The Architecture of Friction: Active Inference, Pathological Attractors, and the Evolutionary Genesis of Agency

Modern cognitive science and user interface design increasingly favor frictionless environments. However, systems-level analysis via Active Inference reveals that the systematic elimination of cognitive friction leads not to optimal adaptation, but to systemic collapse. In this paper, I bridge the gap between clinical pathology—sharply contrasting passive depressive withdrawal (Πₐ → 0) with active obsessive pursuit/stalking as a hyper-active closed-loop cybernetic deadlock (Πₐ ≫ 0 at Πₛ → 0)—and the computational-architectural derivation of human agency. I propose that agency did not emerge via teleological design, but as a non-stationary thermodynamic filter of niche construction where systems capable of buffering temporal tension avoided Markov blanket dissipation. I formalize Cognitive Friction (R) as an ecological boundary-condition operator that dictates the availability of environmental feedback. I demonstrate how the collapse of this feedback (R → 0) forces the selective down-regulation of disconfirming sensory precision (Πₛ → 0) while internal priors dominate (Πᵤ ≫ Πₛ). Rather than arresting system dynamics, this failure triggers a catastrophic forced convergence to a degenerate false attractor and destructive overfitting over a finite relaxation time (θ(t) ⟶ θ*), locking adaptive structures into self-sustaining pathological attractors. Furthermore, I establish a rigorous empirical validation protocol, defining quantitative falsification thresholds in hierarchical agent-based modeling (ABM).