From Regulatory Dependence to Other-Valuing: A Computational Investigation of Evolutionary Value Theory

Evolutionary Value Theory (EVT) distinguishes the capacity to model another system's regulatory dynamics from recognizing its independently maintained interests and assigning those interests motivational significance. We investigated these distinctions in a sequence of synthetic computational experiments, EVT 7.1a–z, focusing here on the archived, inspectable pilots 7.1r–x and 7.1z, with 7.1y providing a theoretical audit rather than a numerical experiment. Agents learned dynamical representations, intervention predictions, and actor-centered regulatory rewards under different coupling, decoupling, conflict, and updating conditions. A supervised diagnostic probe applied to a system-identification representation distinguished identifiable simulated recipients from objects, but failed on observationally exact mimics. A later reward-learning pilot showed selective intervention, but its reward was engineered from a hidden adaptation parameter. Replacing that proxy with simulated endogenous coupling produced extensive object interventions. Frozen policies could retain assistance despite costs they did not actually observe, whereas online self-reward learning tended to extinguish costly assistance after decoupling. An adversarial consolidation pilot acquired identical commitments to a recipient and a matched object. Taken together, the experiments illustrate the difference between prediction, instrumental helping, motivational persistence, and independently other-directed valuation. They do not establish the spontaneous acquisition of independent other-valuing, nor a computational impossibility theorem. The results inform the capacity architecture of EVT while leaving the origins of motivationally autonomous concern unresolved.