Technological Approach to Mind Everywhere: An Experimentally-Grounded Framework for Understanding Diverse Bodies and Minds

Developmental bioelectricity helps explain how free-living cells scaled their cell-level homeostatic pathways to whole body-level anatomical homeostasis (Levin, 2019). It has long been appreciated that evolvability is potentiated by modularity—the ability to trigger complex morphogenetic cascades by a simple “master” trigger that can be re-deployed in various contexts in the body (von Dassow and Munro, 1999). Recent advances reveal that bioelectric states can form very powerful master inducers that initiate self-limiting organogenesis. For example, the action of a single ion channel can induce an eye-specific bioelectric state that creates complete eyes in gut endoderm, spinal cord, and posterior tissues (Pai et al., 2012)—locations where genetic “master regulators” like the Pax6 transcription factor are insufficient in vertebrates (Chow et al., 1999). Likewise, misexpression of a proton pump (or a 1-h ionophore soak) to trigger bioelectric changes in an amputation wound can induce an entire 8-day cascade of building a complete tadpole tail (Adams et al., 2007; Tseng et al., 2010). This is control at the level of organ, not single cells’ fate specification, and does not require the experimenter to provide all of the information needed to build the complex appendage. Thus, bioelectric states serve as effective master regulators that evolution can exploit to make modular, large-scale changes in anatomy.

Moreover, because the same Vmem dynamics can be produced by many different ion channel combinations, and because bioelectric states propagate their influence across tissue distance during morphogenesis (Chernet and Levin, 2014; Pai et al., 2020), evolution is free to swap out channels and explore the bioelectrical state space: simple mutations in electrogenic genes can exert very long-range, highly coordinated changes in anatomy. Indeed, the KCNH8 ion channel and a connexin were identified in the transcriptomic analysis of the evolutionary shift between two functionally different morphologies of fin structures in fish (Kang et al., 2015). The evolutionary significance of bioelectric controls can also be seen across lineages, as some viruses evolved to carry ion channel and gap junction (Vinnexin) genes that enable them to hijack bioelectric machinery used by their target cells (Shimbo et al., 1996; Hover et al., 2017).

The unique computational capabilities of bioelectric circuits likely enabled the evolution of nervous systems, as specialized adaptations of the ancient ability of all cell networks to process electrical information as pre-neural networks (Keijzer, 2015; Fields et al., 2020). A full understanding of nervous system function must involve not only its genetics and molecular biology but also the higher levels of organization comprising dynamic physiology and computations involved in memory, decision-making, and spatio-temporal integration. The same is true for the rest of the body. For example, the realization that epithelia are the generators of bioelectric information (Robinson and Messerli, 1996) suggests models in which they act like a retina wrapped around a whole embryo (and individual organs) to preprocess electrical signals into larger-scale features and compute contrast information for downstream processing (Grossberg, 1978). The investigation of somatic bioelectric states as primitive “pattern memories” and the expansion of computational science beyond neurons will enrich the understanding of cell biology at multiple scales beyond molecular mechanisms, as is currently only done with respect to the brain (Marr, 1982). Generalizing the deep concepts of multiscale neuroscience beyond neurons (Grossberg, 1978; Pezzulo and Levin, 2015; Manicka and Levin, 2019b) is necessary for a better understanding of the tissue-level decision-making that drives adaptive development and regeneration. Conversely, advances in understanding information processing in a relatively simpler anatomical context will feed back to enrich important questions in brain science, shedding light on fundamental mechanisms by which information-processing agents (cells) work collectively to accomplish unified, complex system-level outcomes.

The multi-disciplinary opportunity here is not only to gain insight into the phylogeny of nervous systems and behavior, but to erase the artificial boundaries between scientific disciplines that focus on neurons vs. the rest of the body, with the direct consequence that a more inclusive, gradualist picture emerges of the mechanisms commonly associated with cognitive Selves.

Ion channels and gap junctions are the hardware interface to the bioelectric computational layer within living systems. Like a retina for a brain, or a keyboard for a computer, they allow transient signals to serve as inputs to memory and decision-making networks. For any given agent (cell, tissue, etc.), its bioelectrical interface is accessed by a number of potential users. First are the neighboring agents, such as other tissues, which pass on their bioelectric state during cooperative and competitive interactions in morphogenesis. There are also commensal and parasitic microbes, which have evolved to hijack such control systems to manipulate the anatomy of the host—like the naïve bacteria on planaria that can determine head number and visual system structure in flatworm regeneration (Williams et al., 2020). Moreover, the development of pharmacological, genetic, and optogenetic tools now allows human bioengineers to access bioelectrical circuits for the control of growth and form in regenerative medicine and synthetic bioengineering contexts (Adams et al., 2013, 2014, 2016; Chernet et al., 2016; McNamara et al., 2016; Bonzanni et al., 2020). All of these manipulations can serve as catalysts, enabling an evolutionary lineage to more easily travel to regions of option space that might otherwise be separated by an energy barrier that is difficult for standard evolution to reach. In this sense, cognitive properties of developmental mechanisms help us to understand problem-solving on phylogenetic, not just ontogenetic, timescales. We next look at specific ways in which the architecture of multiscale autonomy, especially as implemented by bioelectric network mechanisms, potentiates evolution.

Deterministic chaos and complexity theory have made it very clear why bottom-up control of even simple systems (e.g., 3-body problem) can be practically impossible. This inverse problem (Lobo et al., 2014)—what control signals would induce the desired change—is not only a problem for human engineers but also for adjacent biological systems such as the microbiome or a fungus that seeks to control the behavior of an ant (Hughes et al., 2016), and most of all, for other parts of a complex system (to help control itself). Evolution tackles this task by exploiting a multiscale competency architecture (MCA), where subunits making up each level of organization are themselves homeostatic agents. It’s built on an extremely powerful design principle: error correction (Fields and Levin, 2017; Frank, 2019a,b).

The key aspect of biological modularity is not simply that complex subroutines can be triggered by simple signals, making it easy to recombine modules in novel ways (Schlosser and Wagner, 2004; Gerhart and Kirschner, 2007), but that these modules are also themselves sub-agents exhibiting infotaxis and socialtaxis, and solving problems in their own spaces (Vergassola et al., 2007; Gottlieb et al., 2013; Karpas et al., 2017). When an eye primordium appears in the wrong place (e.g., a tadpole tail), it still forms a correctly patterned, functional organ, manages to get its data to the brain (via spinal cord connection) to enable vision (Figure 6E), and (if somewhere in the head) moves to the correct place during metamorphosis (Vandenberg et al., 2012). When cells are artificially made to be very large and have several times the normal genetic material, morphogenesis adapts to this and still builds an overall correct animal (Fankhauser, 1945a,b). These are goal-directed (in the cybernetic sense) processes because the system can reach a specific target morphology (and functionality) state despite perturbations or changes in local/starting conditions or the basic underlying components. Regeneration is the most familiar example of this, but is just a special case of the broader phenomenon of anatomical homeostasis. Homeostatic loops operating over large-scale anatomical states have several (closely related) key implications for the power and speed of evolution.

First, it greatly smoothes the fitness landscape. Consider two types of organisms: one whose subsystems mechanically follow a hardwired (genetically-specified) set of steps (A, passive, or merely structural modularity), and one whose modules optimize a reward function (B, multi-scale competency of modules). Mutations that would be detrimental in A (e.g., because they move the eye out of its optimal position) are neutral in B, because the competency of the morphogenetic subsystems repositions the eye even if it starts out somewhere else. Thus, MCA shields from selection some aspects of mutations’ negative effects (which inevitably are the bulk of random mutations’ consequences). The primary reason for the anatomical homeostasis seen in regulative development and regeneration may be for dealing, not with damage, but with deviations from target morphology induced by mutations. This is certainly true at the scale of tissues during the lifetime of an individual [as in the inverse relationship between regeneration and cancerous defection from large-scale target morphology (Levin, 2021b)], but may be true on evolutionary time scales as well.

Second, MCA reduces apparent pleiotropy—the fact that most mutations have multiple effects (Boyle et al., 2017). For example, a change in an important canonical signaling pathway such as Wnt or BMP (Raible and Ragland, 2005) is going to have numerous consequences for an organism. Suppose a mutation appears that moves the mouth off of its optimal position (bad for fitness) but also has some positive (adaptive) feature elsewhere in the body. In creatures of type A, the positive aspects of that mutation would never be seen by selection because the malfunctioning mouth would reduce the overall fitness or kill the individual outright. However, in creatures of type B, the mouth could move to its optimal spot (Vandenberg et al., 2012), enabling selection to independently evaluate the other consequence of that mutation. Creatures possessing MCA could reap the benefit of positive consequences of a mutation while masking its other effects via local adjustments to new changes that reduce the penalties (an important kind of buffering). In effect, evolution doesn’t have to solve the very difficult search problem of “how to improve feature X without touching features Y. Z which already work well,” and reaps massive efficiency (time) savings by not having to wait until the search process stumbles onto a way to directly encode an improvement that is either isolated from other features, or improves them all simultaneously (Wagner et al., 2007; Melo et al., 2016).

Third, MCA allows systems not only to solve problems, but also to exploit opportunities. A lineage has the chance to find out what pro-adaptive things a mutation can do, because competency hides the negative consequences. This gives time for new mutations to appear that hardwire the compensatory changes that had to be applied—an analogy to the proposed Baldwin effect (Hogenson, 2001; Downing, 2004; Robinson and Barron, 2017). This enables the opportunity to exploit the possibility space more freely, providing a kind of patience or reduction of the constraint that evolutionary benefits have to be immediate in order to propagate—it effectively reduces the short-sightedness of the evolutionary process. Indeed, multiscale competency is beneficial not only for natural evolution, but also for soft robotics and synthetic bioengineering because it helps cross the sim-to-real gap: models do not have to be 100% realistic to be predictively useful if the component modules can adaptively make up for some degree of deficiency in the controller design (Brooks, 1986).

Fourth, the homeostatic setpoint-seeking architecture makes the relationship between genotype and anatomical phenotype more linear (Muller and Schuppert, 2011; Lobo et al., 2014), improving controllability (Liu et al., 2011; Gao et al., 2014; Posfai et al., 2016). By using a top-down control layer to encode the patterns to which competent subunits operate, living systems do not need to solve the difficult inverse problems of what signals to send their subsystems to achieve high-level outcomes. Bioelectric pattern memories (such as the voltage distribution that tells wild-type planarian cells whether to build 1 head or 2) exploit a separation of data from the machine itself, which makes it much easier to make changes. Evolution does not need to find changes at the micro level but can also simply change the information encoded in the setpoints, such as the electric face prepattern (Vandenberg et al., 2012), which allows it to re-use the same exact implementation machinery to build something that can be quite different. The ability to rely on a non-zero IQ for your component modules (thus delegating and offloading complex regulatory chains) is an important affordance (Watson et al., 2010; Friston et al., 2012) for the evolutionary process. It means that the larger system’s evolution is in effect searching an easier, less convoluted control, signaling, or reward space—this massive dimensionality reduction offers the same advantages human engineers have with agents on the right side of the persuadability scale. It is no accident that learning in the brain, and behavioral systems, eventually exapted this same architecture and indeed the exact same bioelectrical machinery to speed up the benefits of evolution.

A significant brake on the efficiency of evolution, as on machine learning (indeed, all learning) is credit assignment: which change or action led to the improvement or reward? When a collection of cells known as a “rat” learns to press a level and get a reward, no individual cell has the experience of interacting with a lever and receiving the nutrient. What enables the associative memory in this collective intelligence are the delay lines (nervous system) between the paws and the intestinal lining which provide a kind of patience—a tolerance of the temporal delay between the action and the reward and the ability to link extremely diverse modules on both ends (different kinds of actions can be linked to arbitrary rewards). MCA does the same thing for evolutionary learning (Watson et al., 2014, 2016; Power et al., 2015; Watson and Szathmary, 2016; Kouvaris et al., 2017), making it easier for systems to reap selection rewards for arbitrary moves in genotype space. This effectively raises the IQ of the evolutionary search process. Much as (Figure 5) an agent’s sophistication can be gauged by how expertly and efficiently it navigates an arbitrary search space and its local optima, the traversal of the evolutionary search process can be made less short-sided by homeostatic activity within the physiological layer that sits between genotype and phenotype.

There is an adaptation tradeoff between robustness (e.g., morphogenesis to the same pattern despite interventions and changing conditions) and responsiveness to environment (context sensitivity), perhaps similar to the notion of criticality (Beggs, 2008; Hankey, 2015). The plasticity and goal-directedness of modules (as opposed to hardwired patterns) serve to reduce the sim-to-real gap (Kriegman et al., 2020b): because the current environment always offers novel challenges compared to prior experiences which evolution (or human design) uses to prepare responses, the MCA architecture doesn’t take history too seriously, relying on plasticity and problem-solving more than on fine-tuning micromodels of what to do in specific cases. Biology reaps the benefits of both types of strategies by implementing anatomical homeostasis that coarse-grains robustness by making stability applying to large outcomes, such as overall anatomy, not to the microdetails of cell states. The scaling of homeostatic loops makes it possible to achieve both: consistent results and environmental sensitivity. These dynamics apply in various degrees to the numerous nested, adjacent, and overlapping sub-agents that make up any biological system. Cooperation results not from altruistic actions between Selves, but by the expansion of the borders of a single Self via scaling of the homeostatic loops. On this view, cancer cells are not more selfish than tissues—they are all equally selfish, but maintain goals appropriate to smaller scales of Selves. Indeed, even the parts of one normal body don’t perfectly cooperate—this is as true in development (Gawne et al., 2020) as it is in cognitive science (Dorahy et al., 2014; Reinders et al., 2018, 2019).

A picture is emerging of how evolution exploits the local competency of modules, competing and cooperating, to scale these subsystems’ sensing, actuation, and setpoint memories to give rise to coherent larger-scale Selves. Overall, the TAME framework addresses functional aspects only, and is compatible with several views on phenomenal consciousness in compound Selves (Chalmers, 1996). However, it does have a few implications for the study of Consciousness.

While the TAME framework focuses on 3rd-person observable properties, it does make some commitments to ways of thinking about consciousness. Provisionally, I suggest that consciousness also comes in degrees and kinds (is not binary) for the same reasons argued for continuity of cognition: if consciousness is fundamentally embodied, the plasticity and gradual malleability of bodies suggests that it is a strong requirement for proponents of phase transitions to specify what kind of “atomic” (not further divisible) bodily change makes for a qualitative shift in capacity consciousness. Another implication of TAME is that while “embodiment” is critical for consciousness, it is not restricted to physical bodies acting in 3D space, but also includes perception-action systems working in all sorts of spaces. This implies, counter to many people’s intuitions, that systems that operate in morphogenetic, transcriptional, and other spaces should also have some (if very minimal) degree of consciousness. This in turn suggests that an agent, such as a typical modern human, is really a patchwork of many diverse consciousnesses, only one of which is usually capable of verbally reporting its states (and, not surprisingly, given its limited access and self-boundary, believes itself to be a unitary, sole owner of the body).

What is necessary for consciousness? TAME’s perspective is fundamentally that of the primacy of goal-directed activity. Thus, consciousness accompanies specific types of cognitive processes which exert energy toward goals, but as described above, those processes can take forms very divergent from our typical brain-centered view. Unlike other panpsychist views, TAME does not claim that mind is inevitably baked in regardless of physical implementation or structure. Causal structure and cybernetic properties of the embodiment are key determinants of consciousness capacity. However, as the minimal degree of internal self-determination and goal-directedness is apparently present even in particles (Feynman, 1942; Georgiev and Georgiev, 2002; Ogborn et al., 2006; Kaila and Annila, 2008; Ramstead et al., 2019; Kuchling et al., 2020a), there may be no true “0” on the scale of consciousness in the universe. While simple accretion does not magnify the nano-goal-directed activity and indeterminate action of particles (e.g., rocks are not more conscious, and probably less, than particles in specific contexts), biological organization does amplify it, resulting in scaling up of sentience.

Of course, these implications will be unpalatable conclusions for many. It should be kept in mind that TAME is compatible with several different views on consciousness, and does not need to commit to one specific philosophy. It is fundamentally a framework for enabling empirical experiments, and its practical utility remains, regardless of the above speculations. Moreover, I remain skeptical about being able to say anything definitive about consciousness per se (as distinct from correlates of consciousness) from a 3rd-person, objective perspective. Thus, however unappealing the above view may be, I see no way of rigorously showing why any other claim about consciousness and what it requires is demonstrably better.

An emphasis on somatic plasticity has additional practical implications, being focused on the functional splitting and joining of agents’ parts. For example, the ancient question of “where does it all come together?” in the brain, with respect to the unified character of consciousness, is one of those pseudo-problems that is dispelled by a framework like TAME that focuses on multi-scale architecture. How big should a place where it all comes together be? If it can be ∼140 mm wide, then the answer is, the whole brain. One could decide that it should be smaller (the human pineal gland is ∼7 mm wide), but then the question is, why not smaller still—given the cellular components of the pineal (or any piece of the brain) and the molecular organelles inside a pineal gland cell, one would always need to ask “but where does it all come together inside there?” of whatever piece of the brain is taken to be the seat of consciousness. The multi-scale nature of biology means that there is no privileged size scale for any homunculus.

Another important idea with respect to consciousness is “What is it like to be” a given agent (Nagel, 1974). Sensory augmentation, neural link technologies, and bioengineering produce tractable model systems in novel cognitive architectures, such as 2-headed planaria where the brains are connected by a central nervous system (Figure 7B), to help study the functional aspects of this cognitive re-shuffling. TAME’s focus on the fact that all cognitive architectures are inevitably composites emphasizes that the parts can be rearranged; thus, the Subject of cognition can change “on the fly,” not merely during evolutionary timescales. Thus, the basic question of philosophy of mind—what’s it like to be animal X (Nagel, 1974)—is just a first-order step on a much longer journey. The second-order question is, what’s it like to be a caterpillar, slowly changing into a butterfly as its brain is largely dissolved and reassembled into a different architecture for an animal whose sense organs, effectors, and perhaps overall Umwelt is completely different. All of this raises fascinating issues of first person experience not only in purely biological metamorphoses (such as human patients undergoing stem cell implants into their brains), but also technological hybrids such as brains instrumentized with novel sensory arrays, robotic bodies, software information systems, or brains functionally linked to other brains (Warwick et al., 1998; Demarse et al., 2001; Potter et al., 2003; Bakkum et al., 2007a,b; Tsuda et al., 2009; Cohen-Karni et al., 2012; Giselbrecht et al., 2013; Aaser et al., 2017; Ricotti et al., 2017; Ding et al., 2018; Mehrali et al., 2018; Anderson et al., 2020; Ando and Kanzaki, 2020; Merritt et al., 2020; Orive et al., 2020; Saha et al., 2020; Dong et al., 2021; Li et al., 2021; Pio-Lopez, 2021).

The developmental approach to the emergence of consciousness on short, ontogenetic timescales complements the related question on phylogenetic timescales, and is likely to be a key component of mature theories in this field.

Most surprisingly, the plasticity and capacity for bioengineering and chimerization (recombination of biological and engineered parts in novel configurations) erases the sharp divide between first person and third person perspectives. This has been a fundamental, discrete distinction ever since Descartes, but the capacity for understanding and creating new combinations shows a continuum even in this basic distinction (Figure 11). The fact that Selves are not monadic means we can share parts with our subject of inquiry. If one has to be a system in order to truly know what it’s like to be that system (1st person perspective), this is now possible, to various degrees, by physically merging one’s cognitive architecture with that of another system. Of course, by strongly coupling to another agent, one doesn’t remain the same and experience the other’s consciousness; instead, a new Self is created that is a composite of the two prior individuals and has composite cognition. This is why consciousness research is distinct in strong degree from other scientific topics. One can observe gauges and instruments for 3rd-person science and remain the same Self (largely; the results of the observation may introduce small alterations in the cognitive structure). However, data on 1st person experiential consciousness cannot be taken in without fundamentally changing the Self (being an effective homunculus by watching the neuroscience data corresponding to the movies inside the heads of other people is impossible for the same reason that there is no homunculus in each of our heads). The study of consciousness, whether done via scientific tools or via the mind’s own capacity to change itself, inevitably alters the Subject.

Thus, standard (3rd-person) investigations of this process leave open the ancient question as to whether specific upgrades to cognition induce truly discontinuous jumps in consciousness. The TAME framework is not incompatible with novel discoveries about sharp phase transitions, but it takes the null hypothesis to be continuity, and it remains to be seen whether contrary evidence for truly sharp upgrades in consciousness can be provided. Future, radical brain-computer interfaces in human patients are perhaps one avenue where a subject undergoing such a change can convince themselves, and perhaps others, that a qualitative, not continuous, change in their consciousness had occurred.

With respect to the question of consciousness per se, as opposed to neural or behavioral correlates of consciousness, we have one major functional tool: general anesthesia. It is remarkable that we can readily induce a state in which all the individual cells are fine and healthy, but the larger Self is simply gone [although, some of the parts can continue to learn during this time (Ghoneim and Block, 1997)]. Interestingly, general anesthetics are gap junction blockers (Wentlandt et al., 2006): consistent with the cognitive scaling example above, shutting down electrical communication among the cells leads to a disappearance of the higher-level computational layer while the cellular network is disrupted. GJ blockers are used to anesthetize living beings ranging across plants, Hydra, and human subjects (Gremiaux et al., 2014). It is amazing that the same Self (with memories and other properties) returns, when the anesthetic is removed. Of course, the Self does not return immediately, as shown by the many hallucinatory (Saniova et al., 2009; Kelz et al., 2019) experiences of people coming out of general anesthesia—it takes some time for the brain to return to the correct global bioelectric state once the network connections are allowed again (meta-stability) (Rabinovich et al., 2008). Interestingly, and in line with the proposed isomorphism between cognition and morphogenesis, gap junction blockade has exactly this effect in regeneration: planaria briefly treated with GJ blocker regenerate heads of other species, but eventually snap out of it and remodel back to their correct target morphology (Emmons-Bell et al., 2015). It is no accident that the same reagents cause drastic changes in the high-level Selves in both behavioral and morphogenetic contexts: evolution uses the same scheme (GJ-mediated bioelectrical networks) to implement both.

The epistemic problem of Other Minds has been framed to imply that we cannot directly ever be sure how much or what kind of consciousness exists in any particular system under study. The TAME framework reminds us that this is true even for components of ourselves (like the non-verbal brain hemisphere). Perhaps the confabulation system enables one part of our mind to estimate the agency of other parts (the feelings of consciousness and free will) and develop models useful for prediction and control, applying in effect the empirical criteria for persuadability internally. The ability to develop a “theory of mind” about external agents can readily be turned inward, in a composite Self.

Are all cognitive systems conscious? The TAME framework is compatible with several views on the nature of consciousness. However, the evolutionary conservation of mechanisms between brains and their non-neural precursors has an important consequence for the question of where consciousness could be found. To the extent that one believes that mechanisms in the brain enable consciousness, all of the same machinery and many similar functional aspects are found in many other places in the body and in other constructs. TAME emphasizes that there is no principled way to restrict consciousness to “human-like, full-blown sophisticated brains,” which means one has to seriously consider degrees of consciousness in other organs, tissues, and synthetic constructs that have the same features neurons and their networks do (Trewavas and Baluska, 2011; Baluska et al., 2016, 2021; Baluska and Reber, 2019). The fundamental gradualism of this framework suggests that whatever consciousness is, some variant and degree thereof has to be present very widely across autopoietic systems. TAME is definitely incompatible with binary views that cut off consciousness at a particular sharp line and it suggests no obvious way to define cognitive systems that have no consciousness whatsoever. A big open question is whether the continuum of cognition (and consciousness) contains a true “0” or only infinitesimal levels for very modest agents. One is tempted to imagine what properties a truly minimal agent would have to have; not being fully constrained by local forces, and ability to pursue goals, both seem key, and both of these are present to a degree in even single particles (via quantum indeterminacy and least action behavior). The type and degree of scaling (or lack thereof) of these capacities in bulk inorganic matter vs.

highly-organized living forms is a fertile area for future development of TAME and will be explored in forthcoming work.

Regenerating, physiological, and behaving systems use effort (energy) to achieve defined, adaptive outcomes despite novel circumstances and unpredictable perturbations. That is a key invariant for cognition; differences in substrate, scale, or origin story among living systems are not fundamental, and obscure an important way to unify key properties of life: the ability to deploy intelligence for problem-solving in diverse domains. Modern theories of Mind must eventually handle the entire option space for intelligent agents, which not only contains the familiar advanced animals we see on Earth, but can also subsume ones consisting of radically different materials, ones created by synthetic bioengineering or combinations of evolution and rational design in the lab, and ones of exobiological as well as possible terrestrial origins. The advances of engineering confirm and put into practice an idea that was already entailed by evolution: that cognitive traits, like all other traits, evolved from humbler variants, forming a continuum. There are no biologically-valid binary categories in this space. Take the prevalent legal definition of human “adults,” who snap into being at the age of 18; such binary views on cognitive properties are fictitious coarse-grainings useful for our legal system to operate, but no more than that. There is no bright line between “truly cognitive” and “pseudo cognitive” that can ever be drawn between two successive members of an evolutionary lineage. The error of “committing Anthropomorphism” is a pseudo-scientific “folk” notion useful for only the most trivial examples of failure to scale down complex claims proportionally to simpler systems; engineering requires us to determine what level of cognitive model enables the most fruitful prediction and control.

Every intelligence is a collective intelligence, and the modular, multi-scale architecture of life means that we are a holobiont in more than just the sense of having a microbiome (Chiu and Gilbert, 2015)—we are all patchworks of overlapping, nested, competing, and cooperating agents that have homeostatic (goal-directed) activity within their self-constructed virtual space at a scale that determines their cognitive sophistication. A highly tractable model system for unconventional cognition, in which these processes and the scaling of Selves can not only be seen but can also be manipulated, is morphogenetic homeostasis. The process of construction and remodeling (toward anatomical features) of cellular collectives shows crucial isomorphism to cognitive aspects of the many-into-one binding like credit assignment, learning, stress reduction, etc. The partial wiping of ownership information on permanent signals makes gap junctional coupling an excellent minimal model system for thinking about biological mechanisms that scale cognition while enabling co-existence of subunits with local goals (multiple levels of overlapping Selves, whose scale and borders are porous and can change during the lifetime of the agent). However, many other substrates can no doubt fulfill the same functions.

The TAME framework is conceptually incomplete in important ways. On-going development is proceeding along lines including merging with other frameworks such as Active Inference (Friston, 2013; Badcock et al., 2019; Ramstead et al., 2019), Rosen’s (M,R) and Anticipatory Systems (Rosen, 1973, 1979, 1985; Nasuto and Hayashi, 2016), and recent advances in information theory as applied to individuality and scaling of causal power (Hoel et al., 2013, 2016; Krakauer et al., 2014; Daniels et al., 2016). It will be critical to more rigorously develop the waypoints along the Persuadability Continuum, including understanding of what an “increased capacity” human (or non-human) would be like, in contrast to the “diminished capacity” with which we are well familiar from legal proceedings [the right side of the continuum, corresponding to radically expanded cognitive light cones (Śāntideva Bstan ’dzin rgya m and Comité de traduction Padmakara, 2006)].

The TAME framework suggests numerous practical research directions immediately within reach (some of which are already pursued in our group), including developing biomedically-actionable models of morphogenetic plasticity and robustness as meta-cognitive error correction mechanisms, tissue training paradigms for anatomical and physiological outcomes, exploiting learning properties of pathway models for regenerative medicine (Herrera-Delgado et al., 2018; Biswas et al., 2021), and creation of AI platforms based on multi-scale agency architectures that do not rely on neuromorphic principles.