Abstract
Abstract concepts are systematically associated with body-centered space, yet it remains unclear how they are represented across the body itself. Here, we used Body Mapping to characterize the bodily organization of four conceptual dimensions—time, length, quantity, and physical size—in humans and generative artificial intelligence (AI). In Experiment 1, human participants reported significant increases in bodily activity for “future”, “long”, “more”, and “large”, whereas no significant activation or deactivation survived correction for their conceptual opposites. “Future”, “more”, and “large” were primarily associated with the head and central torso, whereas long showed a distinct limb-centered pattern. In Experiment 2, six image-generation models were tested; five produced sufficient valid samples for the main comparison and yielded statistically reliable body maps whose organization differed markedly from the human pattern and across model families. GPT Image models showed either strongly lateralized or widespread activation patterns, whereas Nano Banana models more clearly expressed conceptual opposition through complementary activation and deactivation. These findings show that bodily organization of abstract concepts cannot be inferred directly from conventional spatial compatibility effects and that sensitivity to conceptual polarity does not imply human-like bodily organization. More broadly, Body Mapping provides a common representational framework for comparing conceptual organization across human and artificial systems, opening a route from behavioral comparison toward representation-level comparison—and potentially training—using human-derived bodily data.