LSLars Sandved-Smith

Lars Sandved-Smith

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Lars Sandved-Smith has a total of 45 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Computational Mechanisms of Gratitude Practice

    by Elizabeth L. Fisher, Lars Sandved Smith, Jakob Hohwy

    Positive psychology interventions, such as gratitude practice, aim to improve well-being through simple techniques. One widely used practice involves reflecting daily on three things one is grateful for. Although this practice is promising, empirical evidence for its efficacy is mixed, and the underlying mechanisms remain poorly understood. Here, we develop a computational model of gratitude practice to gain insight into the potential mechanisms of the practice. Employing the active inference framework, we present three simulation studies. First, we formalize gratitude practice as the deliberate allocation of high precision (i.e., attention) to three positive observations. This allows us to show how agents form beliefs about their environment and how these beliefs shift following gratitude practice. Second, to link the model to empirical findings, we simulate an optimism assessment task before and after the intervention, demonstrating an increase in optimism following gratitude practice. Finally, we simulate an abstract proof-of-concept task inspired by the Yarbus paradigm to model attention and find that gratitude practice may affect habitual attention patterns, where agents attend to positive elements over neutral or negative elements in a painting. Our model provides a conceptual understanding of gratitude practice that can be used in future research to gain insights into who may benefit from gratitude practice.

  • Open Access

    There is no self-evidence: A physics of emptiness realisation

    by Lars Sandved-Smith, Chris Fields, Thomas Doctor, Ruben Eero Laukkonen, Jakob Hohwy

    Any system that persists must minimise surprisal, thereby gathering evidence for its own generative model, a process known as self-evidencing. However, one dimension of self-evidencing is provably impossible: evidencing the boundary that would constitute the agent as an entity separate from their environment. Recent results in quantum information theory demonstrate that no finite system can measure the entanglement entropy across its own boundary, rendering the separability of agent from environment permanently unevidenceable. Subjectively therefore, there can be no evidence for a self separate from the world. We propose that the Buddhist notion of awakening, in the sense of stable realisation of emptiness, can be understood as the embodied recognition of this impossibility. We formalise the belief in separation as a structural prior over an agent's quantum reference frame (QRF) deployments, constraining all measurement frames to respect a self/environment partition. We describe how contemplative practice progressively opacifies this constraint by developing a model of the agent's own QRF dynamics, revealing the partition as a contingent modelling choice rather than a given feature of reality. Once visible, the prior is eliminated via Bayesian model reduction resulting in a post-dual agent with unconstrained QRF deployments. We argue that such priors suppress the inherent contextuality of the boundary and that their removal constitutes a formal counterpart to the Buddhist notion of emptiness realisation. Self-evidencing may continue unimpeded after this transition, grounded in structural realism about causal regularities, while the ontological commitment to a bounded self is relinquished. We discuss empirical predictions, including altered dynamical regimes in neural systems, and the formal relationship between emptiness realisation and compassion.

  • Open Access

    Computational spirits: a neuroscientific account of psychedelic entity encounters

    by Jonas Mago, George Deane, Lars Sandved-Smith, Adam Safron, Christopher Timmermann, Maxwell Ramstead, David Dupuis, Robin Carhart-Harris, Kyle T Greenway, Michael Lifshitz

    Under the influence of psychedelics, people often report encountering “entities” who seem to have their own autonomous agency. Depending on the cultural milieu, these entities are reported to take a variety of forms, including spirits, elves, ancestors, or fragments of the self. Encounters with such beings hold a central place in many traditions of psychedelic practice around the world. And yet, mechanistic accounts of these experiences are scarce in the neuroscientific literature. Here, we propose a neurocomputational model to account for experiences of entities, focusing primarily on those occasioned by the serotonergic psychedelic N,N-dimethyltryptamine. Our model builds on earlier theoretical accounts, including the entropic brain model of psychedelics, computational accounts of the felt presence of other minds, and theories of self-other discrimination based on sensory attenuation. We synthesize and expand on these perspectives through an overarching physics-based approach to cognition and brain function—the active inference framework. We propose that the general effects of psychedelics on large-scale neural dynamics may shape the way the brain comes to infer and interpret agentic presences. In particular, the reduction in the predictability of sensory perceptions during the psychedelic state may incline the brain to interpret perceptions, both internal and external, as resulting from non-self-agentic sources. In specifying the neurocomputational mechanisms, our model aims to explain how the brain supports entity encounters while also accounting for the diversity (and similarity) of these experiences across cultural contexts.

  • Open Access

    Computational mechanisms of gratitude practice

    by Elizabeth L Fisher, Lars Sandved-Smith, Jakob Hohwy

    Positive psychology interventions, such as gratitude practice, claim to substantially improve an individual’s well-being. By focusing on three things one is grateful for each day, gratitude practice may be a promising simple and scalable intervention to support well-being. However, there is mixed evidence of the efficacy of gratitude practice and we lack a mechanistic understanding of the underlying processes. Here we develop a computational model of gratitude practice to gain insight into the potential mechanisms of the practice. Employing the active inference framework, we present three simulation results. The first is an instantiation of gratitude practice as the deliberate deployment of high precision (i.e. attention) of three good observations throughout the day. We demonstrate theoretically how agents form their beliefs about their world from the environment they are in, and how these beliefs change positively after gratitude practice. To understand the impact on the agent’s worldview and to connect the modelling results to the empirical literature, we simulate an optimism assessment task before and after the intervention, showing an increase in optimism after gratitude practice. Finally, we simulate a modified version of the Yarbus eye movement task and find that gratitude practice may affect habitual attention patterns, such that agents attend to more positive elements compared to neutral or negative elements in a painting after gratitude practice. Our model provides a conceptual understanding of gratitude practice that can be used in future research to gain insights into who may benefit from gratitude practice, and why some individuals may not.

  • Open Access

    Active Inference, Computational Phenomenology, and Advanced Meditation: Toward the Formalization of the Experience of Meditation

    by Hagar Tal, Malcolm Wright, Shawn Prest, Lars Sandved-Smith, Matthew Sacchet

    Computational phenomenology has emerged as a powerful framework for investigating advanced meditation states and stages, and meditative development and endpoints. Various models have been proposed to mechanistically explain the diverse experiences associated with these practices, including enhanced well-being, attentional shifts, and defabrication, as well as minimal phenomenal experiences and meditative endpoints. However, these models have developed in disparate directions, and an integrative understanding of the underlying mechanisms of advanced meditation remains elusive. This review examines how computational models attempt to account for the phenomenology of advanced meditation, with a particular focus on Active Inference as a modeling framework. We find that a key point of convergence across models is the role of precision weighting as a driver of experiential shifts. Furthermore, we observe a marked difference between early models, which emphasize top-down attentional modulation toward interoception or specific focus objects, and later models which center on layer-specific precision re-weighting within the meditator’s hierarchical generative model. These differences arise from variations in the models' aims, scope, and definitions of contemplative practice. Few models address reported increases in cognitive flexibility and learning from meditation, fundamental mechanisms behind informal practice and affective processes, as well as processes underlying compassion traditions, remain underexplored. Addressing these gaps is crucial for refining computational models of advanced meditation and informing our understanding of its cognitive, affective, and experiential effects.


  • Top Cited Papers
    Open Access

    Deep computational neurophenomenology: a methodological framework for investigating the how of experience

    by Lars Sandved-Smith, Juan Diego Bogotá, Jakob Hohwy, Julian Kiverstein, Antoine Lutz

    The context for our paper comes from the neurophenomenology (NPh) research programme initiated by Francisco Varela at the end of the 1990s. Varela’s working hypothesis was that, to be successful, a consciousness research programme must progress by relating first-person phenomenological accounts of the structure of experience and their third-person counterparts in neuroscience through “mutual constraints”. Leveraging Bayesian mechanics, in particular deep parametric active inference, we demonstrate the potential for epistemically advantageous mutual constraints between phenomenological, computational, behavioural, and physiological vocabularies. Specifically, the dual information geometry of Bayesian mechanics serves to establish, under certain conditions, generative passage between lived experience and its physiological instantiation. This paper argues for the epistemological necessity of such a passage and the inclusion of trained reflective awareness in neurophenomenological empirical approaches. In particular, it showcases incremental explanatory gains for the scientist that arise from incorporating the participants’ epistemic insights, shifting the focus from the contents of experience (i.e. what a subject experiences in a given experimental set-up) to the how of experience (i.e. the activities of consciousness that allow for a meaningful world to appear to us as such in lived experience). The explanatory power of the resulting ‘meta-Bayesian’ framework, deep computational NPh, arises from the disciplined circulation between first and third-person perspectives enabled by the formalism of deep parametric active inference, where parametric depth refers to a property of generative models that can form beliefs about the parameters of their own modelling process. Hence, this computational formalism contributes to understanding consciousness by bridging phenomenological descriptions and physiological instantiations, whilst also highlighting the significance of trained first-person investigation in experimental protocols.


  • Open Access

    A Computational Model of Minimal Phenomenal Experience (MPE)

    by Lars Sandved-Smith

    Minimal phenomenal experience (MPE), or "pure consciousness," represents a fundamental form of conscious experience characterised by reflexive meta-awareness and the absence of many features of regular phenomenology. It has been described as e.g. non-conceptual, atemporal, non-egoic and aperspectival. This paper aims to develop a computational model of MPE using the mathematics of variational free energy minimization derived from the free energy principle (FEP). I employ a computational neurophenomenology approach, formalising key phenomenological features of MPE within the active inference framework. The model incorporates parametric depth, allowing for higher-order inferences about generative model parameters. I relate specific model parameterisations to reported MPE qualities such as meta-awareness, equanimity, effortlessness, and non-conceptuality. The proposed model suggests that MPE arises when an agent achieves very low free energy through self-directed awareness and modulation of their generative model, particularly by emphasising awareness of awareness itself. The model predicts elements of MPE phenomenology including a sense of effortlessness, timelessness, and the potential for a "zero-person perspective”. The implementation details for a simulation of the proposed model are outlined, as well as directions for empirical validation.


  • Open Access

    Making the Thermodynamic Cost of Active Inference Explicit

    by Chris Fields, Adam Goldstein, Lars Sandved-Smith

    When describing Active Inference Agents (AIAs), the term “energy” can have two distinct meanings. One is the energy that is utilized by the AIA (e.g., electrical energy or chemical energy). The second meaning is so-called Variational Free Energy (VFE), a statistical quantity which provides an upper bound on surprisal. In this paper, we develop an account of the former quantity—the Thermodynamic Free Energy (TFE)—and its relationship with the latter. We highlight the necessary tradeoffs between these two in a generic, quantum information-theoretic formulation, and the macroscopic consequences of those tradeoffs for the ways that organisms approach their environments. By making this tradeoff explicit, we provide a theoretical basis for the different metabolic strategies that organisms from plants to predators use to survive.


  • Open Access

    Deep computational neurophenomenology: A methodological framework for investigating the how of experience

    by Lars Sandved-Smith, Juan Diego Bogotá, Jakob Hohwy, Julian Kiverstein, Antoine Lutz

    The context for our paper comes from the neurophenomenology research program initiated by Francisco Varela at the end of the 1990s. Varela’s working hypothesis was that, to be successful, a consciousness research program must progress by relating first-person phenomenological accounts of the structure of experience and their third-person counterparts in neuroscience through reciprocal or mutual constraints. Leveraging Bayesian mechanics, in particular deep parametric active inference, we demonstrate the potential for epistemically advantageous mutual constraints between phenomenological, computational, behavioural and physiological vocabularies. Specifically, the dual information geometry of Bayesian mechanics serves to establish, under certain conditions, generative passages between lived experience and its physiological instantiation. This paper argues for the epistemological necessity of such passages and the inclusion of trained reflective awareness in neurophenomenological empirical approaches, showcasing incremental epistemic gains that come from by shifting the focus from the contents of experience (i.e. what a subject experiences in a given experimental set-up) to the how of experience - the activities of consciousness that allow for a meaningful world to appear to us as such in lived experience. The explanatory power of the resulting framework, deep computational neurophenomenology, arises from the disciplined circulation between first and third-person perspectives enabled by the formalism of deep parametric active inference, where parametric depth refers to a property of generative models that can form beliefs about the parameters of their own modelling process. Hence, this computational formalism contributes to understanding consciousness by bridging phenomenological descriptions and physiological instantiations, whilst also highlighting the significance of trained first person investigation in experimental protocols.


  • Open Access

    The computational unconscious: Adaptive narrative control, psychopathology, and subjective well-being

    by George Deane, Jonas Mago, Aikaterini Fotopoulou, Matthew Sacchet, Robin Carhart-Harris, Lars Sandved-Smith

    This paper introduces the notion of adaptive narrative control, a conception of how subpersonal computational processes shape the contents of conscious experience to realize adaptive behavior. We unpack the implications of the theory for understanding the computational mechanisms underwriting psychopathology and improvements in subjective well-being associated with psychedelic therapy and meditation. The core idea of adaptive narrative control is that systems equipped with an ‘attention schema’ — a model of its own attentional states and how attentional states can be controlled — can come to anticipate not only the epistemic implications of certain attentional states, but also the pragmatic consequences, such as how certain attentional states potentiate certain affective responses. In anticipating affective states, the system is able to regulate affective states through ‘mental action’ — the endogenous control of attention. We argue that using mental action to bias the sampling of evidence to control the ‘narrative’ — the upshot of inference understood to correspond to the contents of conscious experience — allows the system to regulate affective and physiological states in ways that potentiate adaptive behavior. However, it is this adaptive capacity which gives rise to the computational mechanism — ‘avoidant mental action’, or equivalently ‘motivated inattention’ — which we argue is a core mechanism underlying psychopathology. We unpack this approach within the active inference framework to provide specification of the candidate mechanisms. We argue this conception can be used to account for the rigid belief formation characterizing ‘canalization’ and show how the decrements in subjective well-being come as a consequence of reduced recognition and categorisation of emotions (i.e., alexithymia or impaired emotional granularity). We argue that while avoidant mental action facilitates adaptive behavior, certain environmental conditions can lead it to resulting in decreased subjective well-being and psychopathology. Our account partially echoes a Freudian perspective on the function and effects of avoidant defense mechanisms like repression, and brings into view a novel computational conception of the dynamic unconscious— the ‘computational unconscious’. Finally, we explore how this conceptualisation can expand and refine the ‘REBUS’ model of psychedelic action and therapy, and explain some of the increments in subjective well-being associated with meditation.