We propose a thermodynamic reframing of the empirical phenomenon of AI-driven labor displacement. Existing models of automation exposure (Acemoglu and Restrepo 2020; Eloundou et al. 2023) explain which occupations are technically substitutable, but leave residual variation in the speed of substitution. We hypothesise that occupations characterised by high human metabolic entropy production per unit time are differentially preserved under AI-driven substitution; the temporal order of displacement is inversely related to the human dissipation rate of the occupation. Equivalently, surviving occupations are those for which the entropy-production gap between human worker and AI substitute is small or negative, so that the marginal entropy gain from substitution is too small to drive MEPP-consistent selection. Building on the Maximum Entropy Production Principle (Kleidon 2010; Martyushev and Seleznev 2006) and on Jeremy England's dissipative-adaptation programme (England 2013, 2015), we develop the framework, propose an operational definition based on metabolic-equivalent ratings (O*NET / Ainsworth Compendium), sketch a regression strategy using BLS occupational data for 2019–2025, and illustrate the framework with a case-study comparison of call-center work and skilled trades. We then survey five strands of existing empirical evidence — the routinisability gradient (Autor, Levy and Murnane 2003), capability-based AI exposure measures (Frey and Osborne 2017; Eloundou et al. 2023), recent generative-AI experiments (Noy and Zhang 2023; Brynjolfsson, Li and Raymond 2023), BLS occupational data, and the energy-economy linkage (Garrett 2014) — all consistent with H4. The cosmological proposals of Smolin (1992) and Crane (1994) inspire but do not constrain the empirical claim. We conclude with policy implications for redistribution and for the regulation of AI energy consumption.
Semin UM (Sun,) studied this question.