Mindscape Collective is now The Consciousness Library. Same library, new name. You may need to sign in again. About the change
Skip to content

Karl Friston

45 papers in the library · 4,199 citations · publishing 1995-2026

Papers

Active interoceptive inference and the emotional brain

Philosophical Transactions of the Royal Society B Biological Sciences October 11, 2016 Anil K. Seth, Karl Friston 936 citations

Interoception—the perception of internal bodily states—is being reconceptualized through the lens of hierarchical inference in the brain. Bodily states are regulated by autonomic reflexes enslaved by descending predictions from deep generative models of our internal and external environment. This re-conceptualization, termed interoceptive inference, illuminates issues in cognitive and clinical neuroscience, including experiences of selfhood and emotion. The review contextualizes interoception in active Bayesian inference, highlighting its enactivist aspects, the role of uncertainty or precision in neuromodulation, implications for the functional anatomy of the emotional brain, and the role of interoception in shaping a sense of embodied self and feelings. Links are drawn between homeostasis, allostasis, cybernetic predictive control, and predictive processing. The explanatory scope ranges from autism and depression to consciousness.

Broadband Cortical Desynchronization Underlies the Human Psychedelic State

Journal of Neuroscience September 18, 2013 Matthew J. Brookes, David Errtizoe, Ben Sessa et al. 501 citations

Psychedelic drugs like psilocybin produce profound changes in consciousness by desynchronizing ongoing oscillatory rhythms in the cortex. Using magnetoencephalography in healthy participants, psilocybin reduced spontaneous cortical oscillatory power from 1 to 50 Hz in posterior association cortices and from 8 to 100 Hz in frontal association cortices, with large decreases in default-mode network areas. Low-level visually induced and motor-induced gamma-band oscillations were unaffected, suggesting some basic oscillatory activity is preserved. Dynamic causal modeling indicated that posterior cingulate cortex desynchronization results from increased excitability of deep-layer pyramidal neurons rich in 5-HT 2A receptors.

A tale of two densities: active inference is enactive inference.

Adaptive Behavior August 1, 2020 Maxwell Jd Ramstead, Michael D Kirchhoff, Karl Friston 387 citations

The free-energy principle (FEP) and active inference are often conflated with predictive processing frameworks, leading to misunderstandings about their core constructs, generative models and variational densities. This article argues that these models have been systematically misrepresented as structural representations. Instead, under the FEP, generative and recognition models function to realize inference and control—the self-organizing, belief-guided selection of action policies—not as representations with the properties structural representationalists ascribe. The authors propose an enactive interpretation, termed enactive inference, as a more accurate account of these constructs.

From cognitivism to autopoiesis: towards a computational framework for the embodied mind.

Synthese January 1, 2018 Micah Allen, Karl Friston 387 citations

Predictive processing (PP) approaches to the mind vary widely, from cognitivist views that rely on modular, internal mental representations to radical enactive and embodied theories. This review maps the continuum of PP theories, showing that some emphasize body-representations while others align with dynamic, enactive accounts. The Free Energy Principle (FEP) offers a formal framework that reconciles internalist and externalist perspectives by explaining how internal representations arise from autopoietic self-organization. The FEP thus provides a foundation for empirically productive process theories, such as PP, that guide research through formal modeling of the embodied mind.

Thinking through other minds: A variational approach to cognition and culture.

The Behavioral and brain sciences May 30, 2019 Samuel P. L. Veissière, Axel Constant, Maxwell J D Ramstead et al. 362 citations

A unifying account of how humans acquire shared cultural habits, norms, and expectations is developed by integrating the variational free-energy principle from theoretical neuroscience with concepts of cultural evolution and implicit learning. Humans construct social niches that provide epistemic resources called cultural affordances. Through immersive participation in patterned cultural practices, agents learn by inferring what other people expect—a process termed "thinking through other minds" (TTOM). This makes information about others' expectations the primary statistical regularity humans use to predict and organize behavior. The model aims to resolve debates in cognitive science between internalist and externalist accounts of theory of mind and between dynamical and representational views of enactivism.

Effective connectivity changes in LSD-induced altered states of consciousness in humans.

Proc Natl Acad Sci U S A January 28, 2019 Katrin H. Preller, Adeel Razi, Peter Zeidman et al. 303 citations

LSD alters communication within brain pathways that filter sensory information, according to a brain imaging study. Using a double-blind, placebo-controlled design with 25 healthy participants, researchers found that LSD increased signaling from the thalamus to the posterior cingulate cortex—an effect dependent on serotonin 2A receptor activation—and decreased signaling from the ventral striatum to the thalamus independently of those receptors. These changes in directed connectivity within cortico-striato-thalamo-cortical loops support the thalamic filter model of psychedelic action, which proposes that psychedelics disrupt the gating of sensory information to the cortex. The findings advance understanding of how psychedelics alter consciousness and may inform development of new therapeutics.

Towards a computational phenomenology of mental action: modelling meta-awareness and attentional control with deep parametric active inference.

Neuroscience of Consciousness January 1, 2021 Lars Sandved-Smith, Casper Hesp, Jérémie Mattout et al. 116 citations

Meta-awareness, the ability to notice the current content of consciousness, is crucial for controlling cognitive states like directing attention. This paper models meta-awareness and attentional control using hierarchical active inference, treating mental actions as policy choices over higher-level cognitive states. A further hierarchical level represents meta-awareness states that modulate the expected confidence in the mapping between observations and hidden cognitive states. Simulations of mind-wandering during a sustained selective attention task illustrate how this inferential architecture enables accessing and controlling cognitive states, offering a computational foundation for a phenomenology of mental action and self-monitoring.

Characterising the complexity of neuronal interactions

Human Brain Mapping January 1, 1995 Karl Friston, Giulio Tononi, Olaf Sporns et al. 114 citations

Neuronal interactions in the brain balance two opposing organizational principles: functional segregation, where specialized cortical areas exhibit relatively high entropy (unpredictable dynamics), and functional integration, where distributed influence across areas produces lower entropy overall. A measure of complexity is highest when small brain regions have high entropy on average relative to the whole system's entropy, equivalent to the average mutual information between small regions and the rest of the system. Applied to nonlinear simulations and fMRI data during photic stimulation, complexity peaked between high-dimensional chaotic behavior and low-dimensional orderly behavior—between asynchronous oscillations and global synchrony—confirming the hypothesis.

Modeling Ketamine Effects on Synaptic Plasticity During the Mismatch Negativity

Cerebral Cortex August 8, 2012 André Schmidt, Andreea O. Diaconescu, Michael Kometer et al. 110 citations

Using dynamic causal modeling and Bayesian model selection on data from a double-blind, placebo-controlled, crossover ketamine study, the authors investigated how the NMDA-receptor antagonist ketamine reduces mismatch negativity (MMN) amplitudes. Guided by a predictive coding framework that unifies adaptation and model adjustment theories, they compared models allowing different expressions of neuronal adaptation and synaptic plasticity. Results replicated that both adaptation and short-term plasticity are necessary for MMN generation. Ketamine significantly affected synaptic plasticity but not adaptation, with a selective effect on the forward connection from left primary auditory cortex to superior temporal gyrus. This model-based estimate of ketamine's effect on synaptic plasticity correlated with ratings of ketamine-induced impairments in cognition and control, suggesting a concrete mechanism linking ketamine effects on MMN to drug-induced psychopathology.

The mixed serotonin receptor agonist psilocybin reduces threat-induced modulation of amygdala connectivity

NeuroImage Clinical August 22, 2015 Rainer Kraehenmann, André Schmidt, Karl Friston et al. 107 citations

Psilocybin reduces the brain's threat response by weakening top-down signals from the amygdala to the primary visual cortex. Using dynamic causal modeling of fMRI data, researchers found that psilocybin decreased the threat-induced modulation of this specific connection within the visual-limbic-prefrontal network. This neural mechanism may help explain how psilocybin shifts emotional processing away from negative toward positive stimuli, which could be relevant for treating mood and anxiety disorders.

Canalization and plasticity in psychopathology

Neuropharmacology December 27, 2022 Robin Carhart-Harris, Shamil Chandaria, David Erritzøe et al. 106 citations

A theoretical model proposes that psychopathology arises from a defensive process called canalization, which narrows an individual's range of thoughts, feelings, and behaviors by increasing precision or reducing variance in neural responses. This contrasts with an early form of plasticity, TEMP (Temperature or Entropy Mediated Plasticity), which increases variance and learning rate. Canalization entrenches pathology as the agent develops expertise in their disorder, while TEMP, combined with gentle psychological support, may counter this entrenchment. The model distinguishes adaptive from maladaptive canalization and suggests concrete experiments to test its hypotheses.

In the Body’s Eye: The Computational Anatomy of Interoceptive Inference

bioRxiv Preprint Server April 10, 2019 Micah Allen, Andrew Levy, Thomas Parr et al. 100 citations preprint

A formal model of cardiac active inference explains how signals from the heart influence perception of the outside world and confidence in that perception. Simulated experiments reproduce the defensive startle reflex and the link between the cardiac cycle and fear perception. Simulated interoceptive lesions reduce fear expectations, cause psychosomatic hallucinations, and worsen metacognitive biases. Synthetic heart-rate variability analyses show how the balance of arousal-priors and visceral prediction errors creates distinct patterns of physiological reactivity. The model provides a way to computationally characterize disordered brain-body interaction.

The Active Inference Approach to Ecological Perception: General Information Dynamics for Natural and Artificial Embodied Cognition

Frontiers in Robotics and AI March 8, 2018 A. Linson, A. Clark, S. Ramamoorthy et al. 94 citations

The active inference framework (AIF) offers a unified, naturalistic account of life, mind, and consciousness by grounding them in the principle of free energy minimization. It bridges computational neuroscience, robotics, ecological psychology, and phenomenology, treating particles, organisms, and artificial agents under a common information-theoretic foundation. The paper introduces AIF, then examines its implications for evolutionary theory, ecological psychology, embodied phenomenology, and robotics, concluding with considerations for machine consciousness.

The Projective Consciousness Model and Phenomenal Selfhood.

Frontiers in Psychology January 1, 2018 Kenneth Williford, Daniel Bennequin, Karl Friston et al. 85 citations

The Projective Consciousness Model (PCM) combines a projective geometric model of the perspectival structure of conscious experience with a variational free-energy minimization model of active inference, explaining how consciousness serves a cybernetic function: modulating cognitive and affective dynamics to control embodied agents. Projective transformations link geometry and inference, integrating perception, emotion, memory, reasoning, and perspectival imagination to optimize behavior, resilience, and preference satisfaction. The PCM makes empirical predictions, fits a neurocomputational framework, and accounts for pre-reflective self-consciousness, the first-person perspective, the sense of ownership, and social self-consciousness. The authors argue it offers the most complete theory to date of phenomenal selfhood.

Neural Mechanisms and Psychology of Psychedelic Ego Dissolution

Pharmacological Reviews September 9, 2022 Devon Stoliker, Adeel Razi, Gary F. Egan et al. 83 citations

Classic psychedelics work primarily by binding to serotonergic 5-HT2A receptors, and their agonist activity at these receptors changes synaptic efficacy, profoundly affecting hierarchical message-passing in the brain. This review synthesizes cognitive and neuroimaging evidence showing that psychedelics influence selfhood and subject-object boundaries—a phenomenon called ego dissolution—which may underlie their subjective and therapeutic effects. Because 5-HT2A receptors sit at the apex of the cortical hierarchy, their agonism may powerfully affect sentience and consciousness. Effects can last beyond the pharmacological half-life, suggesting psychedelics promote neural plasticity. Psychologically, they may disarm ego resistance, expanding the repertoire of perceptual hypotheses and enabling alternate pathways for thought and behavior. The authors interpret these effects through hierarchical predictive coding, offering testable predictions about effective connectivity in cortical hierarchies.

Understanding visual hallucinations: A new synthesis.

Neuroscience and Biobehavioral Reviews July 1, 2023 Daniel Collerton, James Barnes, Nico J Diederich et al. 80 citations

Eight distinct models of complex visual hallucinations have been proposed since 2000, each based on different views of brain organization. Researchers from each model group have now agreed on an integrated Visual Hallucination Framework that aligns with current theories of both real and hallucinatory vision. The Framework identifies cognitive systems involved in hallucinations and enables systematic investigation of how hallucination experiences relate to changes in underlying cognitive structures. The episodic nature of hallucinations points to separate factors for their onset, persistence, and end, suggesting a complex relationship between temporary states and long-term traits of hallucination risk. The Framework also suggests new research directions and potential treatments for distressing hallucinations.

I overthink—Therefore I am not: An active inference account of altered sense of self and agency in depersonalisation disorder

Consciousness and Cognition April 28, 2022 Anna Ciaunica, Anil K. Seth, Jakub Limanowski et al. 76 citations

Depersonalisation disorder may arise when the brain fails to modulate sensory precision, leading to a loss of the normal sense of control over one's own perceptions and actions. The paper proposes that individuals with depersonalisation adopt the hypothesis that they are embodied perceivers who are not in control of their perception, while still recognizing that the controlling agent is themselves. This account uses predictive processing to explain the phenomenology of depersonalisation, including the feeling of being detached from one's own thoughts and actions. The analysis suggests specific psychophysical and physiological abnormalities that could underlie these experiences.

From Generative Models to Generative Passages: A Computational Approach to (Neuro) Phenomenology.

Review of Philosophy and Psychology January 1, 2022 Maxwell J D Ramstead, Anil K. Seth, Casper Hesp et al. 75 citations

A version of neurophenomenology is presented that uses generative modelling techniques from computational neuroscience and biology to formally model descriptions of lived experience from the phenomenological tradition (e.g., Husserl, Merleau-Ponty). The approach, called computational phenomenology, is situated within the broader project of naturalizing phenomenology. Philosophical objections to that project are evaluated, and the generative modelling framework is reviewed. The approach differs from previous uses of generative modelling for consciousness by constructing computational models of inferential or interpretive processes that best explain particular kinds of lived experience.

Philosophy of psychiatry: theoretical advances and clinical implications.

World psychiatry : official journal of the World Psychiatric Association (WPA) June 1, 2024 Dan J Stein, Kris Nielsen, Anna Hartford et al. 70 citations

Psychiatric understanding and treatment benefit from integrating both objective facts and subjective values, moving beyond strict scientism toward a softer naturalism. A pluralist approach—embracing ontological, explanatory, and value pluralism—acknowledges the multi-level causal interactions underlying psychopathology and highlights the importance of lived experience and diverse difference-makers in research and practice. Embodied, embedded, and enactive views of the brain-mind offer a conceptual framework for the mind-body problem that clinically integrates cognitive-affective neuroscience with phenomenological psychopathology.

From generative models to generative passages: A computational approach to (neuro)phenomenology

PsyArXiv February 23, 2021 Maxwell James Ramstead, Anil K. Seth, Casper Hesp et al. 21 citations preprint

A new approach called computational phenomenology uses generative modeling techniques from computational neuroscience to study conscious experience. The paper reviews efforts to naturalize phenomenology, addresses philosophical objections, and explains how generative models can simulate the inferential processes underlying specific types of lived experience. This differs from prior uses of generative modeling for consciousness by focusing on modeling the interpretive process that best accounts for particular phenomenal experiences.

Forgetting ourselves in flow: an active inference account of flow states and how we experience ourselves within them

Frontiers in Psychology June 3, 2024 Darius Parvizi-Wayne, Lars Sandved-Smith, Riddhi J. Pitliya et al. 20 citations

Flow is a state of optimal performance experienced across domains like art, athletics, gaming, and writing. Its puzzling features include a reported loss of self-awareness despite skilled agency, and effortlessness despite task complexity. Using the active inference framework—where action and perception minimize variational free energy—the authors propose that flow arises from high precision weighting on expected sensory consequences of action and beliefs about sequential action. This draws the embodied system to exploit pragmatic affordances while restricting counterfactual planning, leading to inhibition of the sense of self as a temporally extended object and higher-order self-conceptualization. However, self-awareness is not entirely lost; it remains pre-reflective and bodily.

A beautiful loop: An active inference theory of consciousness.

Neuroscience and Biobehavioral Reviews September 1, 2025 Ruben Laukkonen, Karl Friston, Shamil Chandaria 17 citations

A theoretical paper proposes that active inference can model consciousness through three conditions: a world model (epistemic field) defining what can be known, inferential competition (Bayesian binding) selecting only coherent inferences that reduce long-term uncertainty, and epistemic depth—a recursive sharing of beliefs throughout a hierarchical system like the brain. This loop allows the world model to know itself non-locally and continuously evidence that knowing, distinct from self-consciousness. The authors formally propose a hyper-model for precision-control whose latent states encode global weighting rules, enacting epistemic agency and flexibility reminiscent of general intelligence. The theory also addresses altered states, meditation, and the full spectrum of conscious experience.

The paradox of the self-studying brain.

Physics of Life Reviews March 1, 2025 Simone Battaglia, Philippe Servajean, Karl Friston 14 citations

The brain's attempt to study itself creates a paradox of self-reference, raising questions about consciousness, psychiatric disorders, and the limits of science. Historically a philosophical issue, modern techniques like functional and structural brain imaging and neurostimulation now allow researchers to probe this inquiry. However, the broader implications remain unclear. The need to use both perception and introspection has led to different formulations of consciousness, and evidence for one does not necessarily support another. Deconstructing this paradox from philosophical and neuroscientific perspectives may yield insights into self-awareness and consciousness.

Towards a computational (neuro)phenomenology of mental action: modelling meta-awareness and attentional control with deep-parametric active inference

PsyArXiv June 10, 2020 Lars Sandved-Smith, Casper Hesp, Antoine Lutz et al. 11 citations preprint

This theoretical paper proposes a computational model of mental action—the deliberate control of one's own thoughts and attention—by combining insights from phenomenology and active inference. The authors develop a deep-parametric active inference framework to simulate meta-awareness and attentional control, showing how agents can learn to monitor and regulate their own cognitive processes. The model suggests that meta-awareness emerges from hierarchical inference about attentional states, enabling flexible control of attention. This work bridges phenomenological philosophy and computational neuroscience, offering a formal account of how conscious agents can intentionally shape their own mental activity.

Cross-Frequency Coupling as a Neural Substrate for Prediction Error Evaluation: A Laminar Neural Mass Modeling Approach

bioRxiv (Cold Spring Harbor Laboratory) March 19, 2025 Giulio Ruffini, Edmundo Lopez-Sola, Raul P. Aristides et al. 8 citations preprint

Cross-frequency coupling (CFC), where brain rhythms at different speeds interact, may be the mechanism the brain uses to compare sensory input with internal predictions. Using a laminar neural mass model, the authors show that two forms of CFC—signal-envelope coupling and envelope-envelope coupling—can implement hierarchical prediction-error computation and precision-weighting. In Alzheimer's disease, disruptions in fast-spiking interneurons lead to aberrant prediction errors: inflated early on, then attenuated. Serotonergic psychedelics reduce the influence of predictions, increasing prediction-error signals. These findings suggest that CFC across multiple timescales is a key computational mechanism supporting predictive coding, with disruptions central to certain disorders.