Emotion and Prediction
Online Workshop | 31 March - 1 April 2021 12h-14h GMT
Speakers: Regina Fabry, Daphne Demekas, Slawa Loev, Pablo Fernández, Mark Miller, Abby Tabor, José M. Araya
Emotion permeates all mental life - it is a reflection of how we are faring in our various pursuits, it imbues our activities and our environments with meaning and purpose, and it motivates and modulates our behaviours. We are emotional creatures through and through. While there has been a tremendous amount of work done on this topic, to date an integrative account capable of unifying the various theoretical perspectives and experimental results is still lacking.
Recently, a possible unifying account of cognition (and perhaps emotion) has begun to emerge within computational neuroscience. According to the so-called predictive processing framework the brain is constantly attempting to minimize the discrepancy between its sensory expectations and its actual incoming sensory signals. This framework offers an architecture in which distinct functions can be explained at their different time-scales by the same computational principles, and where distinct theories can find a common language, which brings fruitful modelling advantages. As such it is quickly becoming an attractive way of carrying out theoretical and experimental research in cognitive science.
More recently this framework has most recently begun to serve as the architectural basis for an exciting and very promising new account of feelings, emotions and moods. This workshop will bring together philosophers and cognitive scientists working on predictive processing and emotion in order to promote an interdisciplinary dialogue about the nature of emotion in predictive system like us.
DAY 1: 12h to 12h40
What is the Relationship between Emotions and Moods?
Predictive Processing and the Affective Mind
Day 1: 12h40 to 13h20
Active Inference for Emotion Recognition
Neuroscientific and philosophical research on affectivity suggests a distinction between emotions and moods. Emotions structure occurrent phenomenal experiences and are characterised by the intentional directedness towards objects or persons in the individual’s local environment (Damasio, 1999; Stephan, 2012). By contrast, moods can be understood as persistent, pre-intentional states that pre-structure phenomenal experiences (Ratcliffe, 2015). Moods, philosophical research indicates, pre-structure or facilitate occurrent emotional experiences (Ratcliffe, 2010; Stephan, 2017). Recently, it has been suggested that predictive processing (and related) accounts have the potential to capture the crucial contributions of emotions (Barrett, 2017; Miller & Clark, 2018; Seth, 2013; Wilkinson, Deane, Nave, & Clark, 2019) and moods (Clark, Watson, & Friston, 2018; Kiverstein, Miller, & Rietveld, 2020) to our mental lives. Emotions would be enabled by precision-modulated interoceptive inference across multiple subcortical and cortical levels of the hierarchical generative model implemented in the human brain. By contrast, moods would be enabled by the estimation of expected rates of prediction error minimisation (prediction error dynamics). In this talk, I will explore the question as of how emotions and moods could be related under predictive processing (for a starting point, see Fernandez Velasco & Loev, 2020). In particular, I will explore the option that prediction error dynamics (enabling the onset and maintenance of moods) could causally constrain interoceptive inference (enabling emotional experiences). If feasible, this option would be consistent with phenomenological work on the mood-emotion nexus.
Recently great strides are being made in our understanding of how to model the brain and behaviour, and in particular emotional states and inference. In “An Investigation of the Free Energy Principle for Emotion Recognition” we propose that emotion recognition and inference devices will evolve from state-of-the-art deep learning models into active inference schemes that go beyond marketing applications and become adjunct to psychiatric practice. This presentation will cover some background on the FEP and modeling of emotions, and then outline the three prospective waves of emotion recognition technology that can be achievable by further developing active inference methods.
Day 1: 13h20 to 14h
Slawa Loev and Pablo Fernández
Affective Experience in the Predictive Mind
We review existing accounts of affective experience within the Predictive Processing framework and propose and integrated theory, which holds that valence corresponds to the expected rate of prediction error reduction and that the formal object of a feeling is a predictive model of the expected changes in prediction error rate.
Day 2: 12h to 12h40
Active Inference and the Paradox of Horror
Starting from the active inference framework, I will explore in this talk why an agent striving to minimize surprise would find pleasure in engaging with horror films. Why do we sometimes find it enjoyable to engage with fictions that evoke negative emotions such as fear and disgust? The answer I will propose lies in what I will call “consumable error” - prediction errors with just the right amount of complexity and novelty for the agent to make progress in its learning. Films carefully curate error slopes for their viewers - they play at frustrating our expectations and then resolving those prediction errors over the course of the film. I will argue that it is these carefully curated changes in the rate of error reduction which produce the primary source of aesthetic pleasure for the error minimizing mind.
Day 2: 12h40 to 13h20
Pain & Suffering: an active inference perspective
The experience of pain fundamentally alters an individual’s orientation in the world. Anticipating a loss of bodily integrity, the individual in pain is afforded a narrowed repertoire of action (Tabor & Burr, 2019). On the one hand, this constraint presents an opportunity to reduce the likelihood of potential bodily harm, on the other, it risks compromising the ability to attune to a changing environment (Ramstead, Kirchhoff & Friston, 2019). In applying a predictive account to the experience of pain, the concept of suffering will be discussed in relation to constrained action. Here, the experience of pain is considered both embodied and embedded (Tabor, Keogh & Eccleston, 2018). In this talk, I will draw on persistent experiences, particularly focussing on pain, to explore the mechanisms underpinning associated suffering.
Day 2: 13h20 to 14h
José M. Araya
Following your guts: interoceptive expectations and the loops of emotion
Emotion researchers tend to account for emotion generation by drawing an analogy between the mechanisms and dynamics of perceptual experiences in the visual modality and emotion. This approach captures many aspects of emotion. However, I argue that this approach fails to account for the motivational character of emotion and its dynamics of self-organisation. I suggest that emotion generation is rather a form of regulatory action analogous to drives such as ‘air hunger’, affective touch, and itch.
Regina Fabry is a philosopher of mind and cognition. She works on 4E cognition, enculturation, and predictive processing. Her research focuses on narrative practices, literacy, sense of self, mental disorders, mind-wandering, and mathematical cognition. Regina currently holds a Lecturer position in the Department of Philosophy II at the Ruhr University Bochum and is a member of the interdisciplinary Research Training Group Situated Cognition (Osnabrück University & Ruhr University Bochum). She is the recipient of an ARC Discovery Early Career Research Award for her project Living to tell, telling to live: Experience, narrative and the self (Macquarie University, 2021-2024). For more information, see http://www.reginaefabry.de
Daphne is a postgraduate student of machine learning at Imperial College London. Her interest lies at the intersection between mathematics and neuroscience; she has been actively involved in research in theoretical models of emotion under the Free Energy Principle, and is now developing modeling skills for work on multi-scale representations of opinion dynamics with Markov blankets and active inference.
Slawa Loev received his PhD from the Institut Jean Nicod (ENS, CNRS) in Paris and is now a postdoctoral researcher at the Cognition, Value, Behaviour group (CVBE) at the LMU Munich. His research is situated at the crossroads of philosophy of mind and cognitive science, with a particular focus on emotion, intuition and metacognition.
Pablo Fernández is pursuing his research on the phenomenology of disorientation at Institut Jean Nicod (ENS, EHESS, CNRS), and he is a visitor at the Spatial Cognition Lab of University College London. The focus of his work is on how space structures human experience. He specialises in spatial cognition and in phenomenology. He follows an interdisciplinary approach, actively collaborating with neuroscientists, architects, geographers and anthropologists.
Mark Miller is a philosopher of cognition. His recent research investigates the implications of a leading new perspective on cognition, which conceptualizes the mind as an engine of knowledge-driven predictions (i.e., the “predictive mind”). Drawing on theoretical insights from active inference and the free-energy principle Mark aims to develop new perspectives on the nature of human well-being in our socio-technological niche. Mark is currently a part of Andy Clark's 4-year ERC project Expecting Ourselves: Embodied Prediction and the Construction of Conscious Experience ( and is a newly appointed member of the University of Hokkaido's Center for Human Nature, Artificial INtelligence and Neuroscience (
Abby Tabor is a behavioural scientist, focussing on persistent aversive experiences from an embodied perspective. She is currently a Wallscourt Fellow in Mental Health at the University of the West of England.
José M. Araya
José M. Araya works at the intersection of philosophy and the affective sciences. He obtained his PhD in Philosophy at the University of Edinburgh, and he is currently a postdoctoral researcher (Fondecyt) at Instituto de Filosofía y Ciencias de la Complejidad (IFICC, Chile). His current research aims to develop an account of self-control by integrating insights from predictive processing, cybernetics, and complexity sciences.