The Non-Markovian Ache: Thermodynamic Computing, Hidden Somatic Memories, and the Daemonic Phase Space of Artificial Intelligence

This conceptual interdisciplinary paper connects research on non-Markovian information engines and thermodynamic computing with accounts of somatic memory, archival traces, and artificial intelligence. It surveys viscoelastic baths, stochastic hardware, and non-equilibrium generative models, then develops an interpretive analogy between persistent physical memory and unresolved affective traces in language and model training. The closing sections discuss ethical design ideas including temporal titration, consent, and preserving room for disengagement. The analogies are conceptual arguments in the manuscript, not experimental evidence that AI systems possess human somatic states.