The Computational Boundary of a "Self": Developmental Bioelectricity Drives Multicellularity and Scale-Free Cognition

All epistemic agents physically consist of parts that must somehow comprise an integrated cognitive self. Biological individuals consist of subunits (organs, cells, and molecular networks) that are themselves complex and competent in their own native contexts. How do coherent biological Individuals result from the activity of smaller sub-agents? To understand the evolution and function of metazoan creatures’ bodies and minds, it is essential to conceptually explore the origin of multicellularity and the scaling of the basal cognition of individual cells into a coherent larger organism. In this article, I synthesize ideas in cognitive science, evolutionary biology, and developmental physiology toward a hypothesis about the origin of Individuality: “Scale-Free Cognition.” I propose a fundamental definition of an Individual based on the ability to pursue goals at an appropriate level of scale and organization and suggest a formalism for defining and comparing the cognitive capacities of highly diverse types of agents. Any Self is demarcated by a computational surface – the spatio-temporal boundary of events that it can measure, model, and try to affect. This surface sets a functional boundary - a cognitive “light cone” which defines the scale and limits of its cognition. I hypothesize that higher level goal-directed activity and agency, resulting in larger cognitive boundaries, evolve from the primal homeostatic drive of living things to reduce stress – the difference between current conditions and life-optimal conditions.

The mechanisms of developmental bioelectricity - the ability of all cells to form electrical networks that process information - suggest a plausible set of gradual evolutionary steps that naturally lead from physiological homeostasis in single cells to memory, prediction, and ultimately complex cognitive agents, via scale-up of the basic drive of infotaxis. Recent data on the molecular mechanisms of pre-neural bioelectricity suggest a model of how increasingly sophisticated cognitive functions emerge smoothly from cell-cell communication used to guide embryogenesis and regeneration. This set of hypotheses provides a novel perspective on numerous phenomena, such as cancer, and makes several unique, testable predictions for interdisciplinary research that have implications not only for evolutionary developmental biology but also for biomedicine and perhaps artificial intelligence and exobiology.

Why did some competent unicellular organisms join together to form complex bodies, and how do they cooperate during highly robust embryogenesis and regeneration of anatomical structures? Why does this process break down during carcinogenic defection from the body plan? How can we best understand and control biological systems that consist of numerous nested levels of organization, such as bacteria and biofilms, which can functionally interact with the host’s cells, tissues, and organs? How are lower level (molecular and cellular) activities harnessed toward adaptive system-level outcomes during regulative development and adaptation to novel stressors? What is the relationship between the ability of cells to implement invariant organ-level morphogenetic goal states and the purposive activity of brains? What dynamics enable the scaling of cognitive capacities from the simple memory functions found in bacteria to those of sophisticated minds?

All these fundamental biological problems have one thing in common: the need to understand and formalize what a coherent Individual or Agent is, in a way that is compatible with a gradual co-evolution of minds and bodies (Tarnita et al., 2013). While significant work has addressed this topic from an evolutionary perspective, I suggest a different and complementary view called Scale-Free Cognition, synthesizing ideas from theories of computation and control to identify common information-processing events occurring at multiple levels of organization. I propose a semi-quantitative metric, based on the spatio-temporal boundaries of events that systems measure and try to control, that can be used to define and compare the cognitive boundaries for highly diverse types of agents (which could be biological, exo-biological, or artificial). Ideas from the fields of proto-cognition, developmental biophysics, and information theory offer a novel lens with which to understand the evolution, development, physiology, and behavior of a wide range of living systems. Focusing on information processing and decision-making enables a unifying conception of goal-directedness in biological systems, which naturally scales from simple homeostatic pathways to complex cognition via evolutionarily ancient physiological mechanisms of cell-cell communication. Here, I illustrate these hypotheses from the perspective of developmental bioelectricity, which evolution has robustly exploited for cognitive scaling; however, the same general scheme applies to any similar mechanism, whether chemical, physical, or other.

These ideas are explored here from the perspective of the following three core assumptions. First is a commitment to evolution: every capacity has a natural history and emerged from simpler variants. Closely related is the idea that cognition, like “organism status” (Child, 1915; Queller and Strassmann, 2009), is not a binary capacity that exists in higher organisms only (Figure 1). Thus, it is fundamentally incorrect to view functional capacities such as memory, prediction, goal-directedness, etc. as entirely new features appropriate only to advanced forms of life, or to view descriptions of cognitive capacities in primitive or aneural organisms as category mistakes (Baluška and Levin, 2016). Indeed, failure to appreciate cognitive processes in a system that would have otherwise improved prediction and control is a “neganthropomorphic fallacy” – a type 1 error that is as bad for empirical research as is profligate anthropomorphic reasoning. This is a kind of “intentional stance” (Dennett, 1987) approach, generalized beyond brains and behavior. Thus, here it is assumed that there is no such thing as magical “true cognition” that is the province of humans, without a smooth history of precursor capacities in simpler forms, stretching back to the base of the tree of life (Dennett, 2017; Ginsburg et al., 2019). Second, it is assumed that all metaphors are to be judged by their utility in driving scientific progress and that there is not a binary categorization of scientific pictures which should be taken literally or not, which can be decided a priori.

Thus, whether a way of thinking about a system is correct or mere metonymy is to be determined by whether the specific metaphor gives rise to new, robust research programs – it is an empirical question to be answered in time, based on whether a given metaphor improves prediction and control in novel cases at the bench, compared to other existing metaphors. Third, much of the discussion centers around “goals”, related to teleology, a hotly debated topic (Nagel, 1979; Ruse, 1989; Teufel, 2011; McShea, 2012, 2013, 2016; Kolchinsky and Wolpert, 2018): here goal-directedness is taken in the non-magical, cybernetic, engineering, and control theory sense, of a feedback system that operates to maximize some specific state of affairs (which can be modeled as a dynamical system with attractors in its state space). Finally, except for a few brief remarks at the end, no claims about consciousness (first person experience or a sense of self as qualia) are made: all of the examples concern functional, third-person, objective capacities, computations, and behaviors.

“Of course there is no question that a tree or an elephant is one individual, and we have a very clear mental picture of what this means, for we ourselves are individuals. But there are lower forms in the borderland between one-celled organisms and multicellular organisms that are more bothersome in this respect.”

Neuroscientists (and philosophers of mind) have long wrestled with the question of how many can add up to one – a “conceptual self that is composed of interacting neural regions” with centralized agency and planning (Murray et al., 2015). But this is not just a key question for neurobiology: as will be seen below, important aspects of biomedicine also hinge on the ability to identify higher-order control structures composed of cells or pathways. Epistemology requires a subject (whoever it is that does or does not know something), and this is critical not only for human philosophers but also for many animals. Survival depends on evolving finely tuned agency detectors that allow them to parse their world into agents that pursue recognizable goals and can be interacted with in a way different than with their parts (Mar et al., 2007).

Exobiologists may at some point be confronted with the task of understanding what aspect of an alien ecosystem constitutes some sort of coherent biological Individual. This is likely to be a very non-trivial task, as even life on Earth presents a significant challenge in the cases of metazoan microbiomes and colonial/symbiotic organisms (Daniels et al., 2016). In addition to evolved/natural Agents, the field of artificial intelligence has (or soon will) give rise to numerous constructs that may represent kinds of Selves. The goal then is not to attempt to draw a sharp line but to understand the factors that go into empirically useful ways to demarcate Individuals in a given context.

The question of defining an “organism” has been long discussed in biology (Huxley, 1852; Loeb, 1916, 1937; Sober, 1991; Maynard Smith and Szathmáry, 1995; West et al., 2015). Much work has been done on defining compound Individuals, from perspectives of evolution (the forces which drive long-term changes in the parts and how they relate), game theory (competition between and within the parts), thermodynamics and dissipative systems, systems theory, and even immunology (Prigogine, 1980; Rosen, 1985; Pradeu and Carosella, 2006; Strassmann and Queller, 2010; Pradeu, 2012, 2016; Godfrey-Smith, 2016, 2017). Below, I first present some background on aspects of flexible decision-making at different levels of biological organization and then argue for a definition based on information and goal-directedness (Walker et al., 2016) from a developmental biology perspective. I suggest candidates for proximate mechanisms that drive major transitions and provide a plausible evolutionary story of how primitive homeostasis leads to advanced agency (Man and Damasio, 2019). This perspective is complementary to others, not incompatible with them. There is likely more than one useful definition of what constitutes a cognitive agent, in terms of making predictions and optimizing control policies for the origin and behavior of specific agents in various circumstances. As will be seen below, the information-centered definition of the Self has specific advantages for dealing with nested (multi-scale) structures observed in the biosphere and for understanding the shifting (not fixed) boundaries between Self and environment that can change within the timeline of a single individual (not only on evolutionary timescales).

It is a well-known fact that the biosphere is a set of nesting dolls (Smythies, 2015). Eco-systems consist of groups that are comprised of organisms, which in turn are made of organs composed of tissues, which consist of cells made up of biochemical networks.

Remarkably, flexible and adaptive behavior is found at every level, which provides an important background for thinking about scale-invariant, essential features of Individuals in the broadest sense. Do integrated Selves only exist at the level of “organisms” (bodies), or could they arise and co-exist at multiple levels of organization and be recognized in novel contexts and implementations? In preparation for a proposed definition of Selves as goal-directed computational agents regardless of implementation, it is helpful to begin by considering novel embodiments of capacities usually associated with brains.

Single cells are composite agents that exhibit extremely rich patterning and behaviors and can be divided into even smaller independent units (such as cytoplast fragments with autonomous activity) because of their dynamic cytoskeletal and protein network subsystems (Novák and Bentrup, 1972; Albrecht-Buehler, 1985; Ford, 2017; Siccardi and Adamatzky, 2017; Barvitenko et al., 2018; Graham et al., 2018). These capacities of cells presage their swarm behavior as metazoan organisms (Gregg, 1959). Most features observed in the anatomical control of complex organisms, including differentiation (Sogabe et al., 2019), plasticity (Koch et al., 2017), programmed cell death (Gordeeva et al., 2004), regenerative repair (Morgan, 1901), and “neural” machinery (Tasneem et al., 2005; Liebeskind et al., 2011; Burkhardt, 2015) already existed in ancient, unicellular life forms. Cells such as bacteria and yeast, as well as advanced plants, have been studied for their ability to compute and predict future events using patterns inferred from prior experiences (Saigusa et al., 2008; Pezzulo and Castelfranchi, 2009; Stepp, 2009; Shemesh et al., 2010; Stepp and Turvey, 2010; Mossbridge et al., 2012; Dhar et al., 2013; Goel and Mehta, 2013; Bohm et al., 2016; De Berker et al., 2016; Katz and Springer, 2016; Peters et al., 2017; Wilson et al., 2017; Gagliano et al., 2018; Katz et al., 2018). Such aneural systems have also been used to understand neural function (Koshland, 1983; Morimoto and Koshland, 1991; Sarto-Jackson and Tomaska, 2016). Thus, capacities usually assigned to Individuals with nervous systems, such as integrating spatio-temporal information, memory, and ability to pursue specific outcomes via selection from a number of possible behaviors evolved from far older pre-neural origins (Eisenstein, 1975; Boisseau et al., 2016).

The emerging field known as “basal cognition” tracks the evolutionary history of learning and decision-making processes, beginning from the dynamic problem-solving capacities of cellular and subcellular forms (Lyon, 2006, 2015; Ginsburg et al., 2019). Many examples of memory, anticipation, context-dependent decision-making, and learning are exhibited by organisms from yeast and bacteria to plants and somatic cells [reviewed in (Lyon, 2006, 2015; Baluška and Levin, 2016)]. This is even true of subcellular-level components, e.g., gene-regulatory networks can execute similar learning and computational properties as neural networks, as can cytoskeletal networks, cell signaling pathways, reaction-diffusion chemistry, and metabolic networks (Watson et al., 2010; Szabó et al., 2012; Stockwell et al., 2015; Prentice-Mott et al., 2016; Dent, 2017; Stovold and O’Keefe, 2017; Barvitenko et al., 2018; Gabalda-Sagarra et al., 2018; Bulcha et al., 2019). Single cells are very good at managing their morphology, behavior, and physiology as needed for survival, altering their motility, and metabolism in response to, and proactively in, changing environmental conditions. They succeed in exploiting their microenvironment toward optimal reproduction by selecting among numerous possible choices of gene expression patterns and behaviors. While the mechanisms by which unicellular organisms’ ability to accomplish specific adaptive ends is harnessed toward cooperative multicellularity is still poorly understood, one thing is clear: somatic cells did not lose their behavioral plasticity and computational capabilities in becoming part of metazoan swarms (bodies): they scaled them to enable pursuit of larger goals consisting of creation and upkeep of massively complex anatomies (Pezzulo and Levin, 2015).

Metazoan embryos are possible because the progeny of a fertilized egg cell can cooperate to create an invariant, large, complex anatomical structure with very high fidelity. Crucially, this is not simply an emergent result of hardwired processes but a very plastic, context-dependent system that achieves invariant patterning outcomes in an uncertain world. Regulative development (e.g., the two normal bodies resulting when an early embryo is cut in half) reveals that cellular swarms are able to achieve the same desired end-state (a species-specific target morphology) despite drastic perturbations and highly unexpected starting states. For example, when tadpoles are perturbed in the laboratory such that their craniofacial organs are in abnormal positions, they still make largely normal frog faces because eyes, jaws, and other structures move around in un-natural paths and only stop when a “correct frog face configuration” is reached (Vandenberg et al., 2012; Pinet and McLaughlin, 2019). Similarly, salamander tails grafted to the flank slowly remodel into limbs – a structure more compatible with the large-scale anatomical spec of a salamander (Farinella-Ferruzza, 1956). Many animals, such as axolotls, are able to regenerate whole limbs, no matter where they are amputated, and other organs; a key aspect is that the new appendages grow until the precisely correct structures are made – no more and no less. Developing kidneys will form tubules of the same diameter, whether by cell-cell communication of ~10 cells per cross-section in normal settings or by cytoskeletal bending of single cells around themselves when cell size is artificially increased drastically, revealing the ability of the system to harness diverse molecular mechanisms toward the same anatomical outcome (Fankhauser, 1945). These are but a few of the ubiquitous examples of remodeling toward an invariant end.

They reveal the ability of somatic cells to not only pursue specific target morphologies despite perturbations and novel scenarios, but to pursue collective patterning outcomes that are truly enormous with respect to the scale of size and organization of single cells: the length of a limb or the configuration of a face are simply not defined at the level of single cells – the set points of pattern-homeostatic mechanisms implemented by a cellular collective are large, organ-level macrostates.

Also important to the understanding of compound biological individuals is the ability of cells to make decisions as a single coherent unit. For example, in early embryos, regions on the left and right sides of the midline need to express left- or right-specific genes in order to establish invariant laterality of the heart and visceral organs. Experimental interference with a number of physiological mechanisms upstream of asymmetric gene expression is sufficient to randomize this normally invariant pattern, producing duplicated (LL or RR) or reversed (RL) patterns. Remarkably, however, this randomization is made on a group level – in each case, the entire domain randomly picks L or R identity, not exhibiting a speckled appearance where each cell adopts a stochastic outcome. The decision is random, but all of the cells in the domain are coordinated to make the same random decision (Levin et al., 2002). This is likewise seen in the conversion of normal frog pigment cells to melanoma – the decision to convert is stochastic across a population of animals, but it is made by all of the melanocytes within a body acting as a single coherent deciding unit (Lobikin et al., 2015). The ability of cells to join into domains that execute group decisions, whether deterministic or stochastic, is an essential component of the evolution of complex forms and will be a central component of a proposed definition of a Self.

The empirical utility of pursuing metaphors based on the parallels between adaptive whole organism behavior and the plasticity of cellular activity during construction and repair of a body is discussed in detail elsewhere (Pezzulo and Levin, 2015). However, for the proposed view of agency described below, and for thinking about its evolutionary origin, it is important to realize that the parallels between goal-directed behaviors and morphogenesis are not only functional but reflect deep conservation of molecular mechanisms. Neural networks control the movement of a body in three-dimensional space; this scheme may be an evolutionary exaptation and speed-optimization of a more ancient, slower role of bioelectrical signaling: the movement of body configuration through anatomical morphospace during embryogenesis, repair, and remodeling (Sullivan et al., 2016; Mathews and Levin, 2018; Mclaughlin and Levin, 2018). This is an expansion of previous proposals of minimal cognition as sensorimotor coordination (Van Duijn et al., 2006), to include cell behavior during morphogenesis as a kind of sensorimotor activity of a patterning Agent.

Developmental bioelectricity is the ubiquitous exchange of slowly changing ion-based voltage signals within and among cells (Funk, 2013; Levin and Martyniuk, 2018). All cells are electrically active, and modern neurons evolved from pre-neural precursors that were already reaping the benefits of ionic signaling for computation. How ancient are these mechanisms – how early was bioelectric coordination exploited by evolution?

Bioelectric dynamics, already used for coordination within bacterial biofilms (Prindle et al., 2015), also enable embryonic tissues to implement robust growth and morphogenesis (Bates, 2015), as well as by unicellular organisms to coordinate behavior (Van Houten, 1979). Recent advances in developmental bioelectricity have shown how endogenous dynamics of resting potential changes modify transcriptional cascades and thereby instruct axial patterning, organ determination, and size control, as well as guiding the behavior of individual cells (Sundelacruz et al., 2009; Levin, 2014; Levin and Martyniuk, 2018). It is thus not surprising that drugs that modify traditional cognitive capacities can be strong teratogens (Hernandez-Diaz and Levin, 2014), while anesthetics reduce and modify regenerative capacity (Buchanan, 1922; Tseng et al., 2010). We have previously argued that the deep evolutionary conservation of ion channel and neurotransmitter mechanisms highlights a fundamental isomorphism between developmental and behavioral processes. Consistent with this, membrane excitability has been suggested to be the ancestral basis for psychology (Grossberg, 1978; Cook et al., 2014; Pezzulo and Levin, 2015). Thus, it is likely that the cognitive capacities of advanced brains lie on a continuum with, and evolve from, much simpler computational processes that are widely conserved at both the functional and mechanism (molecular) levels. The information processing and spatio-temporal integration needed to construct and regenerate complex bodies arises from the capabilities of single cells, which evolution exapted and scaled up as behavioral repertoires of complex nervous systems that underlie familiar examples of Selves.

An appreciation of these aspects of developmental biology blurs the distinction between mind and body – a direction already being explored in the field of soft body robotics (Pfeifer et al., 2007a,b; Pfeifer and Gomez, 2009), which expands the possibilities for defining new kinds of Selves. A computational perspective on the task of building specific complex anatomies suggests interesting new ways to think about how an integrated Self can be formed and how its boundaries are maintained or altered during its lifetime and within evolutionary timescales.

The communication that enables cells to join into collectives that make decisions about the growth and form of organ-level structures (i.e., what to sculpt and when to stop) can go awry, resulting in cancer (Chernet and Levin, 2013a; Moore et al., 2017). Despite highly diverse molecular and clinical manifestations, one common aspect points to the key: in carcinogenic transformation (Yamasaki et al., 1995; Ruch and Trosko, 2001), cells become isolated from the physiological signals that bind them into unified networks (the essential role of bioelectricity in this process is discussed in the next section). In the absence of global cues, they revert to their unicellular past, when their behaviors were aimed at optimizing the future of just one cell: proliferate as much as possible, and travel to whatever location affords the best local environment for nutrients and expansion (Davies and Lineweaver, 2011; Bussey et al., 2017; Zhou et al., 2018a). This is a breakdown of multicellularity and highlights the fact that the scale of the structure which cells work to maintain can change rapidly – from cooperation toward an entire organ system or body to the level of a single cell.

Normal bodies consist of networks of cells working together toward a unified goal - create and maintain specific large structures. In an important sense, each cell is integrated, via physiological signaling, into a coherent swarm intelligence with system-level (anatomical) goals. In cancer, the scope of the coherent Self of a cell reduces from being as large as the boundary of a whole body, to that of just one cell’s surface. The scope of the Self – the structure which a cell works to maintain and support – shrinks drastically in two main ways.

First, it shrinks spatially: being electrically isolated from their neighbors by a shutdown of gap junction synapses, a cell can neither measure distant events nor communicate across anatomical distances with other cells and with the outside world (Kull, 2000; Ay and Lohr, 2015). All of the attention (in terms of measurement and activity toward scaled goals) is focused at the single cell level, while the rest of the body is treated as the “external environment” (and all living things exploit their environment for their own benefit). Thus, the boundary between self and non-self (David Krakauer et al., 2014) can shift: multicellularity enlarges it, while carcinogenesis reduces it, and it is readily seen how this shifts the behavior of cells toward a mode that is not beneficial for the organism level.

Moreover, the size of the Self shrinks temporally: the time horizon of activity shifts from decades (somatic cells execute regenerative and repair processes that maintain a body up to ~100 years) to a much shorter time frame, as cancer cells undertake activity which may kill the host (and themselves) within a mere year. Normal cells pursue goals (organ maintenance) that can be much longer than a cell’s individual lifespan, especially in tissues with rapid cellular turnover – the time scale for such homeostatic activity is one that belongs to the collective, not the cells themselves. Shrinking their Self reduces the ability to work toward temporally distant goals (reduces their temporal horizon of concern) and, together with the inability to communicate as part of a distributed electrical network, makes them literally short-sighted – undertaking activity that will result in death of themselves as well as of the organism (except in the very rare cases of transmissible tumors). This view predicts that interference with (or restoration of) physiological communication among cells should be able to trigger (or suppress) cancer and that the relevant parameter (communication) is spread out over considerable distance and not confined to single cells (e.g., genomic damage). The general fact that actively patterning (embryonic and regenerative) environments can reprogram cancer cells to normal histogenesis has been known for decades (Illmensee and Mintz, 1976; Kasemeier-Kulesa et al., 2008; Oviedo and Beane, 2009).

However, the specific prediction of a role of real-time bioelectric communication in cancer has been confirmed in vivo in recent experiments: metastatic transformation of normal melanocytes can be achieved in genetically normal tadpoles simply by depolarizing a specific cell population (Lobikin et al., 2012), while, conversely, human oncogenes which normally induce tumors can be prevented from doing so simply by optogenetic or constitutive hyperpolarization (Chernet and Levin, 2013b, 2014; Chernet et al., 2015, 2016).

One important aspect should be stressed, in connection with the classic evolutionary concept of “selfishness” (Werren, 2011; Kourilsky, 2012): on the above-mentioned view, the cells in a metazoan organism body are not less selfish than unicellular organisms or cancer cells. They are equally selfish, working only for the benefit of themselves, but the relevant self that they defend is bigger in terms of space, time, and complexity. As will be argued below, what defines this Self is the boundary of information being able to pass between the subunits. It should be noted that this change is not binary and can occur at intermediate levels, e.g., teratoma tumors have coherent differentiated tissues (teeth, hair, skin, and muscle) but are not functional embryos, illustrating a level of integrated organization between that of cells and whole organs. It is an essential aspect of Scale-Free Cognition that most biological systems consist of multiple, nested selves [not one, as implied by Integrated Information Theory (Tononi, 2008)]. Each inner agent maintains its own local Self only to the extent that it regulates (restricts) information flow from neighbors and thus establishes unique states that do not coincide with those of the collective. Evolutionary and game-like dynamics occur within and between each level of organization, via cooperation and competition. While competition between cells in a body is only now beginning to be characterized in molecular detail (Gogna et al., 2015), the remarkable fact that different substructures in an otherwise unified organism do not all have the same goals and can work at cross-purposes was realized already at the very dawn of developmental biology by Roux, who presciently wrote of the struggle of the parts in an embryo (Heams, 2012).

Some biologists model cancer cells as individuals, while some see a tumor as its own organ (Egeblad et al., 2010), which can take over other cell types (Gabrilovich, 2017) in the same way that embryonic instructor cells control others in normal development (Vieira and McCusker, 2019). These possibilities could now be distinguished experimentally and analytically using advances in information theory that enable rigorous comparison of the causal power of models at different levels of description (Hoel et al., 2013, 2016; Moore et al., 2017). Identifying the most efficacious level of organization in a given system is critical for selecting targets in biomedical or engineering approaches to understand and control it. Is the best strategy chemotherapy, which seeks to identify and destroy irrevocably broken cells, or could there be normalization strategies that re-connect cells into a collective? The latter is a tantalizing possibility, as it has long been known that embryonic or regenerative environments can reprogram tumors in situ (Mintz and Illmensee, 1975; Kulesa et al., 2006). Motivating a system with inputs and experiences at the appropriate level is always easier than attempting to rewire complex systems with emergent dynamics in attempts to micromanage specific global outcomes. Understanding the decision-making, not only the molecular mechanisms, is essential and is perhaps one path toward resolving the unsatisfactory state of cancer medicine (Soto and Sonnenschein, 2013; Sonnenschein and Soto, 2015). Similar approaches are being taken with respect to swarms of insects, robots, and human beings (Couzin, 2009; Deisboeck and Couzin, 2009; Gomes et al., 2013; Rubenstein et al., 2014).

The ability of subunits to enter into a communication network with variable scale of boundaries is already being recognized in swarm dynamics, where the termite nest “superorganism is marked by a kind of extended physiology”, which supports regenerative repair (taking the large termite nest as the “body”), a kind of swarm cancer, and primitive cognition functions such as hypothesis testing carried out by the swarm during the repair process (Beekman and Oldroyd, 2008; Turner, 2016). As will be seen in the next section, this continuum between somatic pattern and cognitive systems reflects a deep scale invariance of cognitive concepts that apply as much to societies of cells as of societies of animals (Seeley, 2009; Turner, 2011). The view of the extent of active information as central to demarcating what exactly constitutes an Individual Self in a given context, especially the focus on a set of level-specific goal states that a system is able to pursue, suggests a formalism for identifying, categorizing, and comparing highly diverse types of Individuals.

I propose a definition of an Individual based on its information-processing structure (Barandiaran et al., 2009): the scale and types of goals that a system can pursue defines (determines) the boundaries and content of the putative “agent.” On this view, what defines a coherent, unified Self out of its constituent components and the surrounding environment is the set of parts that operate toward reaching specific goal states. Like a theorem, which has a holistic nature not shared by any of the axioms alone, a functional individual is more than the sum of its parts in the sense that goal-directed capacity emerges only from the integrated activity of all of the components (Weiss, 1967). This flows naturally from the conception of a goal-seeking system: having specified goal state(s), it is immediately clear than there must be an integrated system that can perform all of the parts of the Test-Operate-Test cycle, which may be invisible by inspecting each of the components in micro detail. This definition is orthogonal to other definitions, e.g., evolutionary or genetic ones, but having bigger goals (such as organ morphogenesis instead of single-cell proliferation) de-Darwinises cells and Darwinises groups, shifting the locus of selection and competition (Michod, 2007; Godfrey-Smith, 2009). I propose a working definition for the degree to which a system is a coherent Self, in terms of the goals (in the cybernetic sense) that the system seeks to achieve. Thus, a system’s ability to operate toward specific counterfactuals (future states that are not true right now but can be brought about through specific actions) is central to defining, understanding, and communicating with an arbitrary Individual, which might consist of many levels of organized components.