Active inference and the free energy principle model brains as hierarchical generative systems that infer the hidden causes of sensory data and act to minimize expected surprise. Over the past decade this formalism has moved from theoretical neuroscience into psychological science and clinical psychiatry, producing a growing literature on emotion, interoception, selfhood, social cognition, psychotherapy, and psychopathology. The literature remains methodologically young: most models are still at the level of simulation and theory, with relatively few large-scale empirical tests and only nascent translational tools.
Why the domain fits
Psychology is fundamentally about how organisms manage uncertainty: interpreting ambiguous sensory signals, forming beliefs, regulating bodies and emotions, and selecting actions in partially observable situations. Active inference treats these functions as instances of probabilistic inference under a generative model, where perception infers hidden states, learning updates model parameters, and action samples preferred observations by minimizing expected free energy. This gives psychological constructs long described qualitatively—schemas, appraisals, defense mechanisms, coping strategies—an explicit mathematical treatment, for example modeling anxiously biased threat appraisals as priors that overweight danger and drive hypervigilant behavior.
State of the literature
A 2024 Entropy review, "Active Inference in Psychology and Psychiatry: Progress to Date?," surveys applications across emotion, interoception, selfhood, social cognition, and psychopathology, arguing the framework could unify these subfields while stressing the need for rigorous empirical tests. Lisa Feldman Barrett's theory of constructed emotion recasts emotion as inference over interoceptive and exteroceptive inputs using stored concepts; related work extends this to interoceptive psychopathology, anxiety as "learned uncertainty," allostatic accounts of depression, addiction as optimal inference under a suboptimal model, and negative symptoms of schizophrenia as failures to predict the consequences of one's own actions. Psychotherapy itself has been formalized computationally, with active inference models of CBT and of Coherence Therapy simulating how interventions revise priors and policies.
Key projects and tools
The main computational backbone cited is work on active inference in discrete state spaces ("Active inference on discrete state-spaces: A synthesis"), which supplies algorithms for categorical generative models, policy evaluation, and belief updating that underlie later computational-psychiatry applications. The report also notes MATLAB routines associated with Karl Friston's group at UCL and Python-based packages implementing discrete active inference, used to fit models to behavioral task data and infer individual differences in priors and precision. Task-specific code exists for CBT simulation models with adjustable patient-specific priors and preferences, though the Coherence Therapy model remains largely theoretical rather than implemented software; the report explicitly notes that fully deployed clinical systems based on active inference are still rare.
Open problems
The report flags a core tension between active inference's normative claim (agents minimize free energy) and the often suboptimal, culturally and situationally shaped variability of real human behavior, compounded by model identifiability risk given high-dimensional priors, precision terms, and preferences. Empirical validation across levels of description is thin: most models are tested on simple tasks with limited ecological validity, and claims about self, identity, and personality are largely conceptual rather than data-driven. The report calls for longitudinal, multi-modal studies linking neural, physiological, behavioral, and self-report data to model parameters, comparisons against alternative frameworks like reinforcement learning, and controlled trials of active-inference-informed interventions before clinical translation—including attention to interpretability, individual and cultural differences, and the ethics of model-based behavior prediction.
Reference Backbone
Karl J. Friston (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience. DOI: 10.1038/nrn2787. Thomas Parr, Giovanni Pezzulo, Karl J. Friston (2022). Active Inference: The Free Energy Principle in Mind, Brain, and Behavior. MIT Press. Giovanni Pezzulo, Francesco Rigoli, Karl J. Friston (2017). Active Inference, homeostatic regulation and adaptive behavioural control. Progress in Neurobiology. DOI: 10.1016/j.pneurobio.2017.08.001. Maxwell J. D. Ramstead, Karl J. Friston, Axel Constant, Lancelot Da Costa, Casper Hesp, Beren Millidge, Alexander Tschantz (2023). On Bayesian Mechanics: A Physics of and by Beliefs. arXiv. Lancelot Da Costa, Thomas Parr, Noor Sajid, Sebastijan Veselic, Victorita Neacsu, Karl J. Friston (2020). Active inference on discrete state-spaces: A synthesis. Journal of Mathematical Psychology. DOI: 10.1016/j.jmp.2020.102447.