A Proposal Extending the Free Energy Principle

Abstract

A survey of AI-related dreams (Mayer, 2025) reports that 93% of individuals who experienced AI-associated nightmares fixated on the interaction interface itself rather than on narrative content. We propose an Interaction Architecture Internalization Model (IAIM) to account for this finding and for a broader class of phenomena spanning language acquisition, cultural meme propagation, and personality formation. The model is grounded in, and extends, the Free Energy Principle (FEP; Friston et al., 2006): while the FEP's perceptual-learning pathway (optimisation of slow synaptic parameters μₕ) already encodes structural regularities of the environment, we argue that temporal delay is the specific, hitherto under-exploited statistical feature that drives internalisation of interaction architecture. We formalise this via the Learning Time Delay Dose D = Σᵢ[wᵢ(Δtᵢ - Δt꜀)] and show that once D crosses an individual-specific critical threshold Θ꜀, a precision-weighted phase transition occurs in which the external interaction architecture becomes a stable prior in the agent's generative model. A key empirical implication—the Formal Priority Law (FPL)—is that off-line cognition reproduces the formal-structural elements of a repeatedly experienced architecture preferentially over its narrative-semantic content. This prediction is now supported by six independent sources, including the first polysomnographically verified, real-time-signalled lucid-dreaming study of a VR interaction architecture (Morris et al., 2025). The proposal draws on Piaget's constructivist internalisation, Chomsky's architectural nativism, Wiener's cybernetics, and Landau-Hale phase-transition mathematics, all interpreted within the FEP variational framework.