We applaud Veissière and colleagues for pursuing the ambitious goal of situating the free-energy principle within the context of sociocultural processes. This is, indeed, a much needed undertaking, which has only recently started developing, holding promise for advancing not only relevant sociocultural research fields, but also computational psychiatry (cf. Bolis & Schilbach Reference Bolis and Schilbach2017; Reference Bolis and Schilbach2018b; Constant et al. Reference Constant, Ramstead, Veissière and Friston2019b; Friston & Frith Reference Friston and Frith2015b; Gallagher & Allen Reference Gallagher and Allen2018). In fact, human cognition and culture have often been studied in isolation. For instance, the field of computational psychiatry has been developing rigorous experimental protocols and mathematical toolboxes to mechanistically explain human cognition and action. Yet, until recently a rather individualistic perspective has been adopted, which neglects levels of description beyond the individual (cf. Bolis et al. Reference Bolis, Balsters, Wenderoth, Becchio and Schilbach2017; De Jaegher & Di Paolo Reference De Jaegher and Di Paolo2007; Kirmayer & Crafa Reference Kirmayer and Crafa2014; Schilbach et al. Reference Schilbach, Timmermans, Reddy, Costall, Bente, Schlicht and Vogeley2013). On the other hand, sociocultural fields, such as cultural anthropology, have rightfully adopted a more holistic perspective to complex phenomena of life, yet frequently lacking formal descriptions of cognitive and biological mechanisms (cf. Seligman & Brown Reference Seligman and Brown2009).

An artificial dichotomy between the individual and the collective has inevitably led to a “chicken-egg” paradox (cf. Dumas et al. Reference Dumas, Kelso and Nadel2014). However, such causality dilemmas dissolve once one considers the dialectical nature of human-becoming, which is multiscale, reciprocal, dynamic, cumulative, and inherently contradictory (cf. Bolis & Schilbach Reference Bolis and Schilbach2018b; Di Paolo et al. Reference Di Paolo, Cuffari and De Jaegher2018; Dumas et al. Reference Dumas, Kelso and Nadel2014; Vygotsky Reference Vygotsky, Bibou and Vosniadou1930–1935/1978). Processes from evolution and culture to individual development, learning, and sensorimotor activity, can all be viewed as mutually interacting adjustments between the species and the environment. Here, reciprocity is deep, as “it is not only humans who change the environment, but the environment in turn changes them in face of their impact on it” (Bolis & Schilbach Reference Bolis and Schilbach2018b; Levins & Lewontin Reference Levins and Lewontin1985). TTOM, therefore, constitutes an important development because it addresses how human agents learn shared expectations and how they construct their own social niches in complex interaction between the individual and the environment.

We concretely appreciate the consideration of predictive coding and active inference within a framework of circular causality. Indeed, an organism can be viewed as embedded within the dialectic between the two above-mentioned processes, which in order to survive obeys a simple, but fundamental rule: “adjust yourself to reality or change the reality itself” (Bolis & Schilbach Reference Bolis and Schilbach2018b; Friston Reference Friston2010). When it comes to TTOM, it is not only the agent which learns environmental regularities and adjusts accordingly, but also the environment in turn “learns” the agents’ “beliefs” through repeated and culturally regulated actions. TTOM resonates well with the dialectical attunement hypothesis (Bolis & Schilbach Reference Bolis and Schilbach2018b), which views human-becoming as the interplay between internalization and externalization in and through (culturally-mediated) social interaction; internalization being the “co-construction of bodily hierarchical models of the (social) world and the organism” [cf. predictive coding], whereas externalization the “collective transformation of the world” [cf. active inference]. In a nutshell, “interpersonal statistical regularities shape multiscale hierarchical models on an individual level and vice versa.”

To offer a formal description of how the environment “learns,” the authors interestingly suggest twisting the modeling equations by inverting relevant quantities across actions and sensations. This offers various potential modeling scenarios about the degree of interactivity within the system of brain–body–environment–body–brain (cf. Froese et al. Reference Froese, Iizuka and Ikegami2013). Here, a multiscale meta-Bayesian scheme might nicely lend itself for modeling not only individual processes, but also collective and environmental interactions (Bolis & Schilbach Reference Bolis and Schilbach2017; Brandi et al. Reference Brandi, Kaifel, Bolis and Schilbach2019; Ramstead et al. Reference Ramstead, Badcock and Friston2018).

Not only are we in line with the authors on conceptual and computational grounds, but also concerning the need for empirical studies. To make this more concrete, we describe certain experimental directions: Systematically varying social structure, cultural, and socioeconomic background, affective bonds and interpersonal similarity across interacting individuals will enable the mechanistic study of interpersonal attunement. With regard to psychiatric disorders, construed as disorders of social interaction (Schilbach Reference Schilbach2016), two-person (or indeed collective) psychophysiology allows to move beyond the individual (cf. Bolis & Schilbach Reference Bolis and Schilbach2018a). Taking autism as a paradigm example, the dialectical misattunement hypothesis has put forward a research line, which, moving away from an exclusive study of individual differences, considers types of interacting groups: that is, autistic, neurotypical, and mixed groups, expecting smoother interactions within the more homogeneous groups or dyads (Bolis et al. Reference Bolis, Balsters, Wenderoth, Becchio and Schilbach2017). Taken together, such experiments will not only inform TTOM within the “neurotypical social world,” but also open up avenues for evaluating and updating the ontological status of conditions, such as autism, as relational and interactional (cf. double empathy problem; Milton Reference Milton2012).

Apart from praising TTOM, we would also like to point out a fundamental aspect which, in our opinion, would benefit from further elaboration. We feel that the potentially constitutive role of real-time social interaction in sense-making and human-becoming was not sufficiently taken into account within the model (cf. De Jaegher & Di Paolo Reference De Jaegher and Di Paolo2007; Vygotsky Reference Vygotsky, Bibou and Vosniadou1930–1935/1978). It has been suggested that thinking about and with others might be fundamentally different in real-time interactive scenarios, as compared to passive observational situations (cf. second-person perspective; Redcay & Schilbach Reference Redcay and Schilbach2019; Schilbach et al. Reference Schilbach, Timmermans, Reddy, Costall, Bente, Schlicht and Vogeley2013). Crucially, such interactive interpersonal processes have been thought of as dialectically preceding the individual both in evolutionary and developmental regards (cf. Bolis & Schilbach Reference Bolis and Schilbach2018b; Tomasello Reference Tomasello2019). As Vygotsky proclaimed almost a century ago, “through others we become ourselves” (Vygotsky, Reference Vygotsky and Reiber1931/1987). Yet, to do justice to the authors, the field today has not yet reached a conclusive consensus. For instance, although Di Paolo et al. (Reference Di Paolo, Cuffari and De Jaegher2018) suggest that “interactive situations present a richer, more complex set of possibilities” and “the key to our sociality is not in our heads or in our genes,” Schönherr and Westra (Reference Schönherr and Westra2017) claim to have (conceptually) shown that “ersatz interactivity works just as well as the real thing,” by “real thing” denoting genuine, real-time social interaction. We, therefore, conclude our commentary with a question still desperately begging for a definite empirical answer. Does (real-time) social interaction matter… or is it all in our heads?