Perspective Latents as an Architectural Condition for Causal Emergence in Active Inference Agents

A new study tests causal emergence in an active inference agent whose architecture separates a fast perception latent z from a slow global latent g, finding that the integrated information measure Φr concentrates in g and is largely architectural, decreasing with training. The substantive learning effect appears only at the atom-compositional level, where decoupling flips sign from negative to positive and becomes regime-invariant, while downward causation carries regime-dependent adjustment. The results identify g as the architectural locus of Φr-relevant temporal organization and argue against reading scalar Φr as a direct index of learned integration.

arXiv:2607.20708v1 Announce Type: new Abstract: A recent line of work measures causal emergence in reinforcement learning agents through Integrated Information Decomposition, reporting that $\Phi r$ grows with training and tracks reward improvement. For active inference, this raises the question of how reward-free predictive organization relates to such information-theoretic signatures. I test this within an active inference agent whose architecture separates a fast perception latent $z$ from a slow global latent $g$, where $g$ is driven by prediction error and structurally decoupled from policy gradients. In a reward-free environmental regime-switching protocol, $\Phi r$ concentrates in $g$; its aggregate magnitude is largely architectural and decreases with training. The substantive effect of learning becomes legible only at the atom-compositional level: decoupling flips sign from negative to positive and becomes regime-invariant under environmental change, while downward causation carries the regime-dependent adjustment. These results identify $g$ as the architectural locus of $\Phi r$-relevant temporal organization in an active inference agent, and argue against reading scalar $\Phi r$ as a direct index of learned integration.