The fourth law of thermodynamics (LMEP) and cognition from first principles: commentary on Barrett’s “On the nature and origins of cognition as a form of motivated activity”

Nat Barrett’s (2019)“On the nature and origins of cognition as a form of motivated activity” shines a well-focused light on the “truck-size-hole” (cf. Lyon, 2006) that undermines the theoretical core of virtually every species of enactive or “radical” embodied (nonrepresentational/noncomputational) forms of cognitive science (sometimes “E” or “4E Cognition”). None of them has an account of the normativity, end-directedness, or motivated action that actually characterizes cognition and on which any robust cognitive theory must depend. Barrett rightly points to the failure of the core enactivist assertion that “persistence” (or “proscriptive normativity,” negative feedback, or homeostasis) is the fundamental motive of cognition and, likewise to the claim of “radical embodied” theorists, that cognition is explained by dynamics, namely, agent–environment nonlinearly coupled dynamical systems. Barrett’s nose for forensics takes him quickly and correctly to the problem of origins. Cognition is simply given by these theories sui generis, like an act of so much vitalism, all of a sudden appearing in the world, and he cites Lyon’s (2006) call for a “biogenic” account of cognition, for making this point. But as correct as she is for citing the problem, a “biogenic” account per se cannot work. It merely regresses the problem. Classical evolutionary epistemology (CEE; for example, Callebaut & Pinxten, 1987; Swenson, 1989, 1997, 1998) was just such an attempt to understand the naturalistic evolution of cognition from its beginning with single-cell prokaryotes all the way to humans. This made a number of important contributions including the understanding that the first living system was a cognitive system and vice versa. The evolution of life on Earth is a cognitive or epistemic process, clearly motivated activity as Barrett describes it and clearly not defined by homeostasis (proscriptive normativity).

CEE’s problem was that by trying to understand cognition as a “biogenic” process, and more particularly explained by Darwinian theory (natural selection), it also assumed cognition (life) and its defining normativity or particular kind of end-directed behavior to begin with (the “origin of life”) outside its theory (Swenson, 1997, 1998). Following Swenson and Turvey (1991), Barrett recognizes the need for a physical selection principle to account for this which we identified as the Law of Maximum Entropy Production (“LMEP” or later “The fourth Law,” for example, see Martinez-Kahn & Martinez-Castilla, 2010; Swenson, 2009, 2010, 2016) and he turns to recent experiments of branching pattern formation with chromium beads in an oil-filled Petri dish between a source and a sink (Kondepudi, Kay, & Dixon, 2015). Although the experiment shows LMEP (path selection by entropy production), it does not show the kind of symmetry-breaking macroscopic ordering that characterizes cognition (or life) at its origin and throughout its evolution from single cells to human culture, and is thus of limited use for the specific problem at hand. Pointing to what is missing in these experiments, Barrett identifies a kind of circularly causal system as described by enactive theorists but not one governed, as they say, by homeostasis (proscriptive normativity or simply “persistence”), and his point is dead on. Cognitive systems are not defined by proscriptive normativity. They are self-amplifying, deviation amplifying systems—heterostats not homeostats. They are autocatakinetic (“ACK”) systems. ¹ Each ACK system has its origin (entification) in a fluctuation-amplifying symmetry-breaking event whereby a new macroscopic “selfish” entity comes into the world. It is the circular closure of the constitutive relations of an ACK system whereby output works back on input that defines it as an entity, a “self,” and with it a new level of macroscopic behavior operating across space–time dimensions inaccessible to the previously uncoordinated components literally comes into the world (for schematic and the relational ontology of a canonical ACK system, see for example, Swenson, 2010, 2016).

LMEP ² as a physical selection principle explains these origins, which, as Barrett has noted following Swenson and Turvey (1991), happens opportunistically whenever it gets the chance. ACK systems are self-amplifying sinks that puncture the fabric of space–time providing access to otherwise inaccessible dimensions of dissipative space. The nomological basis for end-directed behavior and the “selfish” nature of ACK systems is thus now well understood, but when can end-directed behavior be called cognitive? All cognitive systems are ACK systems, but not all ACK systems are cognitive, and there is no warrant for invoking cognition to explain their behavior. Such simpler ACK systems, for example, tornadoes, dust devils, and the Benard cells, as well as non-ACK simpler end-directed systems, for example, rivers flowing down slopes, heat flowing down a gradient, the “cabin in the woods” example we gave in Swenson and Turvey, or the branching experiments discussed earlier, are all causally determined by their local potentials or gradients. They are “slaves” of their local force-fields. Remove the local potential (e.g. the heat source in the Benard cell) and they “die.” In stark contrast, remove the local potential from a cognitive system (e.g. food source) and it often becomes more active.

The hallmark of cognitive systems is that they can constitute their ACKs in ways that are arbitrary to and often antithetical to local force-fields or potentials. Their end-directed behavior is determined not by local potentials but by meaning. They use their on-board (stored) potential to disconnect from, or ignore, and move across local force-fields to search out, discover, and access new and discontinuously located potentials using the invariant (symmetry) properties of ambient energy flows that nomologically provide information about paths to distal ends or potentials. ³ As we elaborated in Swenson and Turvey (1991), with the opening up of new and otherwise inaccessible dimensions of dissipative space this affords, LMEP readily explains both the rapid origin of cognition on Earth and the new kinds of increasingly differentiated cognitive systems that developed over evolutionary time from single-cell to multi-cell systems with nervous systems, and the rapid encephalization of humankind. This in turn leads to “second-order cognitive systems,” namely, human cultural systems with language and symbol processing marked by “information about information about.”

The distinctive functional capability of cognitive systems is part and parcel of their distinctive ACKs. They are component production systems, namely, “replicative” systems, and their ability to act arbitrarily with respect to local potentials follows from an arbitrary function of the component production process. A primitive of replicative ordering is a set of rate-independent constraints (“RICs”)—internal constraints that are discrete, sequential, and rate-independent relative to the rest of the ACK cycle. The order of the sequences, like the words on this page (a higher order instantiation of this same primitive), or the sequence of base pairs in a DNA string, for example, is thermodynamically arbitrary with respect to the rate at which they are “written” and “read” ⁴ (Swenson, 1998, 1999, 2010, 2016; Swenson & Turvey, 1991). This takes us to a place where we can talk about and fully resolve the issues that Barrett correctly points out regarding the normativity, the intentionality (end-directed behavior determined by meaning), the origin, and in short the nature of cognitive systems as motivated activity from first principles (LMEP or the fourth Law, First Law time-translation symmetry, ACK systems, and RICs). With this, the understanding that cognitive systems are in fact differentiations of the space–time fabric of the world in its own becoming (e.g. Swenson, 2000), we leapfrog the central problems that plague the core of all enactive or 4E cognitive theories that Barrett points out and get to a fully robust alternative to the still widespread computational model of cognition that is nomologically embodied, embedded, enactive, and extended from first principles.