Generative models of living development: an interaction-centered proposal and its implications for artificial life research

Artificial life and integrative systems neuroscience increasingly seek to explain how agency is organized through development, not merely how an already-formed controller optimizes a predefined task. This article advances an interaction-centered theory that links computational generative models, which infer hidden causes and candidate action policies, with experimental ecologies, which generate the developmental histories to be explained. The framework distinguishes reward and prior preference from concern, normativity, habit, and action policy. Concerns denote system-relative dimensions of regulatory significance grounded in an agent’s organization; habits are historically acquired dispositions; and action policies remain formally defined candidates within a computational model. The code-mediated concern hypothesis proposes that events acquire functional significance when adaptor-implemented mappings make cues informative about viability-linked concerns, thereby changing inference, precision, habit, and action selection. Teleonomic and organizational accounts locate the normative force of these concerns in present self-maintaining organization rather than conscious design or backward causation. The Layered Model of Agency differentiates operative, reflective, joint, material-mediated, sign-mediated, and symbolic agency, preventing evidence for low-level control from being generalized to collective, artifact-mediated, symbolic, or phenomenal agency. The Human-Rat Interaction Paradigm (HRIP) supplies an extensible experimental architecture for studying how structured human-rat interaction histories shape multidomain developmental trajectories. The inaugural male-rat study permits only sex-specific inference; the broader HRIP paradigm accommodates female, male, and mixed-sex samples in comparative and mechanistic research designs. Six propositions and eight predictions define tests of model comparison, cue remapping and adaptor perturbation, habit-policy dissociation, reciprocal coupling, material and sign mediation, multimodal trajectories, developmental and sex moderation, and layer-specific evidence. Together, these elements define a testable program for developmental artificial life while preserving the distinction between functional agency-like organization and phenomenal agency.