Figure 7 - uploaded by Joe Dewhurst
Content may be subject to copyright.
The Bayesian network described in Fig. 2, labelled with the Friston blanket notation. Example sensorimotor blankets for different (internal) states, í µí±¥ 10 and í µí±¥ 9 respectively, with labels removed for clarity. External states in lavender, sensory states in magenta, internal states in teal (í µí±¥ 10 in (a) and í µí±¥ 9 in (b)), active states in blue, and putative co-parents in mustard. Notice that the partitions obtained here do not map to the separation of internal states (hidden states í µí±¥ Fig. 2) and world states (observations í µí±¦ Fig. 2). This suggests that sensorimotor blankets are Markov blankets applied under a specific set of assumptions that cannot be traced to standard uses of Markov blankets in variational inference (Jordan et al. 1999; Fox and Roberts 2012).

The Bayesian network described in Fig. 2, labelled with the Friston blanket notation. Example sensorimotor blankets for different (internal) states, í µí±¥ 10 and í µí±¥ 9 respectively, with labels removed for clarity. External states in lavender, sensory states in magenta, internal states in teal (í µí±¥ 10 in (a) and í µí±¥ 9 in (b)), active states in blue, and putative co-parents in mustard. Notice that the partitions obtained here do not map to the separation of internal states (hidden states í µí±¥ Fig. 2) and world states (observations í µí±¦ Fig. 2). This suggests that sensorimotor blankets are Markov blankets applied under a specific set of assumptions that cannot be traced to standard uses of Markov blankets in variational inference (Jordan et al. 1999; Fox and Roberts 2012).

Source publication
Article
Full-text available
[A heavily rewritten version of this paper has been published in BBS in 2021] Markov blankets have been used to settle disputes central to philosophy of mind and cognition. Their development from a technical concept in Bayesian inference to a central concept within the free-energy principle is analysed. We propose to distinguish between instrumenta...

Contexts in source publication

Context 1
... graphical model without first making some additional assumptions. For instance, here we will label a node, say í µí±¥ 10 from Fig. 2, as an 'internal' state (signified by a teal colour as in Fig. 6), which is conditionally separated from all the remaining 'external' variables (in lavender) by a set of nodes constituting its Friston blanket (see Fig. 7a). Following Friston's proposal, the nodes in this Friston blanket can be further separated into 'sensory' (magenta) and 'active' (blue) states, the former corresponding to the parents of the internal state (i.e., node í µí±¥ 10 ) and the latter including its children. Picking a different node, such í µí±¥ 9 , to be labelled 'internal' ...
Context 2
... in this Friston blanket can be further separated into 'sensory' (magenta) and 'active' (blue) states, the former corresponding to the parents of the internal state (i.e., node í µí±¥ 10 ) and the latter including its children. Picking a different node, such í µí±¥ 9 , to be labelled 'internal' generates a correspondingly different blanket (see Fig. 7b). It is clear here that the Friston blanket does 13 Notably, the diagram in Fig. 6 is not a Bayesian network, nor is it intended to approximate one. However, even though the sensorimotor loop assumes cycles (due to both the bidirectional connections between different components and the overall circular causality imposed by the very ...
Context 3
... interested in studying. This issue can be demonstrated by the fact that under the Friston blanket formulation, the location of a node in the graph layout is not sufficient to identify whether or not it is an 'internal' state in Friston's sense (recall that we had to start by arbitrarily selecting an internal state when formulating the graphs in Fig. 7). If we decide to interpret the active/sensory distinction by exclusively following Friston's use of the sensorimotor analogy, we will soon realise that using a Friston blanket as if it was a Pearl blanket falls short of explaining (and in some cases may be directly inconsistent with) the general identification of active and sensory ...
Context 4
... that we presented in 4.2 presents a very different research direction for the FEP and active inference framework. This simulation starts out with a soup of coupled particles and aims to show how a distinction between 'agent' and 'environment' emerges naturally as the dynamics of the system reach equilibrium. Here we will use the example of Fig. 7, where the lavender nodes represent external states, coupled to a set of internal nodes which, under the Friston blanket interpretation, is claimed to imply that the system represented by these internal nodes is an agent performing inference on the external states (Friston 2013b; Friston 2019). Effectively, this Bayesian network ...
Context 5
... graphical model without first making some additional assumptions. For instance, here we will label a node, say í µí±¥ 10 from Fig. 2, as an 'internal' state (signified by a teal colour as in Fig. 6), which is conditionally separated from all the remaining 'external' variables (in lavender) by a set of nodes constituting its Friston blanket (see Fig. 7a). Following Friston's proposal, the nodes in this Friston blanket can be further separated into 'sensory' (magenta) and 'active' (blue) states, the former corresponding to the parents of the internal state (i.e., node í µí±¥ 10 ) and the latter including its children. Picking a different node, such í µí±¥ 9 , to be labelled 'internal' ...
Context 6
... in this Friston blanket can be further separated into 'sensory' (magenta) and 'active' (blue) states, the former corresponding to the parents of the internal state (i.e., node í µí±¥ 10 ) and the latter including its children. Picking a different node, such í µí±¥ 9 , to be labelled 'internal' generates a correspondingly different blanket (see Fig. 7b). It is clear here that the Friston blanket does 13 Notably, the diagram in Fig. 6 is not a Bayesian network, nor is it intended to approximate one. However, even though the sensorimotor loop assumes cycles (due to both the bidirectional connections between different components and the overall circular causality imposed by the very ...
Context 7
... interested in studying. This issue can be demonstrated by the fact that under the Friston blanket formulation, the location of a node in the graph layout is not sufficient to identify whether or not it is an 'internal' state in Friston's sense (recall that we had to start by arbitrarily selecting an internal state when formulating the graphs in Fig. 7). If we decide to interpret the active/sensory distinction by exclusively following Friston's use of the sensorimotor analogy, we will soon realise that using a Friston blanket as if it was a Pearl blanket falls short of explaining (and in some cases may be directly inconsistent with) the general identification of active and sensory ...
Context 8
... that we presented in 4.2 presents a very different research direction for the FEP and active inference framework. This simulation starts out with a soup of coupled particles and aims to show how a distinction between 'agent' and 'environment' emerges naturally as the dynamics of the system reach equilibrium. Here we will use the example of Fig. 7, where the lavender nodes represent external states, coupled to a set of internal nodes which, under the Friston blanket interpretation, is claimed to imply that the system represented by these internal nodes is an agent performing inference on the external states (Friston 2013b; Friston 2019). Effectively, this Bayesian network ...
Context 9
... graphical model without first making some additional assumptions. For instance, here we will label a node, say í µí±¥ 10 from Fig. 2, as an 'internal' state (signified by a teal colour as in Fig. 6), which is conditionally separated from all the remaining 'external' variables (in lavender) by a set of nodes constituting its Friston blanket (see Fig. 7a). Following Friston's proposal, the nodes in this Friston blanket can be further separated into 'sensory' (magenta) and 'active' (blue) states, the former corresponding to the parents of the internal state (i.e., node í µí±¥ 10 ) and the latter including its children. Picking a different node, such í µí±¥ 9 , to be labelled 'internal' ...
Context 10
... in this Friston blanket can be further separated into 'sensory' (magenta) and 'active' (blue) states, the former corresponding to the parents of the internal state (i.e., node í µí±¥ 10 ) and the latter including its children. Picking a different node, such í µí±¥ 9 , to be labelled 'internal' generates a correspondingly different blanket (see Fig. 7b). It is clear here that the Friston blanket does 13 Notably, the diagram in Fig. 6 is not a Bayesian network, nor is it intended to approximate one. However, even though the sensorimotor loop assumes cycles (due to both the bidirectional connections between different components and the overall circular causality imposed by the very ...
Context 11
... interested in studying. This issue can be demonstrated by the fact that under the Friston blanket formulation, the location of a node in the graph layout is not sufficient to identify whether or not it is an 'internal' state in Friston's sense (recall that we had to start by arbitrarily selecting an internal state when formulating the graphs in Fig. 7). If we decide to interpret the active/sensory distinction by exclusively following Friston's use of the sensorimotor analogy, we will soon realise that using a Friston blanket as if it was a Pearl blanket falls short of explaining (and in some cases may be directly inconsistent with) the general identification of active and sensory ...
Context 12
... that we presented in 4.2 presents a very different research direction for the FEP and active inference framework. This simulation starts out with a soup of coupled particles and aims to show how a distinction between 'agent' and 'environment' emerges naturally as the dynamics of the system reach equilibrium. Here we will use the example of Fig. 7, where the lavender nodes represent external states, coupled to a set of internal nodes which, under the Friston blanket interpretation, is claimed to imply that the system represented by these internal nodes is an agent performing inference on the external states (Friston 2013b; Friston 2019). Effectively, this Bayesian network ...

Citations

... The partitioning blanket states map onto the organism's modes of interaction so that sensory states are associated with sensory receptor activity, and active states are associated with the system's influence on its environment, such as action. It remains a current debate to what extent this application of the Markov blanket should be taken literally or instrumentally (van Es 2020; Bruineberg et al. 2020;Hohwy 2016). In this paper, we will remain neutral in this debate, and instead explore only what can be done within the formalism, regardless of how it may or may not be implemented in any real system. ...
... 11 Contrary to, say, Friston (2013), the condition of a system being at NESS with its environment seems to have replaced the initial clause of being locally ergodic (see Friston 2019; but also Hipólito 2020; Ramstead, Badcock et al. 2019) . Discussion of this change and its philosophical implications are outside of the scope of this paper, but see Bruineberg et al. (2020) for preliminary discussions. any Markov blanketed system that is at NESS with its environment can be cast as minimizing free energy ). ...
Article
Full-text available
The free energy principle (FEP) purports to provide a single principle for the organizational dynamics of living systems, including their cognitive profiles. It states that for a system to maintain non-equilibrium steady-state with its environment it must minimise its free energy. It is said to be entirely scale-free, applying to anything from particles to organisms, and interactive machines, spanning from the abiotic to the biotic. Because the FEP is so general in its application, it is for this reason that one might wonder in what sense this framework captures anything specific to biological characteristics, if details at all. We take steps to correct for this here. We do so by taking up a distinct challenge that the FEP must overcome if it is to be of interest to those working in the biological sciences. We call this the pebble challenge: it states that the FEP cannot capture the organisational principles specific to biology, for its formalisms apply equally well to pebbles. We progress in solving the pebble challenge by articulating how the notion of ‘autonomy as precarious operational closure’ from the enactive literature can be unpacked within the FEP. This enables the FEP to delineate between the abiotic and the biotic; avoiding the pebble challenge that keeps it out of touch with the living systems we encounter in the world and is of interest to the sciences of life and mind.
... It simply draws attention to the fact that there are different perspectives on ontological implications of FEP, and not all interpretations are in line with a realist characterisation of FEP. We shall attend to the role of Markov blankets in the statement of the FEP-based account of consciousness soon, but as we will shortly see (with reference to (Bruineberg et al. 2020)), the characterisation of Markov blankets too is prone to diverse ontological interpretations. Non-realist interpretations of Markov blankets, when substantiated adequately, provide grounds for scepticism about Markovian monism, which projects properties of the Markovian models to their target systems. ...
... It is also worth mentioning that there is a disparity between various ontological interpretations of Markov blankets (this is similar and related to the aforementioned disparity between realist and non-realist interpretations of FEP). I draw on Bruineberg et al. (2020) recent study of Markov blankets to pinpoint this disparity. Bruineberg et al. (2020) remarked the divergence between ontic interpretations of Markov blankets in terms of the difference between Pearl blankets and Friston blankets. ...
... I draw on Bruineberg et al. (2020) recent study of Markov blankets to pinpoint this disparity. Bruineberg et al. (2020) remarked the divergence between ontic interpretations of Markov blankets in terms of the difference between Pearl blankets and Friston blankets. Pearl blankets as being presented by Judea Pearl (1988), are powerful formal modelling tools that can regiment conditional independencies/separations in Bayesian networks (of considerable size). ...
Article
Full-text available
Free Energy Principle underlies a unifying framework that integrates theories of origins of life, cognition, and action. Recently, FEP has been developed into a Markovian monist perspective (Friston et al. in BC 102: 227–260, 2020). The paper expresses scepticism about the validity of arguments for Markovian monism. The critique is based on the assumption that Markovian models are scientific models, and while we may defend ontological theories about the nature of scientific models, we could not read off metaphysical theses about the nature of target systems (self-organising conscious systems, in the present context) from our theories of nature of scientific models (Markov blankets). The paper draws attention to different ways of understanding Markovian models, as material entities, fictional entities, and mathematical structures. I argue that none of these interpretations contributes to the defence of a metaphysical stance (either in terms of neutral monism or reductive physicalism). This is because scientific representation is a sophisticated process, and properties of Markovian models—such as the property of being neither physical nor mental—could not be projected onto their targets to determine the ontological properties of targets easily.
... For example, Hohwy's claim that the blanket is best placed around the brain is a conditional arrangement related to, 46 This means that the understanding of Markov's blankets in PP and the active inference framework is far beyond Pearl's approach, in which there is no separation into active and passive states. 47 It should be noticed that there is a divergence between ontic interpretations of Markov blankets in terms of Pearl "instrumental" blankets and Friston "realist" blankets (Bruineberg et al., 2020). Pearl interpretation does not go beyond pure formalism and Markov blankets themselves turn out to be a mathematical construct used for making inferences about the e.g. ...
... living organisms). Such an approach is not "philosophically innocent", as convincingly demonstrated by Bruineberg et al., 2020. among others, the explanatory strategy, the research interests, and the phenomenon under study. ...
Article
Full-text available
The purpose of this paper is to provide a systematic review of the Predictive Processing framework (hereinafter PP) and to identify its basic theoretical difficulties. For this reason, it is, primarily, polemic-critical and, secondarily, historical. I discuss the main concepts, positions and research issues present within this framework (§1-2). Next, I present the Bayesian-brain thesis (§3) and the difficulty associated with it (§4). In §5, I compare the conservative and radical approach to PP and discuss the internalist nature of the generative model in the context of Markov blankets. The possibility of linking PP with the free energy principle (hereinafter FEP) and the homeostatic nature of predictive mechanisms is discussed in §6. This is followed by the presentation of PP's difficulties with solving the dark room problem and the exploration-exploitation trade-off (§7). I emphasize the need to integrate PP with other models and research frameworks within cognitive science. Thus, this review not only discusses PP, but also provides an assessment of the condition of this research framework in the light of the hopes placed on it by many researchers. The Conclusions section discuss further research challenges and the epistemological significance of PP.
... Since writing this paper, we have been encouraged by the enthusiasm with which these issues have been discussed in the literature. For readers interested in delving further into these exchanges, recent papers include [46][47][48][49]. In concluding, we would like to thank Biehl et al. for a thorough and useful deconstruction of [2]. ...
Article
Full-text available
Biehl et al. (2021) present some interesting observations on an early formulation of the free energy principle. We use these observations to scaffold a discussion of the technical arguments that underwrite the free energy principle. This discussion focuses on solenoidal coupling between various (subsets of) states in sparsely coupled systems that possess a Markov blanket—and the distinction between exact and approximate Bayesian inference, implied by the ensuing Bayesian mechanics.
... The Free Energy Principle (FEP) (Friston, 2019a;Friston & Ao, 2012a;Friston, Kilner, & Harrison, 2006; is an emerging theory in theoretical neuroscience which aims to tackle an extremely deep and fundamental question -can one characterise necessary behaviour of any system that maintains a statistical separation from its environment (Bruineberg, Dolega, Dewhurst, & Baltieri, 2020;Friston, 2019a;? Specifically, it argues that any such system can be seen as performing an elemental kind of Bayesian inference where the dynamics of the internal states of such a system can be interpreted as minimizing a variational free energy functional (Beal, 2003) 1 , and thus performing approximate (variational) ...
Preprint
In this PhD thesis, we explore and apply methods inspired by the free energy principle to two important areas in machine learning and neuroscience. The free energy principle is a general mathematical theory of the necessary information-theoretic behaviours of systems that maintain a separation from their environment. A core postulate of the theory is that complex systems can be seen as performing variational Bayesian inference and minimizing an information-theoretic quantity called the variational free energy. The thesis is structured into three independent sections. Firstly, we focus on predictive coding, a neurobiologically plausible process theory derived from the free energy principle which argues that the primary function of the brain is to minimize prediction errors, showing how predictive coding can be scaled up and extended to be more biologically plausible, and elucidating its close links with other methods such as Kalman Filtering. Secondly, we study active inference, a neurobiologically grounded account of action through variational message passing, and investigate how these methods can be scaled up to match the performance of deep reinforcement learning methods. We additionally provide a detailed mathematical understanding of the nature and origin of the information-theoretic objectives that underlie exploratory behaviour. Finally, we investigate biologically plausible methods of credit assignment in the brain. We first demonstrate a close link between predictive coding and the backpropagation of error algorithm. We go on to propose novel and simpler algorithms which allow for backprop to be implemented in purely local, biologically plausible computations.
... Presumably, single, well identified, regions of the cortical hierarchy.34 This issue seems to me importantly related to the "having VS. being" a model in the literature on the free energy principle (seevan Es, 2020;Baltieri et al., 2020; see alsoBruineberg et al., 2020). I must confess, however, that I'm unsure about how to properly articulate such a relation.Content courtesy of Springer Nature, terms of use apply. ...
Article
Full-text available
Philosophers interested in the theoretical consequences of predictive processing often assume that predictive processing is an inferentialist and representationalist theory of cognition. More specifically, they assume that predictive processing revolves around approximated Bayesian inferences drawn by inverting a generative model. Generative models, in turn, are said to be structural representations: representational vehicles that represent their targets by being structurally similar to them. Here, I challenge this assumption, claiming that, at present, it lacks an adequate justification. I examine the only argument offered to establish that generative models are structural representations, and argue that it does not substantiate the desired conclusion. Having so done, I consider a number of alternative arguments aimed at showing that the relevant structural similarity obtains, and argue that all these arguments are unconvincing for a variety of reasons. I then conclude the paper by briefly highlighting three themes that might be relevant for further investigation on the matter.
... We find that, in the class of linear systems explored, the answer to this question is that the statistical structure required by the FEP only arises in a very narrow class of systems, requiring stringent conditions such as fully symmetric agent-environment interactions that we can in general not expect from systems displaying agency [9]. The generality of the FEP has been questioned in the past due to conceptual issues [22,23], or the existence of counterexamples challenging that sensorimotor interfaces, Markov blankets and solenoidal decoupling follow from each other [11]. However, our study is the first to our knowledge that shows that the assumptions of the FEP do not hold for a very broad class of systems, namely linear, weakly coupled systems, except for the limited case of fully symmetric agent-environment interaction. ...
Preprint
The Free Energy Principle (FEP) states that any dynamical system can be interpreted as performing Bayesian inference upon its surrounding environment. Although the FEP applies in theory to a wide variety of systems, there has been almost no direct exploration of the principle in concrete systems. In this paper, we examine in depth the assumptions required to derive the FEP in the simplest possible set of systems - weakly-coupled non-equilibrium linear stochastic systems. Specifically, we explore (i) how general are the requirements imposed on the statistical structure of a system and (ii) how informative the FEP is about the behaviour of such systems. We find that this structure, known as a Markov blanket (i.e. a boundary precluding direct coupling between internal and external states) and stringent restrictions on its solenoidal flows, both required by the FEP, make it challenging to find systems that fulfil the required assumptions. Suitable systems require an absence of asymmetries in sensorimotor interactions that are highly unusual for living systems. Moreover, we find that a core step in the argument relating the behaviour of a system to variational inference relies on an implicit equivalence between the dynamics of the average states with the average of the dynamics. This equivalence does not hold in general even for linear systems as it requires an effective decoupling from the system's history of interactions. These observations are critical for evaluating the generality and applicability of the FEP and point to potential issues in its current form. These issues make the FEP, as it stands, not straightforwardly applicable to the simple linear systems studied here and suggests more development is needed before it could be applied to the kind of complex systems which describe living and cognitive processes.
... And as such, there is no additional information about internal state behaviour that would be gained from knowledge of external states. 7 See Bruineberg et al. (2020) regarding the current debate on whether or not Markov blankets-what they suggest calling 'Friston Blankets'-are best understood instrumentally or ontologically. 8 It is important to note that the Markov blanket formalism is scale-free. ...
Article
Full-text available
Predictive processing theories are increasingly popular in philosophy of mind; such process theories often gain support from the Free Energy Principle (FEP)—a normative principle for adaptive self-organized systems. Yet there is a current and much discussed debate about conflicting philosophical interpretations of FEP, e.g., representational versus non-representational. Here we argue that these different interpretations depend on implicit assumptions about what qualifies (or fails to qualify) as representational. We deploy the Free Energy Principle (FEP) instrumentally to distinguish four main notions of representation, which focus on organizational, structural, content-related and functional aspects, respectively. The various ways that these different aspects matter in arriving at representational or non-representational interpretations of the Free Energy Principle are discussed. We also discuss how the Free Energy Principle may be seen as a unified view where terms that traditionally belong to different ontologies—e.g., notions of model and expectation versus notions of autopoiesis and synchronization—can be harmonized. However, rather than attempting to settle the representationalist versus non-representationalist debate and reveal something about what representations are simpliciter , this paper demonstrates how the Free Energy Principle may be used to reveal something about those partaking in the debate; namely, what our hidden assumptions about what representations are—assumptions that act as sometimes antithetical starting points in this persistent philosophical debate.
... From a biological perspective, for instance the plasmalemma is the natural Markov blanket of the neuron, able to make possible both the differentiation and the communication between the internal and external states (Cieri and Esposito, 2019;Ramstead et al., 2019). Currently there is an active and important discussion about whether this construct should be interpreted in a realistic way, applied to biological systems (Bruineberg et al., 2020;van Es and Hipólito, 2020;Beni, 2021). 3 For example, with reentrant interactions among widely distributed brain regions (Edelman, 1993). ...
Article
Full-text available
Neural complexity and brain entropy (BEN) have gained greater interest in recent years. The dynamics of neural signals and their relations with information processing continue to be investigated through different measures in a variety of noteworthy studies. The BEN of spontaneous neural activity decreases during states of reduced consciousness. This evidence has been showed in primary consciousness states, such as psychedelic states, under the name of “the entropic brain hypothesis.” In this manuscript we propose an extension of this hypothesis to physiological and pathological aging. We review this particular facet of the complexity of the brain, mentioning studies that have investigated BEN in primary consciousness states, and extending this view to the field of neuroaging with a focus on resting-state functional Magnetic Resonance Imaging. We first introduce historic and conceptual ideas about entropy and neural complexity, treating the mindbrain as a complex nonlinear dynamic adaptive system, in light of the free energy principle. Then, we review the studies in this field, analyzing the idea that the aim of the neurocognitive system is to maintain a dynamic state of balance between order and chaos, both in terms of dynamics of neural signals and functional connectivity. In our exploration we will review studies both on acute psychedelic states and more chronic psychotic states and traits, such as those in schizophrenia, in order to show the increase of entropy in those states. Then we extend our exploration to physiological and pathological aging, where BEN is reduced. Finally, we propose an interpretation of these results, defining a general trend of BEN in primary states and cognitive aging.
... And as such, there is no additional information about internal state behaviour that would be gained from knowledge of external states. 7 See Bruineberg et al. (2020) regarding the current debate on whether or not Markov blankets-what they suggest calling 'Friston Blankets'-are best understood instrumentally or ontologically. 8 It is important to note that the Markov blanket formalism is scale-free. ...
Article
Predictive processing theories are increasingly popular in philosophy of mind; such process theories often gain support from the Free Energy Principle (FEP) – a normative principle for adaptive self-organized systems. Yet there is a current and much discussed debate about conflicting philosophical interpretations of FEP, e.g., representational versus non-representational. Here we argue that these different interpretations depend on implicit assumptions about what qualifies (or fails to qualify) as representational. We deploy the Free Energy Principle (FEP) instrumentally to distinguish four main notions of representation, which focus on organizational, structural, content-related and functional aspects, respectively. The various ways that these different aspects matter in arriving at representational or non-representational interpretations of the Free Energy Principle are discussed. We also discuss how the Free Energy Principle may be seen as a unified view where terms that traditionally belong to different ontologies - e.g., notions of model and expectation versus notions of autopoiesis and synchronization - can be harmonized. However, rather than attempting to settle the representationalist vs non-representationalist debate and reveal something about what representations are simpliciter, this paper demonstrates how the Free Energy Principle may be used to reveal something about those partaking in the debate; namely, what our hidden assumptions about what representations are – assumptions that act as sometimes antithetical starting points in this persistent philosophical debate.