Active Inference and Psychoanalysis

The intersection between the Free Energy Principle (FEP), active inference, and psychoanalysis has become a new domain of research, with some speculate that active inference may offer psychoanalytic theory new conceptual tools drawn from neuroscience to revitalize the field that has struggled for institutional legitimacy in many decades. As we have mentioned in our previous posts the FEP provides a formal framework describing how living organisms maintain their structural integrity by minimizing free energy, which represents prediction error within a probabilistic model. Active inference, which extends from the FEP, proposes that brains function as active agents that constantly generate predictions about their environment rather than passively processing sensory input. This framework not only models biological and psychological processes but attempts to formalize them in ways that might compare to psychoanalytic concepts. However, psychoanalysis operates within an interpretive tradition rooted in personal narratives, clinical encounter, and symbolic meaning, while active inference is computational and mechanistic, formulated within a Bayesian framework. Freud’s early Project for a Scientific Psychology (1895) represented an ambitious, but ultimately abandoned, attempt to ground psychology in the neurophysiology of the the time. Despite its historical limitations, the Project anticipated several ideas that active inference and the FEP now formalize, including concepts of energy flow, psychic conflict, and the nervous system’s drive toward equilibrium. Freud’s notion of the mind as a dynamic energy system, influenced by 19th-century thermodynamics, is somehow similar to the FEP’s characterization of the brain as a system that minimizes uncertainty. In Freud’s model, concepts like binding (Bindung) and unbinding (Entbindung) describes how the nervous system processes excitation—which appears to parallel the FEP’s distinction between prediction error (unbound energy) and coherent generative models (bound energy). Both frameworks suggest the mind strives toward a state of minimal excitation or surprise—a kind of homeostatic equilibrium. His later distinction between primary (id-driven, wish-fulfilling) and secondary (ego-driven, reality-oriented) processes finds potential resonance in the FEP’s hierarchical generative models. In active inference, lower hierarchical levels process immediate sensory prediction errors, while higher levels encode abstract, temporally extended representations that shape both perception and action. However, this analogy is fundamentally metaphorical, that is (as we mentioned in several previoous posts) FEP offers a formal, information-theoretic framework grounded in variational Bayesian inference, while Freud’s psychodynamics rests on speculative neurophysiology and clinical observation. The two operate on fundamentally different epistemological planes, and conflating them risks obscuring rather than clarifying either framework.

Desire is another point of potential conceptual alignment, Freud’s theory of desire is based on tension between instinctual drives and the regulatory functions of the ego and superego, where repression keeps unacceptable desires unconscious, yet these desires persist, shaping symptoms, dreams, and behaviors. For Freud, conflict is the engine of psychic life, and its resolution remains perpetually incomplete. Active inference provides what might be considered a computational analog to this dynamic. Desire can be modeled as a prior over preferred states—probabilistic expectations about outcomes the agent seeks to attain (see the post about FEP and emotional valence). These priors influence perception, cognition, and action through policy selection, shaping how the agent engages with the world. As Kruglanski et al. note in their 2020 paper synthesizing epistemic motivation with active inference, “all thinking is wishful thinking” in a predictive brain—prediction is inherently motivated by expected outcomes. Hoewever, it is important to note that active inference see desire as precision-weighted prediction over policies, but psychoanalysis treats desire as symbolic, relational, and historically embedded. To link these two concepts directly would risk reductionism where in the computational framework it is possible to model drive and conflict in abstract, formal terms, but it struggles to capture the dense symbolic texture of psychic life—the meanings embedded in dreams, fantasies, and transference relationships.

Psychoanalysis regards dreams as their “royal road to the unconscious,” that reveals disguised wishes, unresolved conflicts, and repressed material. For Freud, dreams serve not only an expressive function but also a defensive one they transform unacceptable desires into symbolic narratives through dream work mechanisms such as condensation, displacement, and secondary revision. From an active inference perspective, dreaming may serve a different function. During REM sleep, the brain appears to relax sensory constraints and engage in generative modeling, simulating possible scenarios and integrating emotionally significant experiences. This “offline inference” may allow the brain to explore hypothetical states to refine its predictive models and minimize future surprise. Research suggests that REM sleep facilitates pattern extraction from past experiences and enables the consolidation of implicit, emotionally salient memories. This account has some similariy to Freud’s idea of dreams as a site of psychological work. Both perspectives put dreaming as a process that metabolizes affect, rehearses responses, and processes emotionally loaded material. However, the underlying logic is different between two frameworks, Freud emphasizes symbolic transformation and the censorship imposed by psychic defenses, while active inference treats dreams as structurally constrained simulations aimed at optimizing generative models. Both models describe adaptation to internal conflict and emotional intensity, but they explain this adaptation through fundamentally different mechanisms—one prioritizing symbolic interpretation, the other formal simulation.

Despite some suggestive parallels, there are some methodological and epistemological differences that complicate any integration of psychoanalysis and active inference such as Different Scientific Paradigms, psychoanalysis belongs to the hermeneutic tradition, concerned with meaning-making and narrative coherence. Active inference is embedded in formal mechanistic science, concerned with algorithmic explanation and testable prediction. These frameworks operate within what philosophers like Kuhn might call incommensurable paradigms of inquiry. Conceptual Ambiguity: Terms like “energy,” “conflict,” or “desire” carry different connotations in each system. In psychoanalysis, energy is metaphorical and affective; in the FEP, free energy is a precise mathematical quantity derived from information theory. Overextending these analogies risks conceptual confusion rather than clarification. Empirical Generalizability: While active inference generates somewhat testable predictions (hopefully we do a post about falsifiability of active inferences in near future) in experimental paradigms and psychiatric modeling, psychoanalytic constructs have historically resisted quantification. Clinical data from psychoanalysis is rich and nuanced but not easily formalized within computational frameworks. Critics of psychoanalysis often cite its lack of falsifiability—though this criticism has been challenged as both philosophically flawed and empirically outdated. Critics of the FEP, meanwhile, warn that it risks becoming a “theory of everything” that explains too much while predicting too little. Any integration must avoid collapsing into nonsense (cf. early Wittgenstein) either through naive formalism or interpretive excess.

The dialogue between psychoanalysis and active inference is rich but has conceptual risks. The FEP and active inference offer formal, computational lenses through which to reinterpret some of Freud’s insights that is the mind as a system under tension negotiating between competing demands to maintain internal coherence and adapt to its environment. However, this interpretive pairing of two frameworks needs epistemological care. Active inference provides elegant formalisms, but these cannot substitute for the narrative, symbolic, and relational dimensions that psychoanalysis brings to understanding human experience. The analogy between psychic energy and free energy proves fruitful only when treated as metaphor rather than equation. True integration, if it proves possible at all, lies not in collapsing one framework into the other but in allowing each to illuminate the other’s blind spots. Psychoanalysis can challenge active inference to account more adequately for affect, narrative, and transference. Active inference can offer psychoanalysis new tools for modeling conflict, adaptation, and learning. By embracing both the synergies and the tensions, we may develop a richer, more pluralistic understanding of the psyche—one that honors both its symbolic complexity and its computational structure.

References

Anderson, M. C., Ochsner, K. N., Kuhl, B., Cooper, J., Robertson, E., Gabrieli, S. W., et al. (2004). Neural systems underlying the suppression of unwanted memories. Science, 303, 232.

Carhart-Harris, R. L., & Friston, K. J. (2010). The default-mode, ego-functions and free-energy: a neurobiological account of Freudian ideas. Brain, 133, 1265–1283.

Cittern, D., Nolte, T., Friston, K., & Edalat, A. (2018). Intrinsic and extrinsic motivators of attachment under active inference. PLoS ONE, 13(4), e0193955.

Gray, J. R., Bargh, J. A., & Morsella, E. (2013). Neural correlates of the essence of conscious conflict: fMRI of sustaining incompatible intentions. Experimental Brain Research, 229(3), 453–465.

Connolly, P. (2018). Expected free energy formalizes conflict underlying defense in Freudian psychoanalysis. Frontiers in Psychology, 9, 1264.

Freud, S. (1895). A project for a scientific psychology. In The Standard Edition of the Complete Psychological Works of Sigmund Freud (Vol. 1, pp. 283–397). London: Hogarth Press.

Freud, S. (1900). L’Interprétation des Rêves (2010e éd.). Paris: PUF.

Freud, S. (1911). Formulations on the two principles of mental functioning. In The Standard Edition of the Complete Psychological Works of Sigmund Freud (Vol. 12, pp. 218–286).

Hopkins, J. (2016). Free energy and virtual reality in neuroscience and psychoanalysis: a complexity theory of dreaming and mental disorder. Frontiers in Psychology, 7, 922.

Holmes, J., & Nolte, T. (2019). “Surprise” and the Bayesian brain: Implications for psychotherapy theory and practice. Frontiers in Psychology, 10, 592.

Kapur, S. (2003). Psychosis as a state of aberrant salience: a framework. Brain, 126, 13–27.

Parr, T., Pezzulo, G., & Friston, K. J. (2022). Active inference: The free energy principle in mind, brain, and behavior. Cambridge, MA: MIT Press.

Scarfone, D. (2018). Free association, surprise, trauma, and transference. Psychoanalytic Inquiry, 38, 468–477.

Shedler, J. (2010). The efficacy of psychodynamic psychotherapy. American Psychologist, 65, 98–109.

Solms, M. (2019). The hard problem of consciousness and the free energy principle. Frontiers in Psychology, 9, 2714.

Zepf, S. (2010). Libido and psychic energy – Freud’s concepts reconsidered. International Forum of Psychoanalysis, 19, 3–14.

Beren, J. (2024, July 27). A retrospective on active inference. https://www.beren.io/2024-07-27-A-Retrospective-on-Active-Inference/

Bruineberg, J., Kiverstein, J., & Rietveld, E. (2016) The anticipating brain is not a scientist: The free-energy principle from an ecological-enactive perspective. Frontiers in Human Neuroscience, 10, Article 20. https://doi.org/10.3389/fnhum.2016.00020

Clark, A. (2020). Beyond the “Bayesian brain”: Predictive processing and the extended mind. Trends in Cognitive Sciences, 24(7), 517–529. https://doi.org/10.1016/j.tics.2020.03.010

Colombo, M., & Wright, C. (2023). First-person methods in computational psychiatry and cognitive neuroscience. Physics of Life Reviews, 47, 1–30. https://doi.org/10.1016/j.plrev.2023.09.003

Freud, S. (1950/1895). Project for a scientific psychology. Pennsylvania State University. https://bpb-us-e1.wpmucdn.com/sites.psu.edu/dist/9/19778/files/2021/12/freud-project_for_scientific_psychology.pdf

Hohwy, J. (2013). The predictive mind. Oxford University Press.

LessWrong. (2019). Why I’m not into the free energy principle. https://www.lesswrong.com/posts/MArdnet7pwgALaeKs/why-i-m-not-into-the-free-energy-principle

Parr, T., Pezzulo, G., & Friston, K. (2022). Active inference: The free energy principle in mind, brain, and behavior. MIT Press. https://direct.mit.edu/books/oa-monograph/5299/Active-InferenceThe-Free-Energy-Principle-in-Mind

PsychStudies. (2018). Is psychoanalytic psychotherapy empirically validated? https://www.psychstudies.net/is-psychoanalytic-psychotherapy-empirically-validated/

Smith, R., Badcock, P., & Friston, K. (2022). Recent advances in the application of the free-energy principle to psychiatry. Current Opinion in Neurobiology, 76, 102602. https://pmc.ncbi.nlm.nih.gov/articles/PMC9260223/

Williams, D. (2022). Predictive processing and mental disorder. Philosophy, Psychiatry, & Psychology, 29(4), 267–290. https://pmc.ncbi.nlm.nih.gov/articles/PMC9731232/