AI-Discovered Alternative State Representations in Physical Systems: A Methodological Control Case for SHRF/GASD

This entry records and contextualizes claims that an AI system has “independently discovered alternate physics.” The underlying research demonstrates that machine-learning models can infer compact, predictive latent state representations directly from raw observational data (such as video of physical systems) without being supplied with canonical physical variables or equations.The result does not constitute the discovery of new physical laws. Instead, it shows that multiple, non-unique coordinate representations can describe the same underlying dynamical system with comparable predictive power. These representations may differ substantially from human-chosen variables while remaining mathematically…