Dmn_fep

The Default Mode Network and Psychoanalytic Concepts in Active Inference

Introduction

As we discussed in the last post, Freud’s ambition was never purely psychological. From the start, he wanted to anchor the mind in biology. His Project for a Scientific Psychology (1895) sketched neurons, energy flows, and thresholds—a remarkable intuition that outpaced the tools of his era. For most of the 20th century, that project lay dormant. Then came neuroimaging, computational psychiatry, and Karl Friston’s Free Energy Principle (FEP)—and the distance between Freud’s couch and the brain scanner felt shorter. This post highlights can the Default Mode Network (DMN)—a large-scale brain network at the heart of self-referential thought—serve as a neurobiological scaffold for psychoanalytic constructs like the ego, the id, and repression? And can active inference, the behavioral complement of the FEP, give us a computational language precise enough to do justice to both traditions?

In our series of posts, we have discussed the Free Energy Principle, as articulated by Friston (2010) that proposes that all biological systems—from single cells to whole brains—resist disorder by minimizing variational free energy: a measure of the gap between what the organism predicts about its sensory environment and what it actually receives. In plain terms, the brain is a prediction machine that is perpetually trying to reduce surprise. Active inference is how this plays out in behavior. Rather than passively waiting for the world to confirm or disconfirm its models, the organism acts to make the world match its predictions—or updates its internal models when action fails. The key regulatory variable is precision: the confidence assigned to a prediction, which determines how strongly incoming sensory evidence can revise a prior belief. This turns out to be an unexpectedly powerful concept for psychiatry. High precision on prior beliefs, with low weighting of new input, maps neatly onto rigidity—the computational signature of conditions like OCD or trauma-related dissociation. What makes the FEP attractive to psychoanalysts is its formal consonance with Freud’s core claims: that the mind is governed by an economy of bound and unbound energy, that unconscious processes shape behavior by constraining what reaches awareness, and that conflict—between drives and constraints—is the engine of psychopathology. As Solms (2022) and colleagues in the neuropsychoanalytic tradition have argued, free energy can be read as a reformulation of what Freud called psychic energy—”unbound” free energy is aversive and functions like mental pain, while its binding is experienced as relief or pleasure.

The DMN and the Ego

The landmark paper by Carhart-Harris and Friston (2010) placed the DMN at the center of a neuropsychoanalytic model, arguing that its hierarchical organization mirrors the ego’s function, which is integrating internal drives with social and environmental constraints. This remains one of the most cited and debated proposals in neuropsychoanalysis. The DMN is a large-scale network anchored in the medial prefrontal cortex (MPFC), posterior cingulate cortex (PCC), and inferior parietal lobule (IPL). It activates during self-referential thought, autobiographical recall, theory of mind, and future simulation—precisely the capacities that psychoanalysis attributes to the ego. Crucially, MPFC-PCC functional connectivity is absent in neonates and consolidates over early development, tracking the emergence of a coherent self-concept in a way that invites comparison to the ego’s ontogenesis in psychoanalytic theory. More recent work by Yeshurun, Nguyen, and Hasson (2021) reframes the DMN not just as an “intrinsic” system for daydreaming, but as an active sense-making network that integrates incoming information with prior intrinsic models to construct meaning over time. This is an upgrade to the earlier model where the DMN is not just where the ego rests—it is where the ego works, continuously modeling the self in relation to others and to the flow of lived experience. Repression, on this account, is not mystical suppression but a computationally tractable process—the downregulation of precision assigned to signals that would disrupt the model’s coherence. Defense mechanisms, broadly, become strategies for minimizing free energy by selectively constraining what information propagates up the cortical hierarchy.

The Id, Primary and Secondary Process

The id, in Freud’s topography, is the original psychic agency: pre-verbal, affect-laden, indifferent to logic or time. Neurobiologically, this maps onto subcortical structures—particularly the amygdala, hypothalamus, and brainstem nuclei—that generate affective states prior to cortical elaboration. In FEP terms, these structures operate with low-precision priors: they generate urgent, high-entropy signals that demand behavioral response without the contextual filtering that higher cortical regions provide. The clinical evidence for this architecture is compelling. States in which top-down DMN regulation fails such as acute psychosis, temporal lobe seizures, REM dreaming, and high-dose psychedelic experiences—are all characterized by the emergence of primary process material: loose associations, hallucinatory imagery, and affective flooding. These are precisely the states that psychoanalysis describes as a collapse of ego boundaries, where the id’s unconstrained energy breaks through. The phenomenological convergence between these neurologically distinct states is not trivial—it suggests a shared mechanism of DMN disengagement. Solms (2013) makes the argument argument that consciousness itself originates in these subcortical affective systems—that the id, rather than being purely unconscious, is the seat of core subjectivity (cf. blog posts on FEP and consciousness). This inverts Freud’s original topography in a productive way, suggesting that the DMN does not generate consciousness but structures it: shaping raw affective experience into the narrative self we recognize as “I.”

Freud’s distinction between primary process (pleasure-seeking, illogical, timeless) and secondary process (reality-bound, sequential, rational) anticipates the computational distinction between fast, automatic processing and slow, deliberate reasoning that cognitive neuroscience has since formalized (cf. Fast and slow thinking). Within the FEP framework, this maps onto the balance between prior-dominated and likelihood-dominated inference. When priors dominate—when the brain’s internal model overrides incoming evidence—we get primary process-like states: hallucinations, wish fulfillment, confabulation. When sensory precision is high and prior beliefs are updated freely, we get secondary process cognition: evidence-based reasoning, flexible behavior, reality testing. Solms (2013) locates the DMN at the secondary process pole, supporting spontaneous, unconstrained thought, while the Central Executive Network (CEN) drives goal-directed, effortful cognition. The therapeutic implications are real. As we mentioned in our post on PTSD traumatic memories behave like overprecise priors, they override present sensory information, dragging the person back into a past that the body treats as now. Trauma creates, in FEP terms, an overwhelming influx of free energy for which no adequate top-down model exists—leading to defensive operations (dissociation, numbing) that protect the generative model at the cost of impoverished contact with experience.

Interoception, Affect, and Unconscious Conflict

Interoception—the brain’s continuous modeling of the body’s internal physiological state—has emerged as a critical interface between psychoanalytic and neuroscientific accounts of emotion. The predictive brain does not merely model the external world; it maintains ongoing predictions about the body itself, and mismatches between predicted and actual interoceptive states generate the affective signals that drive behavior. Hopkins (2012) argues that interoceptive predictions directly underlie Freud’s account of anxiety: what Freud called “signal anxiety”—the ego’s anticipatory response to potential danger—can be understood as the brain generating a prediction error about its own physiological state before a threatening situation fully unfolds. This is not a metaphorical restatement. It is a mechanistic claim: that anxiety is the phenomenal experience of unresolved interoceptive prediction error propagating through the cortical hierarchy. Depression offers another instructive case. Studies consistently show hypoconnectivity between the DMN and other large-scale networks in depression, alongside abnormal self-referential rumination—a state of rigid, low-precision updating in which negative self-models become entrenched and resistant to disconfirmation. This is the computational portrait of what clinicians describe as a loss of the capacity for new experience: the generative model has collapsed inward, treating its own predictions as facts.

Mapping Psychoanalytic Constructs to Neural and Computational Models

Psychoanalytic Construct Neurobiological Correlate FEP/Active Inference Interpretation
Ego DMN (MPFC, PCC, IPL) Hierarchical generative model regulating precision and prediction error
Id Limbic/subcortical (amygdala, hypothalamus, brainstem) High-entropy affective system generating low-precision, urgent priors
Superego Social priors encoded in DMN/CEN High-level constraints on belief updating, socially scaffolded
Repression DMN modulation of precision Downregulation of conflicting sensory evidence to minimize free energy
Primary Process Limbic-dominated, DMN-disengaged states Fast, affect-driven processing; prior-dominated inference
Secondary Process CEN-supported executive function Deliberative cognition; high sensory precision, flexible model updating
Anxiety Interoceptive prediction error Unresolved mismatch between predicted and actual physiological state
Trauma/PTSD Rigid prior dominance, DMN-CEN dysregulation Overprecise priors overriding present sensory evidence

Conclusion and Future Directions

Although this synthesis and rapprochement seems intellectually tempting but scientific honesty requires a clear view of limitations, for instance Conceptual mapping is not causal explanation and the fact that the DMN activates during self-referential thought, and that the ego is involved in self-regulation, is suggestive but not sufficient. Neural correlates are not neural mechanisms, and the Freudian constructs are complex enough—and slippery enough—that they can be mapped onto almost any sufficiently complex neural system. The mappings in the table above should be read as hypotheses, not established facts. Moreover, empirical validation remains thin most of the proposed correspondences lack direct experimental tests. Neuroimaging studies show correlations between DMN activity and self-referential processing; they do not demonstrate that the DMN implements repression in any mechanistically specific sense. The gap between computational metaphor and neurobiological mechanism is still wide. Liewise, psychoanalytic constructs like the ego or the superego are not simply information-processing functions—they carry a phenomenological weight, an irreducibly first-person dimension, that computational models do not obviously capture. A full account will need to bridge the explanatory gap between third-person neural dynamics and first-person experience. That said, the practical and clinical possibilities here are still present. Psychedelics appear to transiently dissolve the hierarchical precision structure of the DMN, temporarily flattening the self-model in ways that may allow rigid predictive priors to be revised—a computational account of why psilocybin shows early promise in depression and PTSD. Mindfulness practices reduce DMN activity and may work partly by increasing precision on present-moment sensory input, loosening the grip of entrenched prior beliefs. Future work should prioritize developing computational assays that can distinguish adaptive from maladaptive precision regulation in clinical populations; (2) leveraging high-resolution neuroimaging to trace the specific circuits through which DMN-limbic interactions implement defense; and engaging seriously with the phenomenological tradition to ensure that the first-person dimension of psychic life is not lost in the formalism.

References