Neural networks are rapidly becoming ubiquitous, and they seemingly can do anything. But what does it mean to teach a machine to see? Is there any guarantee that machine perception aligns with human perception? These synthetic images show the average representation of 25 object classes as imagined by a computer vision model trained for object classification. From the delicate intricacies of a bonsai tree to the striking geometry of a stop sign, the visualizations provide hints into perceptual details the model associates with the object. By anchoring to the model’s internal understanding of natural objects, my method seeks to overcome a common limitation of earlier techniques, which often produced unrealistic or abstract visualizations. The depicted images are not merely artistic—they are tools of discovery, translating the mind of the machine into something we can comprehend, interpret, and ultimately trust.