A Bayesian framework for happiness, health, and psychological well-being

Classical accounts of well-being have largely been grounded in homeostasis, conceptualizing well-being as the regulation of affective states around defended set points. More recent Bayesian and active inference approaches similarly characterize living systems as predictive regulators that maintain viable states by minimizing expected surprise. Building on these perspectives, this article proposes an integrative Bayesian framework in which well-being is understood not as a fixed scalar quantity, but as a dynamically maintained viability corridor defined by hierarchically organized prior expectations about preferred affective and interoceptive states. Classical accounts of well-being have largely been grounded in homeostasis, conceptualizing well-being as the regulation of affective states around defended set points. More recent Bayesian and active inference approaches similarly characterize living systems as predictive regulators that maintain viable states by minimizing expected surprise. Building on these perspectives, this article proposes an integrative Bayesian framework in which well-being is understood not as a fixed scalar quantity, but as a dynamically maintained viability corridor defined by hierarchically organized prior expectations about preferred affective and interoceptive states. Within this framework, homeostasis and allostasis are viewed as complementary aspects of the same regulatory architecture. Homeostasis refers to the maintenance of vital variables within viable bounds, whereas allostasis denotes the predictive processes through which such stability is achieved under changing conditions. Well-being is interpreted as the phenomenological expression of these regulatory dynamics, closely related to core affect while remaining analytically distinct from the mechanisms that sustain it. At the event level, affective experience is described in information-theoretic terms. Arousal is formalized as information gain arising from the interaction between prediction error and prior uncertainty, whereas valence follows an inverted-U relationship consistent with the Wundt curve. Drawing on the Central Limit Theorem, the framework further suggests that moderate and metabolically efficient regimes are both statistically prevalent and hedonically preferred, helping to explain why organisms tend to gravitate toward intermediate levels of stimulation. Health is reconceptualized as metastable attunement: a dynamic balance between stability and exploration sustained through flexible precision allocation. Within this perspective, valence functions as a control signal that guides precision tuning across timescales, enabling organisms to maintain viable states while adapting to changing environmental demands. Different dimensions of the good life can then be understood as expressions of this broader regulatory architecture. Happiness reflects the experience that regulation is proceeding better than expected at an acceptable energetic cost; eudaimonia emerges from the alignment of action with identity-level priors, values, and long-term goals; and psychological richness reflects epistemic exploration that expands the agent's generative model. Accordingly, mindfulness and psychotherapy are interpreted as complementary routes to adaptive regulation, promoting greater flexibility and integration across levels of the self-model. Situated within health neuroscience, the proposed framework provides a theoretically grounded basis for future research at the intersection of computational neuroscience, clinical psychology, and well-being science.