Despite its having been working so well in explaining many climate change phenomena, the diffusive representation of the moist static energy (MSE) flux in energy balance model (EBM) has yet to be justified. This study shows that the diffusive form of MSE transport can be derived from the column-integrated energy balance equation and can be quantified using a simple machine learning algorithm constrained to satisfy the energy balance. The result suggests that midlatitude baroclinic eddies are more efficient than the tropical overturning circulation in transporting MSE poleward. The machine-learned, seasonally varying diffusivity is then tested against the existing scaling theories for the midlatitude eddy diffusivity. The result happens to validate the moist version of the theory of Held and Larichev (1996), affording a physical interpretation for the seasonal variation of the diffusivity: the midlatitude diffusivity is furnished by mixing motions typified by a meridional mixing length of the Rhines scale and a velocity scale that is set by the requirement of the equipartitioning between the eddy kinetic and available energy. Accounting for an interactive diffusivity formulated following Held and Larichev (1996) in an EBM, it is found that the midlatitude diffusivity must weaken in response to the more isothermal atmospheric temperature under a uniform, positive climate change forcing. This implies an overall decrease of the second moment of the probability distribution of the midlatitude temperature variability, reducing the relative likelihood of the occurrence of temperature extremes.