The Physics of Relevance: Defining and Measuring Intelligence at the Interface Plane

A persistent obstacle in the science and engineering of intelligence is the absence of a definition and measure that generalize across biological and artificial systems. We model intelligence on an intermediate agent-environment interface plane where uncertainty is filtered by a learned relevance partition. By unifying thermodynamic and Shannon entropy into a generalized functional, we establish a formal definition of intelligence as the viability-constrained, physically realized regulation of predictive uncertainty over these relevant states. We introduce an adaptive structure (RUINES) that implements this partition, and provide operational metrics (kinetic intelligence IK , and intelligence potential IP) alongside falsifiable prediction criteria for artificial and biological systems.