Veissière et al., fitting free-energy formalism to human cognition and culture, produce an overdue exciting eco-evolutionary description of culture and associated phenomena with significant scope and eloquence. The authors’ ambition to mathematically unify phylogenetic, ontogenetic, and sociogenetic mechanisms necessary for culture acquisition, expression, adaptation, and transmission is both welcome and admirable. Nevertheless, the precision, power, and internal and external validity of their framework await further testing and elaboration by skilled empiricists and theorists. For more complete, if inconsistent (Gödel Reference Gödel1931), axiomatic accounts, proof and refinement of concepts may be arguably best achieved through the use of broader constructivist approaches that entertain social systems of varying complexity and epistemological likeness to human culture. Veissière et al., for instance, disappointingly avoid considering major evidence-backed cultural features of animal societies and their modifiable relationships to human culture (cf. Clark Reference Clark and Floares2012; Reference Clark2019; Gowdy & Krall Reference Gowdy and Krall2016; Laland & Galef Reference Laland and Galef2009; Russon et al. Reference Russon, Bard and Parker1996), including, but not limited to, pedagogy, tool production and use, mating and mourning rituals, grooming and personal adornment, division of labor, group defense and hunting, parenting, folk medical practice, superstition, communication dialects and mannerisms, abode construction, and sociopolitical hierarchies and maneuvering. Although debated among experts (Laland & Galef Reference Laland and Galef2009), many phylogenetically earlier features of so-called animal culture seem to rely on the same mechanisms commonly attributed to human cultural expression. Indeed, much can be scientifically learned from these antecedent affordances of human culture, perhaps especially characteristics delimiting the origins and evolution of social learning and inferences which, in turn, shape the collective factors embodying and driving social context and culture, a process the authors’ term “Thinking through Other Minds.”

Despite valuable insights gained from contemplating comparative aspects of animal culture, another overlooked or forgotten nascent culture – the culture of intelligent machines – also demands scrutiny and might better inform Veissière et al.'s variational approach to cognition and culture. Intelligent machines notably deliver prospective high-performance computational platforms to model and emulate all known organic forms of Earth life (e.g., Chew et al. Reference Chew, Wenden, Flis, Mengin, Taylor, Davey, Tindal, Thomas, Ougham, de Reffye, Stitt, Williams, Muetzelfeldt, Halliday and Millar2014; Derex & Boyd Reference Derex and Boyd2015; Fan & Markram Reference Fan and Markram2019; Fortuna et al. Reference Fortuna, Zaman, Wagner and Ofria2013; Fung Reference Fung2015; Gillings et al. Reference Gillings, Hibert and Kemp2016; Lenski et al. Reference Lenski, Ofria, Collier and Adami1999; Libin & Libin Reference Libin and Libin2005; Millar et al. Reference Millar, Urquiza, Freeman, Hume, Plotkin, Sorokina, Zardilis and Zielinski2019; Ranjan et al. Reference Ranjan, Logette, Marani, Herzog, Tâche, Scantamburlo, Buchillier and Markram2019; Sarma et al. Reference Sarma, Lee, Portegys, Ghayoomie, Jacobs, Alicea, Cantarelli, Currie, Gerkin, Gingell, Gleeson, Gordon, Hasani, Idili, Khayrulin, Lung, Palyanov, Watts and Larson2018). They, moreover, render virtually-augmented eco-evolutionary spaces to prototype, replicate, and adapt digital life or e-life (i.e., electronic life), new provocative programmable in silico forms of life thought capable of taxa-blurring sophisticated intelligent agency and culture often celebrated and feared by fantasists and futurists (cf. Arbib & Fellous Reference Arbib and Fellous2004; Asada Reference Asada2015; Bostrom Reference Bostrom2014; Briegel Reference Briegel2012; Cardon Reference Cardon2006; Clark Reference Clark2015; Reference Clark2017; in press a; Davies Reference Davies2016; Fung Reference Fung2015; Han et al. Reference Han, Lin and Song2013; Kaipa et al. Reference Kaipa, Bongard and Meltzoff2010; Lake et al. Reference Lake, Ullman, Tenenbaum and Gershman2018; Lumaca & Baggio Reference Lumaca and Baggio2017; Mathews et al. Reference Mathews, Christensen, O'Grady, Mondada and Dorigo2017; McShea Reference McShea2013; Parisi Reference Parisi1997; Thomaz & Cakmak Reference Thomaz, Cakmak and Clark2013; Wallach et al. Reference Wallach, Franklin and Allen2010). Trends in state-of-the-art digital life research and development emphasize dynamic machine-learning construction, operation, and evolution of whole semi-autonomous artificial organisms, such as OpenWorm's simulated nematode Caenorhabditis elegans and the plant-science community's crop-on-a-chip Arabidopsis thaliana. In addition, persistent improvements in machine architecture and software engineering continue to innovate progressively human-like artificial brains and intelligences, such as the Blue Brain Project's intricate virtual brain connectome, IBM's Jeopardy-quiz-show-winning Watson supercomputer, and Amazon's Alexa and Apple's Siri cloud-based digital assistants or digibots.

Without fully autonomous machine intelligences, current state-of-the-art machine form and function, including machine–machine and human–machine interactions, still fail to scale to computational and combinatorial complexities of purposeful sentient humans and their interactive societies. But, the near-seamless, data-exchanging distributed connectivity of smart machines to one another and to humans instantiates a kind of hybrid society built upon quasi model-free/model-based social networks. Networks comprised of social digibots and humans, for example, support local expectation, selective patterning of attention and behavior, and cultural evolution and inheritance. Changes in local network behavior reciprocally affect collective habits, norms, and expectations that typify larger virtually-augmented eco-environmental spaces shared by machines and humans. Critical analysis of this not-so-hypothetical cultural niche of organic and digital life first requires detailing a non-trivial theory of (natural and artificial) minds, through which Veissière et al.'s “Thinking through Other Minds” can be rigorously validated within and across taxonomic and technological margins. To that end, roboticists now carefully employ principles derived from relevant domain-knowledge sources, such as philosophy, computer science, zoology, engineering, neuroscience, anthropology, education, and sociobiology, to identify, craft, and implement machine traits fundamental to emergence and normalization of social–emotional intelligences (cf. Clark Reference Clark2015; Reference Clark2017; Lake et al. Reference Lake, Ullman, Tenenbaum and Gershman2018). Machine intelligences evolve from simple initial state conditions that iteratively generate fuzzy clusters of core complex traits, such as causative inference, personality, self-awareness, empathy, and creativity. These traits help perfect and humanize learner machines via human and machine social modeling and queried instruction, where passive- and active-learning interfaces between learners and demonstrators/teachers correspond to credible primitive digital life manifestations of Veissière et al.'s “Thinking through Other Minds.” Affordances of combined machine and human cultures then map to patterned immersive experiences of response priming/contagion, social facilitation, incentive motivation, and local/stimulus enhancement.

Similar to the authors’ variational approach, free-energy constraints on digital life and virtually-augmented eco-evolutionary spaces dictate statistical regularities leveraged to forecast and organize behavior into machine and, in part, hybrid machine–human cultures (cf. Clark Reference Clark and Floares2012; Reference Clark2015; Langton Reference Langton1990; Wolfram Reference Wolfram1984). However, machine-culture thermodynamic profiles may radically deviate from profiles distinguishing human and even hybrid machine–human cultures. Causes of framework inconsistencies predictably stem from the frequent quasi-nature of machine architectures, algorithms, and social networks, which might inaccurately emulate human brains, neurocognitive processes, and social structures and practices. Furthermore, digital life may never bind or may become unbound from referential human-life conventions, so self-organizing machine performances reflect and harness purer semiotic and cybernetic optimization. Dissociations between digital and organic life thus greatly impact the types of statistics governing respective epistemic machine and human cultural resources, likely invalidating Veissière et al.'s preferred classical Bayesian models for variational frameworks more suited to individual and hybrid taxon/technology cultural niches defined by rarer, exotic, or otherwise different statistics (cf. Clark Reference Clark and Floares2012; Reference Clark2015).