Maladaptive Daydreaming (MD) involves intense, compulsive immersion in internally generated fantasies. Clinicians assess it with the 16-item Maladaptive Daydreaming Scale (MDS-16), yet the computational mechanisms that sustain episodes remain poorly understood. This paper proposes a theoretical model that links Reinforcement Learning (RL) and Active Inference through a precision-weighting dynamic with nonlinear feedback. We argue that MD arises when a pathological precision ratio (ρ ≫ 1) causes internal priors to dominate external sensory evidence. Dynamical analysis then shows how saddle-node bifurcations and hysteresis can trap the system in a high-ρ state, which helps explain why patients find it hard to stop daydreaming voluntarily. We also suggest that stereotypical movements such as pacing act as predictive sensory gating: they suppress external prediction errors and maintain the cortical arousal needed for vivid fantasy. Finally, we extend the same precision-imbalance equation to other behavioral addictions. We propose that MD serves as an endogenous prototype, whereas Internet Gaming Disorder (IGD) and AI-companion dependence represent technology-mediated extensions of the same free-energy minimization strategy. The framework generates testable predictions that can be evaluated with Hierarchical Gaussian Filtering and computational neuroimaging. We discuss its limitations and outline directions for empirical validation.
Bo Lee (Sun,) studied this question.