Deeply Felt Affect: The Emergence of Valence in Deep Active Inference
Open accessThe DOI matches its Crossref record; 142 of the 143 references it cites were found in a registry or an index, and 1 are not checked yet.
Abstract
The positive-negative axis of emotional valence has long been recognized as fundamental to adaptive behavior, but its origin and underlying function have largely eluded formal theorizing and computational modeling. Using deep active inference, a hierarchical inference scheme that rests on inverting a model of how sensory data are generated, we develop a principled Bayesian model of emotional valence. This formulation asserts that agents infer their valence state based on the expected precision of their action model—an internal estimate of overall model fitness (“subjective fitness”). This index of subjective fitness can be estimated within any environment and exploits the domain generality of second-order beliefs (beliefs about beliefs). We show how maintaining internal valence representations allows the ensuing affective agent to optimize confidence in action selection preemptively. Valence representations can in turn be optimized by leveraging the (Bayes-optimal) updating term for subjective fitness, which we label affective charge (AC). AC tracks changes in fitness estimates and lends a sign to otherwise unsigned divergences between predictions and outcomes. We simulate the resulting affective inference by subjecting an in silico affective agent to a T-maze paradigm requiring context learning, followed by context reversal. This formulation of affective inference offers a principled account of the link between affect, (mental) action, and implicit metacognition. It characterizes how a deep biological system can infer its affective state and reduce uncertainty about such inferences through internal action (i.e., top-down modulation of priors that underwrite confidence). Thus, we demonstrate the potential of active inference to provide a formal and computationally tractable account of affect. Our demonstration of the face validity and potential utility of this formulation represents the first step within a larger research program. Next, this model can be leveraged to test the hypothesized role of valence by fitting the model to behavioral and neuronal responses.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Figures










Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“This vector can be used to encode common sense and habits [34], [35], but can also be used to adapt the plans online due for instance to runtime component failure.”
From Active Inference and Behavior Trees for Reactive Action Planning and Execution in Robotics (2023) · cites this paper as [35]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“agents don’t seek only to minimize prediction error in the present but are sensitive to rates of change in prediction error minimization over time, relative to their prior expectations about that rate”
From Active inference goes to school: the importance of active learning in the age of large language models (2024) · cites this paper as [43]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Deane et al. (2020) suggest that a sensitivity to worse than expected rates of prediction error reduction over time (Kiverstein et al. 2017, 2020; Hesp et al. 2021), manifesting phenomenologically as negative affect, drives the system to switch to more tractable goals.”
From Consciousness in active inference: Deep self-models, other minds, and the challenge of psychedelic-induced ego-dissolution (2021) · cites this paper as Hesp et al. 2021
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Inference about endogenous control of self-evidencing outcomes can thus be understood as an inference about ‘subjective fitness’—the expected precision of the organism’s phenotype-congruent action model (Hesp et al. 2021).”
From Consciousness in active inference: Deep self-models, other minds, and the challenge of psychedelic-induced ego-dissolution (2021) · cites this paper as Hesp et al. 2021
The cited source says what the claim attributes to it.
In the citing paper
“Inference about the reliability of the action model allows the system to increase or decrease precision on the current policy (Kiverstein et al. 2019; Hesp et al. 2021).”
From Consciousness in active inference: Deep self-models, other minds, and the challenge of psychedelic-induced ego-dissolution (2021) · cites this paper as Hesp et al. 2021
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Yaşlı Kadınlarda Üreme Sağlığı
Kokanalı, D., & Engin Üstün, Y. (n.d.) · DergiPark (Istanbul University) · Cited by 170,090
Development and validation of brief measures of positive and negative affect: The PANAS scales.
Watson, D., Clark, L. A., & Tellegen, A. (1988) · Journal of Personality and Social Psychology · doi:10.1037/0022-3514.54.6.1063 · Cited by 32,433 · cited 1×
“…standard measures of valence (e.g., self-report scales of pleasant or unpleasant subjective experience, potentiated startle responses; Watson, Clark, & Tellegen, 1988; Bradley & Lang, 1994; Bublatzky, Guerra, Pastor, Schupp, & Vila, 2013) should be correlated with experimental…”
ResultReinforcement Learning: An Introduction
(2005) · IEEE Transactions on Neural Networks · doi:10.1109/tnn.2004.842673 · Cited by 24,496
Russell, J. A. (1980) · Journal of Personality and Social Psychology · doi:10.1037/h0077714 · Cited by 16,334 · cited 3×
“Some perspectives in emotion research and associated neuroscience research posit that valence is unidimensional (Russell, 1980; Barrett & Russell, 1999) and assume (for example) that a single neural system should increase (or decrease) in activity as valence changes along this dimension.”
BackgroundMeasuring emotion: The self-assessment manikin and the semantic differential
Bradley, M. M., & Lang, P. J. (1994) · Journal of Behavior Therapy and Experimental Psychiatry · doi:10.1016/0005-7916(94)90063-9 · Cited by 9,538 · cited 1×
“…valence (e.g., self-report scales of pleasant or unpleasant subjective experience, potentiated startle responses; Watson, Clark, & Tellegen, 1988; Bradley & Lang, 1994; Bublatzky, Guerra, Pastor, Schupp, & Vila, 2013) should be correlated with experimental inductions of uncertainty about the…”
BackgroundA neural substrate of prediction and reward
Schultz, W., Dayan, P., & Montague, P. R. (1997) · Science · doi:10.1126/science.275.5306.1593 · Cited by 9,388 · cited 1×
“AC is formally related to reward prediction error within reinforcement learning models (Friston et al., 2014; Schultz, Dayan, & Montague, 1997; Stauffer, Lak, & Schultz, 2014).”
BackgroundThe organization of the human cerebral cortex estimated by intrinsic functional connectivity
Thomas Yeo, B. T., Krienen, F. M., Sepulcre, J., Sabuncu, M. R., Lashkari, D., Hollinshead, M., Roffman, J. L., Smoller, J. W., Zöllei, L., Polimeni, J. R., et al. (2011) · Journal of Neurophysiology · doi:10.1152/jn.00338.2011 · Cited by 9,182
The free-energy principle: a unified brain theory?
Friston, K. (2010) · Nature Reviews Neuroscience · doi:10.1038/nrn2787 · Cited by 8,120 · cited 4×
“This framework revolves around the notion that all living systems embody statistical models of their worlds (Friston, 2010; Gallagher & Allen, 2018).”
BackgroundMethodInfluentialSelf-citationFunctional Network Organization of the Human Brain
Power, J. D., Cohen, A. L., Nelson, S. M., Wig, G. S., Barnes, K. A., Church, J. A., Vogel, A. C., Laumann, T. O., Miezin, F. M., Schlaggar, B. L., et al. (2011) · Neuron · doi:10.1016/j.neuron.2011.09.006 · Cited by 4,124
Parvalbumin neurons and gamma rhythms enhance cortical circuit performance
Sohal, V. S., Zhang, F., Yizhar, O., & Deisseroth, K. (2009) · Nature · doi:10.1038/nature07991 · Cited by 2,606
Research works printed but not linked: 1
- Gray, J. A. (1994). Three fundamental emotion systems. In P.Ekman & R. J.Davidson (Eds.), The nature of emotion (pp. 243–247). New York: Oxford University Press.
Cited by
Items 201–245 of 245
Modelling and Measuring the Predictive Brain
Whyte, C. J. (2020) · The Sydney eScholarship Repository (The University of Sydney)
Learning Uncertainty: The Free Energy Principle in Anxiety FormationPreprint
McGovern, H. T., Foe, A. D., Leptourgos, P., Corlett, P. R., Bandara, K. H., Biddell, H., & Hutchinson, B. T. (2021) · doi:10.31234/osf.io/8zsv3
How to Prevent Robotic Sociopaths: A Neuroscience Approach to Artificial EthicsPreprint
Christov‐Moore, L., Reggente, N., Vaccaro, A. G., Schoeller, F., Pluimer, B. R., Douglas, P. K., Iacoboni, M., & Kaplan, J. (2022) · PsyArXiv (OSF Preprints) · doi:10.31234/osf.io/6tn42 · Cited by 2
Hipólito, I., & Hesp, C. (2022) · doi:10.31234/osf.io/t9632
Self-citationThe seductive allure of cargo cult computationalism
Allen, M. (2022) · Behavioral and Brain Sciences · doi:10.1017/s0140525x22000279 · Cited by 1
Self-citationEdelman, S. (2022) · Springer eBooks · doi:10.1007/978-3-031-24012-6_3
Computational Modeling in Various Cognitive Fields
(2023) · Cambridge University Press eBooks · doi:10.1017/9781108755610.027
A computational perspective to affect generation in perceptual decisionsPreprint
Voodla, A., Uusberg, A., & Desender, K. (2023) · PsyArXiv (OSF Preprints) · doi:10.31234/osf.io/qgta9
Attuning the World: Ambient Smart Environments for Autistic Fields of AffordancesPreprint
Nešić, J. (2023) · doi:10.31234/osf.io/k6txe
Servajean, P., & Servajean, R. (2023) · PsyArXiv (OSF Preprints) · doi:10.31234/osf.io/2z97y
Adaptive Pacing in Word Segmentation and the Vowel-Onset-Paced Syllable Inference Model
Pittman-Polletta, B., & Dilley, L. C. (2026) · Cambridge University Press eBooks · doi:10.1017/9781009295888.015 · Cited by 1
(2024) · The MIT Press eBooks · doi:10.7551/mitpress/15196.003.0041
Burdening the Predictive Mind: A Predictive Processing Approach to Health-Related BurdensPreprint
Smart, P. R., Fair, N., Fraser, S. D., & Boniface, M. (2024) · PsyArXiv (OSF Preprints) · doi:10.31234/osf.io/h8scp
Proietti, R., Parr, T., Tessari, A., Friston, K. J., & Pezzulo, G. (2024) · PsyArXiv (OSF Preprints) · doi:10.31234/osf.io/m87wk
Self-citationMetacognitive Control of Linear Dynamic Systems with Self-confidence Adaptation
Meera, A. A., & Lanillos, P. (2024) · Springer proceedings in advanced robotics · doi:10.1007/978-3-031-76424-0_19
Surprise! Using Physiological Stress for Allostatic Regulation Under the Active Inference Framework
Khan, I., & Lowe, R. (2024) · International Conference on Affective Computing and Intelligent Interaction and workshops · doi:10.1109/acii63134.2024.00024 · Cited by 5
“AIF offers a robust formalism from which to approach adaptive and allostatic regulation [14]–[16] as well as affective regulation [5], [11], [17], [18] of (biological) systems.”
Cultural bodybuilding: The embodied influence of culture on perception and action
Albarracin, M. (2025) · The Psychology of learning and motivation/The psychology of learning and motivation · doi:10.1016/bs.plm.2025.02.002 · Cited by 1
The Human AffectomePreprint
Schiller, D., Yu, A. N. C., Alia‐Klein, N., Becker, S., Cromwell, H. C., Dolcos, F., Eslinger, P. J., Frewen, P., Kemp, A. H., Pace-Schott, E. F., et al. (2022) · PsyArXiv (OSF Preprints) · doi:10.31234/osf.io/9nu32_v1 · Cited by 5
Waggoner, M. (2025) · Philosophical Psychology · doi:10.1080/09515089.2025.2537896
“…literature points to the fact that emotional valence seems to operate according to optimal Bayesian infer-encing, making use of priors or of stored information and weighting this relative to the precision of incoming information (Hesp et al., 2021; Majumdar et al., 2023; McGovern et al., 2022).”
Brains, Bodies, and Their Worlds: Towards a Model of Predictive Social Actors
Cheadle, J. E., Davidson-Turner, K., & Goosby, B. J. (2025) · Handbooks of sociology and social research · doi:10.1007/978-3-031-95615-7_16
Modeling the Action-Perception loop and its role in phantom limb pain using active inference
Ramne, M., Lundh, T., & Sensinger, J. W. (2026) · Journal of Theoretical Biology · doi:10.64898/2025.12.05.692511
“Recent work further indicates that emotional valence may be encoded in the perceived success or failure of actions [42].”
Information and emotion coevolution in forecasting technology perceptionPreprint
Trieste, L., & Turchetti, G. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.6151455
Emotion and allostatic control: An active inference account of emotion regulation
Carìa, A., & Pezzulo, G. (2026) · Neuroscience & Biobehavioral Reviews · doi:10.1016/j.neubiorev.2026.106639 · Cited by 1
Dream content and slow waves benefit prey against predator in a video game confrontation
Brandão, D. S., Scott, R. N. B., Soares, E. S., Mota, N. B., & Ribeiro, S. G. (2026) · Scientific Reports · doi:10.1038/s41598-026-42759-7
Machhi, V. S., & Shah, A. (2026) · Cognitive Affective & Behavioral Neuroscience · doi:10.3758/s13415-026-01418-0 · Cited by 1
“(cid:129) Valence is often associated with the motivational dimen-sion of approach versus avoidance, which has been consistently linked to regional brain activity, particularly asymmetrical activity in the prefrontal cortex measured via EEG (Hesp et al., 2021).”
Development and study of coadaptive interactive systems
Dechaux, A. (2024) · theses.fr (ABES) · doi:10.70675/24336488z62bdz4b32zbc69z1620e376bc34
Márquez, F. O. S., & Ortiz, G. E. R. (2026) · Advances in Human-Computer Interaction · doi:10.1155/ahci/8435912
Rahman, S., & Pashea, A. (2026) · Frontiers in Sociology · doi:10.3389/fsoc.2026.1646503
Understanding the Burden of Illness: Steps Towards an Ontology of Patient Experience
Smart, P. R., Fair, N., & Boniface, M. (2026) · Applied Ontology · doi:10.1177/15705838261438032
“First, it provides a point of contact with emerging scientific theories of cognitive and affective phenomena, which are increasingly framed in terms of prediction and prediction-error minimization (Fernandez Velasco & Loev, 2021; Hesp et al., 2020; Miller et al., 2022; Smith, Varshney, et al., 2022).”
An information-theoretic account of cognitive effort
Zhang, M. (2026) · Philosophical Psychology · doi:10.1080/09515089.2026.2671371
Srinivasan, H., Cascio, C. J., & Wallace, M. T. (2026) · Journal of Child and Adolescent Psychopharmacology · doi:10.1177/10445463261448490
“Dysregulation can also stem from unstable gain control, as seen in ADHD, where the system fl uctuates in how strongly it weights incoming signals (Hauser et al., 2016; Hesp et al., 2021).”
Emotions Are Folk Concepts in a Predicting Brain
Lindquist, K. A. (2026) · Brain Sciences · doi:10.3390/brainsci16060614
“In contrast, valence is thought to emerge as a product of model fit; when an internal model fits well in the current environment, this is associated with positive valence; in contrast, when the model fits poorly (i.e., there is uncertainty), this is associated with negative valence (see [25]).”
Parvizi-Wayne, D., Pedersen, R., & Rankin, J. (2026) · Review of Philosophy and Psychology · doi:10.1007/s13164-026-00828-w
“…with respect to negative valence, such that, over a period of time, “it feels bad for an agent when they reduce error at a slower than expected rate” (ibid., p. 2866) (see also Hesp et al. 2021 ; Kiverstein and Miller 2023; Kiverstein et al. 2020 ; Miller et al. 2022 ; Van de Cruys 2017, p. 10).”
Consciousness and psychiatric disorders
Northoff, G., Lechner, S., & Arantes-Gonçalves, F. (2026) · Elsevier eBooks · doi:10.1016/b978-0-443-29258-3.00041-0
Affective Modelling and Valenced Mixtures Within a Hierarchical Bayesian Architecture
Araya, J. M. (2026) · Topoi · doi:10.1007/s11245-026-10465-8
McFarnell, S. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-33195-3_5
Klimaj, V., & Safron, A. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-33010-9_28
Lutz, A., Abdoun, O., Dor‐Ziderman, Y., Trautwein, F., Sandved-Smith, L., Bogotá, J. D., Kiverstein, J. D., & Berkovich‐Ohana, A. (2026) · Elsevier eBooks · doi:10.1016/b978-0-443-29258-3.00046-x
Conative inference as a mechanism of allostatic resilience: wanting to be and being to want
Krupnik, V. (2026) · Cognitive Affective & Behavioral Neuroscience · doi:10.3758/s13415-026-01485-3
Unveiling the Aha! moment: A computational account of insight in active inference
Doulfoukar, Y., Pezzulo, G., & Stuyck, H. (2026) · Psychonomic Bulletin & Review · doi:10.3758/s13423-026-02983-8
“This framework has been employed to account for various brain phenomena, including attention (Mirza et al., 2019), affect (Hesp et al., 2021) psychopathologies (Bad-cock&Davey,2024;Jeganathan&Breakspear,2021;Linson & Friston, 2019; Paulus et al., 2019) and neuromodulation (Friston et al., 2012;…”
Aesthetic taste in humans and machines
Piovesan, F., Frascaroli, J., & de Cruys, S. V. (2026) · Current Opinion in Psychology · doi:10.1016/j.copsyc.2026.102461
Self-Efficacy Under the Free Energy Principle: A Mathematical Model for Adaptive Coaching AI
Inamura, T. (2026) · IEEE RO-MAN · doi:10.1109/ro-man72286.2026.11714682
“…casts it as precision— Sandved-Smith et al. [9] model meta-awareness as higher-level states modulating lower-level precision, and Hesp et al. [10] formalize an efficacy-like valence as the expected precision of the action model—but both treat efficacy and metacognition as a single global…”
Beyond Cortical Signatures: Measurement Gaps Across Theories of Consciousness
Svet, M., & Arachchige, A. S. P. M. (2026) · Brain Sciences · doi:10.3390/brainsci16101087
“Formal accounts deriving valence from prediction-error dynamics or deep active-inference hierarchies remain proposals rather than settled results [98–100].”
Reframing the lived experience of persistent dizziness: insights from active inference
Broberg, M. A., Thacker, M., Friston, K. J., & Esteves, J. E. (2026) · Frontiers in Psychology · doi:10.3389/fpsyg.2026.1867836
“Moreover, affective valence is argued to represent one’s inference about their ability to successfully produce adaptive actions (Hesp et al., 2021).”
InfluentialSelf-citationSocially embedded predictive construction: a model of emotional experience
Grimaldi, R. (2026) · Academia Neuroscience and Brain Research · doi:10.20935/acadneurosci8594
Metadata
- Authors
- Hesp, C., Smith, R., Parr, T., Allen, M., Friston, K. J., & Ramstead, M. J. D.
- Affiliations
- Wellcome Centre for Human Neuroimaging (GB)
- Laureate Institute for Brain Research (US)
- Aarhus University (DK)
- Corresponding author
- Hesp, Casper
- Smith, Ryan
- Parr, Thomas
- Allen, Micah
- Friston, Karl J.
- Ramstead, Maxwell J. D.
- Venue
- Neural Computation
- Publisher
- MIT Press
- Published
- 2021-02-01
- Volume
- 33
- Issue
- 2
- Pages
- 398-446
- ISSN
- 0899-7667 (ISSN-L)
- 1530-888X
- PMC
- PMC8594962
- OpenAlex
- W3108742966
- Cited by
- 246
- References
- 126
- Citations by year
- 2026: 32 · 2025: 29 · 2024: 40 · 2023: 40 · 2022: 47 · 2021: 38
- FWCI
- 19.22
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Neural Computation (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by-nc)
- PubMed (publishedVersion)
- UvA-DARE (University of Amsterdam) (submittedVersion, cc-by)
- UCL Discovery (University College London) (submittedVersion)
- PsyArXiv (OSF Preprints) (submittedVersion)
- PsyArXiv (OSF Preprints) (submittedVersion, cc-by)
- Data Archiving and Networked Services (DANS) (submittedVersion, other-oa)
- Funders
- William K. Warren Foundation
- Aarhus Universitet
- McGill University
- Wellcome Trust (R272-2017-4345, 088130, 754513, 088130/Z/09/Z)
- European Commission (HORIZON2020, 754513)
- Rosetrees Trust (173346)
- Aarhus Universitets Forskningsfond (754513, R272-2017-4345)
- Nederlandse Organisatie voor Wetenschappelijk Onderzoek (406.18, 406.18.535)
- Lundbeckfonden (R272‐2017‐4345, 754513)
- Canada First Research Excellence Fund
- Social Sciences and Humanities Research Council of Canada
- H2020 Marie Skłodowska-Curie Actions (754513, HORIZON2020, R272-2017-4345)
- Topics
- Life Sciences › Neuroscience › Cognitive Neuroscience › Embodied and Extended Cognition
- Life Sciences › Neuroscience › Cognitive Neuroscience › Cognitive Science and Education Research
- Physical Sciences › Computer Science › Artificial Intelligence › Psychiatry, Mental Health, Neuroscience
- Keywords
- emotional valence, second-order beliefs, metacognition, Bayesian inference, top-down modulation, T-maze
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.