Researchers propose a new approach to machine intelligence that focuses on the 'artificial physical world', where objects are intentionally designed and documented. They argue that this domain is different from the symbolic world, which has been conquered by AI. The authors present four contributions: a legitimacy criterion for constitutive prior frameworks, a layering lower bound, deployment claims across five industrial domains, and falsifiable predictions. The framework aims to address the 'cold-start deadlock' in physical AI, where no intelligence exists without data, and no data exists without deployed intelligence. --- Why it matters: This research matters because it attempts to bridge the gap between symbolic and physical AI, which has been a long-standing challenge. Understanding how to apply machine learning to real-world objects and infrastructure could have significant implications for fields like robotics, computer vision, and autonomous systems. Source: https://arxiv.org/abs/2608.15147