Joe Dewhurst

The 35 commentaries cover a wide range of topics and take many different stances on the issues explored by the target article. We have organised our response to the commentaries around three central questions: Are Friston blankets just Pearl blankets? What ontological and metaphysical commitments are implied by the use of Friston blankets? What kind of explanatory work are Friston blankets capable of? We conclude our reply with a short critical reflection on the indiscriminate use of both Markov blankets and the free energy principle.

The free energy principle, an influential framework in computational neuroscience and theoretical neurobiology, starts from the assumption that living systems ensure adaptive exchanges with their environment by minimizing the objective function of variational free energy. Following this premise, it claims to deliver a promising integration of the life sciences. In recent work, Markov blankets, one of the central constructs of the free energy principle, have been applied to resolve debates central to philosophy (such as demarcating the boundaries of the mind). The aim of this paper is twofold. First, we trace the development of Markov blankets starting from their standard application in Bayesian networks, via variational inference, to their use in the literature on active inference. We then identify a persistent confusion in the literature between the formal use of Markov blankets as an epistemic tool for Bayesian inference, and their novel metaphysical use in the free energy framework to demarcate the physical boundary between an agent and its environment. Consequently, we propose to distinguish between “Pearl blankets” to refer to the original epistemic use of Markov blankets and “Friston blankets” to refer to the new metaphysical construct. Second, we use this distinction to critically assess claims resting on the application of Markov blankets to philosophical problems. We suggest that this literature would do well in differentiating between two different research programmes: “inference with a model” and “inference within a model.” Only the latter is capable of doing metaphysical work with Markov blankets, but requires additional philosophical premises and cannot be justified by an appeal to the success of the mathematical framework alone.

Mechanism realists assert the existence of mechanisms as objective structures in the world, but their exact metaphysical commitments are unclear. We introduce Local Hierarchy Realism (LHR) as a substantive and plausible form of mechanism realism. The limits of LHR reveal a deep tension between two aspects of mechanists’ explanatory strategy. Functional decomposition identifies locally relevant entities and activities, while these same entities and activities are also embedded in a nested hierarchy of levels. In principle, a functional decomposition may identify entities engaging in causal interactions that crosscut the hierarchical structure of composition relations, violating the mechanist’s injunction against interlevel causation. We argue that this possibility is realized in the example of ephaptic coupling, a subsidiary process of neural computation that crosscuts the hierarchy derived from synaptic transmission. These considerations undermine the plausibility of LHR as a general view, yet LHR has the advantages that (i) its metaphysical implications are precisely stateable; (ii) the structure it identifies is not reducible to mere aggregate causation; and (iii) it clearly satisfies intuitive and informal definitions of mechanism. We conclude by assessing the prospects for a form of mechanism realism weaker than LHR that nevertheless satisfies all three of these requirements.

Structuralism about mathematical objects and structuralist accounts of physical computation both face indeterminacy objections. For the former, the problem arises for cases such as the complex roots i and \, for which a automorphism can be defined, thus establishing the structural identity of these importantly distinct mathematical objects. In the case of the latter, the problem arises for logical duals such as AND and OR, which have invertible structural profiles :369–400, 2001). This makes their physical implementations indeterminate, in the sense that their structural profiles alone cannot establish whether a given physical component is an AND-gate or an OR-gate. Doherty has recently shown both problems to be analogous, and has argued that computational structuralism is threatened with the absurd conclusion that computational digits might be indiscernible, such that, if structural properties are all that we have to go on, the binary digit 0 must be treated as identical to the binary digit 1. However, we think that a solution to the indiscernibility problem for mathematical structuralists, drawing on the work of David Hilbert, can be adapted for the analogous problem in the computational case, thereby rescuing the structuralist approach to physical computation.

Mechanism realists assert the existence of mechanisms as objective structures in the world, but their exact metaphysical commitments are unclear. We introduce Local Hierarchy Realism (LHR) as a substantive and plausible form of mechanism realism. The limits of LHR reveal a deep tension between two aspects of mechanists’ explanatory strategy. Functional decomposition identifies locally relevant entities and activities, while these same entities and activities are also embedded in a nested hierarchy of levels. In principle, a functional decomposition may identify entities engaging in causal interactions that crosscut the hierarchical structure of composition relations, violating the mechanist’s injunction against interlevel causation. We argue that this possibility is realized in the example of ephaptic coupling, a subsidiary process of neural computation that crosscuts the hierarchy derived from synaptic transmission. These considerations undermine the plausibility of LHR as a general view, yet LHR has the advantages that (i) its metaphysical implications are precisely stateable; (ii) the structure it identifies is not reducible to mere aggregate causation; and (iii) it clearly satisfies intuitive and informal definitions of mechanism. We conclude by assessing the prospects for a form of mechanism realism weaker than LHR that nevertheless satisfies all three of these requirements.

The past few years have seen several novel information-theoretic measures of causal emergence developed within the scientific community. In this paper I will introduce one such measure, called ‘effective information’, and describe how it is used to argue for causal emergence. In brief, the idea is that certain kinds of complex system are structured such that an intervention characterised at the macro-level will be more informative than one characterised at the micro-level, and that this constitutes a form of causal emergence. Having introduced this proposal, I will then assess the extent to which it is genuinely ‘causal’ and/or ‘emergent’, and argue that it supports only an epistemic form of causal emergence that is not as exciting as it first seems.

Chalmers (2018) raises three challenges for Michael Graziano's attention schema theory. Our aim in this paper is to bolster Graziano's attention schema theory with some tools and insights from the predictive processing framework, in order to respond to the challenges raised by Chalmers and more generally strengthen the theory. We will first introduce the attention schema theory and the three challenges raised by Chalmers, before outlining our application of predictive processing to the theory and how it can resolve these challenges, thus offering a solution to the meta-problem.

It is generally acknowledged by proponents of ‘new mechanism’ that mechanistic explanation involves adopting a perspective, but there is less agreement on how we should understand this perspective-taking or what its implications are for practising science. This paper examines the perspectival nature of mechanistic explanation through the lens of the ‘mechanistic stance’, which falls somewhere between Dennett’s more familiar physical and design stance. We argue this approach implies three distinct and significant ways in which mechanistic explanation can be interpreted as perspectival: ‘phenomenon perspectivism’, ‘pattern perspectivism’ and ‘hierarchy perspectivism’. We evaluate the strength of the perspective-dependency implied by each of these, and along the way, discuss their significance for wider debates within the new mechanism literature, such as the nature of function attribution and an ontic vs epistemic understanding of explanation.

Our aim in this paper is to explore two possible directions of interaction between normative folk psychology and decision theory. In one direction, folk psychology plays a regulative role that constrains practical decision‐making. In the other direction, decision theory provides novel tools and norms that shape folk psychology. We argue that these interactions could lead to the emergence of an iterative “decision theoretic spiral," where folk psychology influences decision‐making, decision‐making is studied by decision theory, and decision theory influences folk psychology. Understanding these interactions is important both for the theoretical study of social cognition and decision theory, and also for thinking about how to implement practical interventions into real‐world decision‐making.

Mechanistic explanation involves the attribution of functions to both mechanisms and their component parts, and function attribution plays a central role in the individuation of mechanisms. Our aim in this paper is to investigate the impact of a perspectival view of function attribution for the broader mechanist project, and specifically for realism about mechanistic hierarchies. We argue that, contrary to the claims of function perspectivalists such as Craver, one cannot endorse both function perspectivalism and mechanistic hierarchy realism: if functions are perspectival, then so are the levels of a mechanistic hierarchy. We illustrate this argument with an example from recent neuroscience, where the mechanism responsible for the phenomenon of ephaptic coupling cross-cuts the more familiar mechanism for synaptic firing. Finally, we consider what kind of structure there is left to be realist about for the function perspectivalist.

The proposal that probabilistic inference and unconscious hypothesis testing are central to information processing in the brain has been steadily gaining ground in cognitive neuroscience and associated fields. One popular version of this proposal is the new theoretical framework of predictive processing or prediction error minimization, which couples unconscious hypothesis testing with the idea of ‘active inference’ and claims to offer a unified account of perception and action. Here we will consider one outstanding issue that still looms large at the core of the PEM framework: the lack of a clear criterion for distinguishing conscious states from unconscious ones. In order to fulfill the promise of becoming a unifying framework for describing and modeling cognition, PEM needs to be able to differentiate between conscious and unconscious mental states or processes. We will argue that one currently popular view, that the contents of conscious experience are determined by the ‘winning hypothesis’, falls short of fully accounting for conscious experience. It ignores the possibility that some states of a system can control that system’s behavior even though they are apparently not conscious. What follows from this is that the ‘winning hypothesis’ view does not provide a complete account of the difference between conscious and unconscious states in the probabilistic brain. We show how this problem for the received view can be resolved by augmenting PEM with Daniel Dennett’s multiple drafts model of consciousness. This move is warranted by the similar roles that attention and internal competition play in both the PEM framework and the multiple drafts model.

Recently, some authors have begun to raise questions about the potential unity of 4E (enactive, embedded, embodied, extended) cognition as a distinct research programme within cognitive science. Two tensions, in particular, have been raised:(i) that the body-centric claims embodied cognition militate against the distributed tendencies of extended cognition and (ii) that the body/environment distinction emphasized by enactivism stands in tension with the world-spanning claims of extended cognition. The goal of this paper is to resolve tensions (i) and (ii). The proposal is that a form of ‘wide computationalism’can be used to reconcile the two tensions and, in so doing, articulate a common theoretical core for 4E cognition.

This paper spells out the attention schema theory of consciousness in terms of the predictive processing framework. As it stands, the attention schema theory lacks a plausible computational formalization that could be used for developing possible mechanistic models of how it is realized in the brain. The predictive processing framework, on the other hand, fails to provide a plausible explanation of the subjective quality or the phenomenal aspect of conscious experience. The aim of this work is to apply the formal tools of predictive processing to outline a possible model of cognition that details a plausible way in which organisms may acquire or construct the kinds of schemas postulated by the attention schema theory. This is done in terms of Bayesian control models, which tend to favour representing information in a frugal or 'sparse' manner that strips away detail in precisely the way that the attention schema requires. Once integrated, these two accounts can offer a plausible explanation of the supposedly irreducible and mysterious 'feel' of conscious experience, in terms of the frugal manner in which perceptual mechanisms are represented by the attention schema.

The target article criticises reductionist programs in cognitive science for failing to take into account important explanatory features of the organism's physical embodiment and task environment. My aim in this commentary is to show how such features are increasingly being taken seriously by (some) researchers in cognitive neuroscience, who describe the functional activity of neural structures in terms that are context-sensitive rather than intrinsic. This approach can allow us to take seriously the concerns presented in Gallagher’s [2019] target article without having to completely give up on neuroscientific explanations of human behaviour.

The aim of this paper is to begin developing a version of Gualtiero Piccinini’s mechanistic account of computation that does not need to appeal to any notion of proper functions. The motivation for doing so is a general concern about the role played by proper functions in Piccinini’s account, which will be evaluated in the first part of the paper. I will then propose a potential alternative approach, where computing mechanisms are understood in terms of Carl Craver’s perspectival account of mechanistic functions. According to this approach, the mechanistic function of ‘performing a computation’ can only be attributed relative to an explanatory perspective, but such attributions are nonetheless constrained by the underlying physical structure of the system in question, thus avoiding unlimited pancomputationalism. If successful, this approach would carry with it fewer controversial assumptions than Piccinini’s original account, which requires a robust understanding of proper functions. Insofar as there are outstanding concerns about the status of proper functions, this approach would therefore be more generally acceptable.

ABSTRACT Shagrir and Sprevak explore the apparent necessity of representation for the individuation of digits in computational systems.1 1 I will first offer a response to Sprevak’s argument that does not mention Shagrir’s original formulation, which was more complex. I then extend my initial response to cover Shagrir’s argument, thus demonstrating that it is possible to individuate digits in non-representational computing mechanisms. I also consider the implications that the non-representational individuation of digits would have for the broader theory of computing mechanisms. 1 The Received View: No Computation without Representation 2 Computing Mechanisms and Functional Individuation 3 Against Computational Externalism 4 Implications for the Mechanistic Account

Varela, Thompson, and Rosch illustrated their original presentation of the enactive theory of cognition with the example of a simple cellular automaton. Their theory was paradigmatically anti-computational, and yet automata similar to the one that they describe have typically been used to illustrate theories of computation, and are usually treated as abstract computational systems. Their use of this example is therefore puzzling, especially as they do not seem to acknowledge the discrepancy. The solution to this tension lies in recognizing a hidden background assumption, shared by both Varela, Thompson, and Rosch and the computational theories of mind which they were responding to. This assumption is that computation requires representation, and that computational states must bear representational content. For Varela, Thompson, and Rosch, representational content is incompatible with cognition, and so from their perspective the automaton that they describe cannot, despite appearances, be computational. However, there now exist several accounts of computation that do not make this assumption, and do not characterize computation in terms of representational content. In light of these recent developments, we will argue that it is quite straightforward to characterize the enactive automaton as a non-representational computing mechanism, one that we do not think they should have any objections to.

Fresco and Primiero’s recent article, ‘Miscomputation’ , provides a useful framework with which to think about miscomputation, as well as an admirably broad taxonomy of different kinds of miscomputation. However, it also misconstrues the mechanistic approach to miscomputation, which I will argue should not recognise design errors as miscomputations per se. I argue that a computing mechanism, if it is functioning correctly in the physical sense, cannot miscompute on the basis of an error made by an external agent, such as a programmer. This is partially acknowledged in the distinction that Fresco and Primiero make between errors of functioning and errors of design, but they go on to describe both as cases of miscomputation, which I will argue is a mistake, at least with regard to the analysis made by the mechanistic account

Hobson & Friston (2014) outline a synthesis of Hobson's work on dreaming and consciousness with Friston’s work on the free energy principle and predictive coding. Whilst we are sympathetic with their claims about the function of dreaming and its relationship to consciousness, we argue that their endorsement of the Cartesian theatre metaphor is neither necessary nor desirable. Furthermore, if it were necessary then this endorsement would undermine their positive claims, as the Cartesian theatre metaphor is widely regarded as unsustainable. We demonstrate this point and then develop an alternative formulation of their position that does not require the Cartesian theatre metaphor. Our positive goal is to clarify Hobson & Friston’s confusing usage of philosophical terminology, replacing it where possible with the more transparent language of the forward models framework. This will require some modifications to their account, which as it stands is philosophically and empirically unsustainable.

Physical Computation is the summation of Piccinini’s work on computation and mechanistic explanation over the past decade. It draws together material from papers published during that time, but also provides additional clarifications and restructuring that make this the definitive presentation of his mechanistic account of physical computation. This review will first give a brief summary of the account that Piccinini defends, followed by a chapter-by-chapter overview of the book, before finally discussing one aspect of the account in more critical detail.

Whilst much has been said about the implications of predictive processing for our scientific understanding of cognition, there has been comparatively little discussion of how this new paradigm fits with our everyday understanding of the mind, i.e. folk psychology. This paper aims to assess the relationship between folk psychology and predictive processing, which will first require making a distinction between two ways of understanding folk psychology: as propositional attitude psychology and as a broader folk psychological discourse. It will be argued that folk psychology in this broader sense is compatible with predictive processing, despite the fact that there is an apparent incompatibility between predictive processing and a literalist interpretation of propositional attitude psychology. The distinction between these two kinds of folk psychology allows us to accept that our scientific usage of folk concepts requires revision, whilst rejecting the suggestion that we should eliminate folk psychology entirely.

This chapter draws an analogy between computing mechanisms and autopoietic systems, focusing on the non-representational status of both kinds of system (computational and autopoietic). It will be argued that the role played by input and output components in a computing mechanism closely resembles the relationship between an autopoietic system and its environment, and in this sense differs from the classical understanding of inputs and outputs. The analogy helps to make sense of why we should think of computing mechanisms as non-representational, and might also facilitate reconciliation between computational and autopoietic/enactive approaches to the study of cognition.

Traditionally, computational theory (CT) and dynamical systems theory (DST) have presented themselves as opposed and incompatible paradigms in cognitive science. There have been some efforts to reconcile these paradigms, mainly, by assimilating DST to CT at the expenses of its anti-representationalist commitments. In this paper, building on Piccinini’s mechanistic account of computation and the notion of functional closure, we explore an alternative conciliatory strategy. We try to assimilate CT to DST by dropping its representationalist commitments, and by inviting CT to recognize the functionally closed nature of some computational systems.

In this paper we will demonstrate that a computational system can meet the criteria for autonomy laid down by classical enactivism. The two criteria that we will focus on are operational closure and structural determinism, and we will show that both can be applied to a basic example of a physically instantiated Turing machine. We will also address the question of precariousness, and briefly suggest that a precarious Turing machine could be designed. Our aim in this paper is to challenge the assumption that computational systems are necessarily heteronomous systems, to try and motivate in enactivism a more nuanced and less rigid conception of computational systems, and to demonstrate to computational theorists that they might find some interesting material within the enactivist tradition, despite its historical hostility towards computationalism.