arXiv Paper Proposes Legitimacy Theory for Physical AI via Constitutive Priors

A recent study published on arXiv (2608.15147) contends that physical AI encounters a fundamental cold-start dilemma—intelligence cannot exist without data, and data cannot be generated without existing intelligence. However, the authors note that this dilemma is not uniformly experienced. They suggest that the artificial physical environment, encompassing buildings, industrial sites, and infrastructure, serves as an exception due to its intentional design and documentation, which includes accessible archives that precede their creation. The authors propose a legitimacy standard for these foundational frameworks: prior extraction is valid if the domain is deliberately structured and has an accessible archive, which can be evaluated through direction of fit. The paper outlines four key contributions, including deriving this criterion from a four-world ontology, establishing a theoretical base for machine intelligence in physical contexts, with significant implications for AI applications in architecture, industry, and infrastructure.