In this work, I introduce R-Law AI, an artificial intelligence framework grounded on a thermodynamic–informational principle that I call the Principle of Informed Organizational Efficiency (IOE). I provide the mathematical formulation of the framework, discuss its theoretical implications, and present illustrative experiments on simple datasets (such as Iris) showing that models trained under R-Law dynamics tend to exhibit smoother convergence, reduced internal entropy, and more stable parameter evolution. I argue that this IOE-based perspective opens a path toward physically grounded, self-organizing machine learning systems.