We raise an issue relating to the physics that undergird the free energy principle (FEP), and one relating to whether the FEP actually is relevant to perception, and to phenomena of animacy, in general.
The authors argue the merits of various formulations of the FEP in relation to animacy (e.g., Markov blankets, Friston blankets, Pearl blankets). They acknowledge that “the core of the FEP rests upon an intertwined web of mathematical constructs borrowed from physics” (sect. 1, para. 4). They do not question the validity of the underlying physics of the FEP. However, a consistent thread of scholarship raises doubts about the validity of the FEP as a description of physical reality (rather than as a mathematical abstraction that is not meant to be taken as a claim about reality; Schrodinger, Reference Schrodinger1952a, Reference Schrodinger1952b).
All versions of the FEP assume that time is discrete. That is, the mathematical equations of the FEP are defined only if we assume that time is discrete. In most physics, models and theories are structured in ways that assume that time exists as discrete temporal units. This assumption is accepted even by scholars who have criticized other aspects of the FEP (e.g., Colombo and Palacios, Reference Colombo and Palacios2021; Raja, Valluri, Baggs, Chemero, and Anderson, Reference Raja, Valluri, Baggs, Chemero and Anderson2021; Unnikrishnan, Reference Unnikrishnan2020). Yet not everyone accepts this assumption. Bergson (Reference Bergson and Jacobson1922/1999) claimed that time does not exist in discrete units but, rather, exists as a continuum that cannot be sectioned into discrete units (Robbins, Reference Robbins2014). At minimum, Bergson's alternative conception of time alerts us to the fact that mainstream views of discrete time are assumptions or descriptions, rather than established facts (Schrodinger, Reference Schrodinger1952a, Reference Schrodinger1952b). Claims that are based on this assumption, such as Friston's FEP and the current authors' treatment, should more explicitly acknowledge their reliance on these contingent assumptions. It is also important to carefully evaluate the validity of Bergson's alternative perspective and the implications it may have for our understanding of both physics and animacy.
Bergson's (Reference Bergson and Jacobson1922/1999) conception of time is consistent with Gibson's conception of physics, including time. Gibson argued that traditional physics, including electromagnetism, thermodynamics, quantum mechanics, and abstract, discrete time, cannot account for the phenomena of animacy (Gibson, Reference Gibson, Fraser and Lawrence1975, Reference Gibson1979). More broadly, Rosen (Reference Rosen1991) argued that living things rely on physical principles (what he referred to as “new physics”) that are primary to the physics of inanimacy. Put plainly, each of these scholars raised deep questions, not only about the presumed primacy of traditional physics, but also about its literal accuracy as a description of reality. Friston's FEP is part of an ancient tradition by which the physics of inanimacy are assumed to be basic, with the physics of animacy being derivative. Bergson, Gibson, Rosen, and others argue just the opposite: That the physics of animacy are primary, and the physics of inanimacy derivative.
The above considerations relate intimately to our second issue, which concerns the authors' assumption that perception is inferential. They offer as options only inference with a model, or inference within a model. But other options exist. The ecological approach to perception and action claims that the animal–environment interaction lawfully structures patterns in ambient energy such that reality is specified (e.g., Gibson, Reference Gibson1966; Turvey, Reference Turvey2019). If reality is specified, then perception can be direct and, consequently, there is no requirement for inference. Bruineberg, Chemero, and Rietveld (Reference Bruineberg, Chemero and Rietveld2019) accepted this logic, but argued that social affordances cannot be specified and that, therefore, perception of social affordances cannot be direct. It would follow that knowledge of social affordances must depend upon inference. The sole basis for their argument was the fact that social affordances emerge from social conventions, such as linguistic grammar and syntax, or highway speed limits. However, they offered no evidence, either logical or empirical, that social conventions or social affordances actually cannot be specified.
The fact that social affordances emerge from social conventions does not imply that they are free of physical law, such that they cannot be specified. Social conventions are constrained by physical law. For example, all phonetic systems must conform to the acoustic capabilities of the speech organs. Similarly, grammar and syntax, which vary widely across languages, nevertheless exhibit consistencies, and cannot operate outside physical law. Language is used to communicate about physical reality, such that grammar and syntax may be constrained by the physical laws that constrain the events that are the principal subject of linguistic interaction (e.g., Anthony, Reference Anthony2007). Even metaphor is grounded in embodied experience (Gibbs, Lima, & Francozo, Reference Gibbs, Lima and Francozo2004). In short, the claim that social affordances cannot be specified and that, therefore, perception must be inferential, is a claim, rather than a self-evident fact. It may be that social affordances are specified in conformity with physical law, such that all perception can be direct (e.g., Stoffregen and Bardy, Reference Stoffregen and Bardy2001; Stoffregen, Mantel, and Bardy, Reference Stoffregen, Mantel and Bardy2017).
Empirical research can help to address the continuity or unity of perception. Empirical research is consistent with the idea that perception of social affordances may be direct. As one example, human observers can transition easily between perception of personal and interpersonal affordances (e.g., Richardson, Marsh, and Baron, Reference Richardson, Marsh and Baron2007), suggesting that perception of personal and interpersonal affordances may have a similar basis. Perception of social affordances may be grounded in the perception and control of affordances for the individual. For example, locomotor experience (typically, learning to crawl) causally drives the infant's developing understanding of referential communication (e.g., Campos et al., Reference Campos, Anderson, Barbu-Roth, Hubbard, Hertenstein and Witherington2000), while the physical experience of interpersonal synchrony has causal influence on the development of prosocial behavior (Cirelli, Reference Cirelli2018), and social conventions are taught through guided interactions (Nonaka & Stoffregen, Reference Nonaka and Stoffregen2020; Reed, Reference Reed1996). These findings suggest that perception of social affordances may emerge from the kinds of physical interactions that Bruineberg et al. (Reference Bruineberg, Chemero and Rietveld2019) accepted as being amenable to direct perception.
Acceptance of Friston's FEP mandates rejection of any form of direct perception (e.g., Friston, Reference Friston2013). This stark requirement may explain the uncritical nature of the authors' views on specification (Bruineberg et al., Reference Bruineberg, Chemero and Rietveld2019). The alternative is equally stark: If perception is direct, then Friston's FEP cannot be a factual description of animate systems.
We raise an issue relating to the physics that undergird the free energy principle (FEP), and one relating to whether the FEP actually is relevant to perception, and to phenomena of animacy, in general.
The authors argue the merits of various formulations of the FEP in relation to animacy (e.g., Markov blankets, Friston blankets, Pearl blankets). They acknowledge that “the core of the FEP rests upon an intertwined web of mathematical constructs borrowed from physics” (sect. 1, para. 4). They do not question the validity of the underlying physics of the FEP. However, a consistent thread of scholarship raises doubts about the validity of the FEP as a description of physical reality (rather than as a mathematical abstraction that is not meant to be taken as a claim about reality; Schrodinger, Reference Schrodinger1952a, Reference Schrodinger1952b).
All versions of the FEP assume that time is discrete. That is, the mathematical equations of the FEP are defined only if we assume that time is discrete. In most physics, models and theories are structured in ways that assume that time exists as discrete temporal units. This assumption is accepted even by scholars who have criticized other aspects of the FEP (e.g., Colombo and Palacios, Reference Colombo and Palacios2021; Raja, Valluri, Baggs, Chemero, and Anderson, Reference Raja, Valluri, Baggs, Chemero and Anderson2021; Unnikrishnan, Reference Unnikrishnan2020). Yet not everyone accepts this assumption. Bergson (Reference Bergson and Jacobson1922/1999) claimed that time does not exist in discrete units but, rather, exists as a continuum that cannot be sectioned into discrete units (Robbins, Reference Robbins2014). At minimum, Bergson's alternative conception of time alerts us to the fact that mainstream views of discrete time are assumptions or descriptions, rather than established facts (Schrodinger, Reference Schrodinger1952a, Reference Schrodinger1952b). Claims that are based on this assumption, such as Friston's FEP and the current authors' treatment, should more explicitly acknowledge their reliance on these contingent assumptions. It is also important to carefully evaluate the validity of Bergson's alternative perspective and the implications it may have for our understanding of both physics and animacy.
Bergson's (Reference Bergson and Jacobson1922/1999) conception of time is consistent with Gibson's conception of physics, including time. Gibson argued that traditional physics, including electromagnetism, thermodynamics, quantum mechanics, and abstract, discrete time, cannot account for the phenomena of animacy (Gibson, Reference Gibson, Fraser and Lawrence1975, Reference Gibson1979). More broadly, Rosen (Reference Rosen1991) argued that living things rely on physical principles (what he referred to as “new physics”) that are primary to the physics of inanimacy. Put plainly, each of these scholars raised deep questions, not only about the presumed primacy of traditional physics, but also about its literal accuracy as a description of reality. Friston's FEP is part of an ancient tradition by which the physics of inanimacy are assumed to be basic, with the physics of animacy being derivative. Bergson, Gibson, Rosen, and others argue just the opposite: That the physics of animacy are primary, and the physics of inanimacy derivative.
The above considerations relate intimately to our second issue, which concerns the authors' assumption that perception is inferential. They offer as options only inference with a model, or inference within a model. But other options exist. The ecological approach to perception and action claims that the animal–environment interaction lawfully structures patterns in ambient energy such that reality is specified (e.g., Gibson, Reference Gibson1966; Turvey, Reference Turvey2019). If reality is specified, then perception can be direct and, consequently, there is no requirement for inference. Bruineberg, Chemero, and Rietveld (Reference Bruineberg, Chemero and Rietveld2019) accepted this logic, but argued that social affordances cannot be specified and that, therefore, perception of social affordances cannot be direct. It would follow that knowledge of social affordances must depend upon inference. The sole basis for their argument was the fact that social affordances emerge from social conventions, such as linguistic grammar and syntax, or highway speed limits. However, they offered no evidence, either logical or empirical, that social conventions or social affordances actually cannot be specified.
The fact that social affordances emerge from social conventions does not imply that they are free of physical law, such that they cannot be specified. Social conventions are constrained by physical law. For example, all phonetic systems must conform to the acoustic capabilities of the speech organs. Similarly, grammar and syntax, which vary widely across languages, nevertheless exhibit consistencies, and cannot operate outside physical law. Language is used to communicate about physical reality, such that grammar and syntax may be constrained by the physical laws that constrain the events that are the principal subject of linguistic interaction (e.g., Anthony, Reference Anthony2007). Even metaphor is grounded in embodied experience (Gibbs, Lima, & Francozo, Reference Gibbs, Lima and Francozo2004). In short, the claim that social affordances cannot be specified and that, therefore, perception must be inferential, is a claim, rather than a self-evident fact. It may be that social affordances are specified in conformity with physical law, such that all perception can be direct (e.g., Stoffregen and Bardy, Reference Stoffregen and Bardy2001; Stoffregen, Mantel, and Bardy, Reference Stoffregen, Mantel and Bardy2017).
Empirical research can help to address the continuity or unity of perception. Empirical research is consistent with the idea that perception of social affordances may be direct. As one example, human observers can transition easily between perception of personal and interpersonal affordances (e.g., Richardson, Marsh, and Baron, Reference Richardson, Marsh and Baron2007), suggesting that perception of personal and interpersonal affordances may have a similar basis. Perception of social affordances may be grounded in the perception and control of affordances for the individual. For example, locomotor experience (typically, learning to crawl) causally drives the infant's developing understanding of referential communication (e.g., Campos et al., Reference Campos, Anderson, Barbu-Roth, Hubbard, Hertenstein and Witherington2000), while the physical experience of interpersonal synchrony has causal influence on the development of prosocial behavior (Cirelli, Reference Cirelli2018), and social conventions are taught through guided interactions (Nonaka & Stoffregen, Reference Nonaka and Stoffregen2020; Reed, Reference Reed1996). These findings suggest that perception of social affordances may emerge from the kinds of physical interactions that Bruineberg et al. (Reference Bruineberg, Chemero and Rietveld2019) accepted as being amenable to direct perception.
Acceptance of Friston's FEP mandates rejection of any form of direct perception (e.g., Friston, Reference Friston2013). This stark requirement may explain the uncritical nature of the authors' views on specification (Bruineberg et al., Reference Bruineberg, Chemero and Rietveld2019). The alternative is equally stark: If perception is direct, then Friston's FEP cannot be a factual description of animate systems.
Financial support
Thomas A. Stoffregen was supported by NSF-1901423, CHS: Medium: Prediction, Early Detection, and Mitigation of Virtual Reality Simulator Sickness.
Conflict of interest
None.