Stefano Coelati Rama

Why do living systems represent only a sliver of their environment? This paper argues that the answer is thermodynamic: because information is physical, every bit a system acquires and maintains is paid for in free energy that could otherwise buy persistence. It introduces the Semantic Free Energy of the Expected Future (S-FEEF), a criterion that selects an organism's perceptual policy by trading the metabolic cost of information against viability. Building on Kolchinsky and Wolpert's account of semantic information, the paper refounds viability as the non-equilibrium free energy of the agent's state, its Kullback–Leibler distance from the Boltzmann ("death") equilibrium. This single move eliminates a free parameter, places cost and benefit in the same physical units, gives the metabolic exchange rate a Landauer floor, and lets the criterion be derived from Millidge et al.'s Free Energy of the Future under four explicitly stated approximations. An exhaustive computational model yields four results, each stated with its limits: the optimal agent compresses below the point of informational saturation ("less is more"); more information can reduce viability, but only for agents with bounded perception–action coupling; the "dark room" is a sharp, computable threshold rather than a paradox; and morphology non-monotonically gates the value of a mind. The same marginal-return logic extends to evolutionary life history (semelparity, the emergence of senescence, and bet-hedging) and to a thermodynamic account of where the perceptual filter comes from in the first place. The framework is a deliberately small, fully reproducible toy model: it demonstrates mechanisms rather than universal laws, and proposes (though it does not yet run) a concrete empirical test.