Neural and phenotypic representation under the free-energy principle

Abstract

Highlights We develop a generalizable model of the representational capacities of living creatures based on the free-energy principle. We review the free-energy formulation and its account of the emergence of representational capacities in living creatures. We review formulations of self-assembly and pattern formation in living systems, premised on the free-energy principle. We provide numerical demonstrations for the neuronal packet hypothesis (that Markov blankets encode stimulus features).

Psychophysical identity and free energy

“…An intimately related route to transparency appeals to the scale-free nature of variational free energy dynamics. As suggested in [11,12,13], the relations between global and local free energy minimization in a supersystem can be understood in terms of subsystems themselves undergoing the same type of energy minimization/inference process at their own scales, yielding hierarchies of embedded systems all of which obey the FEP, and in which the dynamics of superordinate systems constrain those of the subsystems [13] 25 (see [62] for an application to neuronal ensembles). Since the thermodynamic potential energies of a complex system and of its subsystems are constitutively related, the identity thesis naturally yields such mutual constraints on inferential dynamics across levels of organization.…”

Natural language syntax complies with the free-energy principle

“…There is precedence in recent literature for these suggestions. The FEP may be compatible with an instrumentalist theory of mental representations, via which representations are useful fictions for explanatory goals (Ramstead et al 2020(Ramstead et al , 2021. This is compatible with certain models in philosophy of language (Collins 2017, Pietroski 2018, which assume that lexical items have no one-to-one direct referent in the external world, but are composites of distinct representational domains that are used for successful, efficient interpretation.…”

Thinking like a State - Embodied intelligence in the deep history of our collective minds

“…The attraction of the agent-niche system toward its most probable regions can therefore be interpreted as "self-evidencing" (Hohwy, 2016), in the sense that it constitutes an epistemical activity by which the agent proves that their world is indeed what they expect it to be. Because this synchronisation drive the agent's activity and structure to incorporate statistical features of their niche through behavioural and evolutionary time, they come to embody models of the world they enact Ramstead, Kirchhoff, et al, 2020;Ramstead, Hesp, Tschantz, et al, 2021).…”

“…While this conceptual similarity between Active Inference and the enactive approach was widely noted, the deep complementarity between their respective formalisms remains a somewhat underground topic. Formal models grounding enactive autonomy are purely syntactical (ie they're interested in the structural properties of the system per se, see eg Aguilera & Bedia (2018); Montévil & Mossio (2015b)), while the formalism of Active Inference shows how semantical content (ie meaning) can arise from basic synchronisation between an organism and their niche Ramstead, Hesp, Tschantz, et al, 2021). This simple feature provides a clear grip on how agents can come to embody knowledge about their ecological niche, and leverage this knowledge by enacting a meaningfully integrated reality.…”