the functioning of the mammalian visual cortex. A CNN consists of multiple layers of processing, each playing a specific role in visual analysis. The initial layers process simple image features, such as edges and contrasts, while the intermediate and final layers combine this information to recognize more complex structures and identify objects within images. Essentially, these rat-inspired programs will, chronologically, see a block of cheese as a wireframe simulation before they recognize it as a viable meal. The SISSA researchers trained rats with a reward to recognize and discriminate objects under increasingly challenging and potentially confusing conditions. Objects were rotated, resized, or partially obscured to assess both the animals’ and the neural networks’ abilities to recognize them, despite the transformations. In simpler scenarios, such as changes in position, the neural network managed to replicate the rats’ accuracy using only half of the layers of analysis. However, as complexity increased, the rats maintained quite a high success rate in all tests, while the network needed increasingly programmed layers and more resources to compete, achieving comparable results only by using the entire depth of the convolutional architecture. In addition, the study found considerable differences in how the neural network and rats process visual information, despite the biological inspiration infused in the former. This suggests that obtaining a more complete understanding of the mechanisms by which rats and other mammals visually recognize objects is complex, and seemingly ambiguous settings could inspire improvements in AI models. And one potential result of research increasingly directed toward rat vision, could be one day finding an AI model scurrying on the ground as it tries to find a single scrap of laser-based cheddar. The research was published in Patterns (www.doi.org/10.1016/j.patter.2024.101149).