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

For example, tadpoles turn into frogs by rearranging their craniofacial structures: the eyes, nostrils, and jaws move as needed to turn a tadpole face into a frog face (Figure 6B). Guided by the hypothesis that this was not a hardwired but an intelligent process that could reach its goal despite novel challenges, we made tadpoles in which these organs were in the wrong positions—so-called Picasso Tadpoles (Vandenberg et al., 2012). Amazingly, they tend to turn into largely normal frogs because the craniofacial organs move in novel, abnormal paths [sometimes overshooting and needing to return a bit (Pinet et al., 2019)] and stop when they get to the correct frog face positions. Similarly, frog legs that are artificially induced to regenerate create a correct final form but not via the normal developmental steps (Tseng and Levin, 2013). Students who encounter such phenomena and have not yet been inoculated with the belief that molecular biology is a privileged level of explanation (Noble, 2012) ask the obvious (and proper) question: how does it know what a correct face or leg shape is?

Examples of remodeling, regulative development (e.g., embryos that can be cut in half and produce normal monozygotic twins), and regeneration, ideally illustrate the goal-directed nature of cellular collectives. They pursue specific anatomical states that are much larger than any individual cells and solve problems in morphospace in a context-sensitive manner—any swarm of miniature robots that could do this would be called a triumph of collective intelligence in the engineering field. Guided by the TAME framework, two questions come within reach. First, how does the collective measure current state and store the information about the correct target morphology? Second, if morphogenesis is not at the clockwork level on the continuum of persuadability but perhaps at that of the thermostat, could it be possible to re-write the setpoint without rewiring the machine (i.e., in the context of a wild-type genome)?

Deer farmers have long known of trophic memory: wounds made on a branched antler structure in 1 year, will result in ectopic tines growing at that same location in subsequent years, long after the original rack of antlers has fallen off (Bubenik and Pavlansky, 1965; Bubenik, 1966; Lobo et al., 2014). This process requires cells at the growth plate in the scalp to sense, and remember for months, the location of a transient damage event within a stereotypical branched structure, and reproduce it in subsequent years by over-riding the wild-type stereotypical growth patterns of cells, instead guiding them to a novel outcome. This is an example of experience-dependent, re-writable pattern memory, in which the target morphology (the setpoint for anatomical homeostasis) is re-written within standard hardware.

Planarian flatworms can be cut into multiple pieces, and each fragment regenerates precisely what is missing at each location (and re-scales the remaining tissue as needed) to make a perfect little worm (Cebrià et al., 2018). Some species of planaria have an incredibly messy genome—they are mixoploid due to their method of reproduction: fission and regeneration, which propagates any mutations that don’t kill the stem cell and expands it throughout the lineage [reviewed in Fields et al. (2020)]. Despite the divergence of genomic information, the worms are champion regenerators, with near 100% fidelity of anatomical structure. Recent data have identified one set of mechanisms mediating the ability of the cells to make, for example, the correct number of heads: a standing bioelectrical distribution across the tissue, generated by ion channels and propagated by electrical synapses known as gap junctions (Figures 7A–D). Manipulation of the normal voltage pattern by targeting the gap junctions (Sordillo and Bargmann, 2021) or ion channels can give rise to planaria with one, two, or 0 heads, or heads with shape (and brain shape) resembling other extant species of planaria (Emmons-Bell et al., 2015; Sullivan et al., 2016). Remarkably, the worms with abnormal head number are permanently altered to this pattern, despite their wild-type genetics: cut into pieces with no further manipulations, the pieces continue to regenerate with abnormal head number (Oviedo et al., 2010; Durant et al., 2017). Thus, much like the optogenetic techniques used to incept false behavioral memories into brains (Vetere et al., 2019), modulation of transient bioelectric state is a conserved mechanism by which false pattern memories can be re-written into the genetically-specified electrical circuits of a living animal.

A key feature of morphogenesis is that diverse underlying molecular mechanisms can be deployed to reach the same large-scale goal. This plasticity and coarse-graining over subunits’ states is a hallmark of collective cognition, and is also well known in neuroscience (Prinz et al., 2004; Otopalik et al., 2017). Newt kidney tubules normally have a lumen of a specific size and are made up (in cross section) of 8–10 cells (Fankhauser, 1945a,b). When the cell size is experimentally enlarged, the same tubules are made of a smaller number of the bigger cells. Even more remarkable than the scaling of the cell number to unexpected size changes (on an ontogenetic, not evolutionary, timescale) is the fact that if the cells are made really huge, just one cell wraps around itself and still makes a proper lumen (Figure 6D). Instead of the typical cell-cell interactions that coordinate tubule formation, cytoskeletal deformations within one cell can be deployed to achieve the same end result. As in the brain, the levels of organization exhibit significant autonomy in the details of their molecular activity but are harnessed toward an invariant system-level outcome.

The plasticity of morphogenesis is significantly isomorphic to that of brains and behavior because the communication dynamics that scale individual neural cells into a coherent Self are ones that evolution honed long before brains appeared, in the context of morphogenic control (Fields et al., 2020), and before that, in metabolic control in bacterial biofilms (Prindle et al., 2015; Liu et al., 2017; Martinez-Corral et al., 2019; Yang et al., 2020). Each genome specifies cellular hardware that implements signaling circuits with a robust, reliable default “inborn” morphology—just as genomes give rise to brain circuits that drive instinctual behavior in species that can build nests and do other complex things with no training. However, evolution selected for hardware that can be reprogrammed by experiences, in addition to its robust default functional modes—in body structure, as well as in brain-driven behavior. Many of the brain’s special features are to be found, unsurprisingly, in other forms outside the central nervous system. For example, mirror neurons and somatotopic representation are seen in limbs’ response to injury, where the type and site of damage to one limb can be read out within 30 s from imaging the opposite, un-injured limbs (Busse et al., 2018). Table 2 shows the many parallels between morphogenetic and cognitive systems.

The view of anatomical homeostasis as a collective intelligence is not a neutral philosophical viewpoint—it makes strong predictions, some of which have already borne fruit. It led to the discovery of reprogrammable head number in planaria (Nogi and Levin, 2005) and of pre-neural roles for serotonin (Fukumoto et al., 2005a,b). It explains the teratogenicity for pre-neural exposure to ion channel or neurotransmitter drugs (Hernandez-Diaz and Levin, 2014), the patterning defects observed in human channelopathies in addition to the neurological phenotypes [reviewed in Srivastava et al. (2020)], and the utility of gap junction blockers as general anesthetics.

Prediction derived from the conservation and scaling hypotheses of TAME can be tested via bioinformatics. Significant and specific overlap are predicted for genes involved in morphogenesis and cognition (categories of memory and learning). This is already known for ion channels, connexin (gap junction) genes, and neurotransmitter machinery, but TAME predicts a widespread re-use of the same molecular machinery. Cell-cell communication and cellular stress pathways should be involved in internal conflict between psychological modules (Reinders et al., 2019) and social behavior, while memory genes should be identified in genetic investigations of cancer, regeneration, and embryogenesis.

Another key prediction that remains to be tested (ongoing in our lab) is trainability of morphogenesis. The collective intelligence of tissues could be sophisticated enough to be trainable via reinforcement learning for specific morphological outcomes. Learning has been suggested by clinical data in the heart (Zoghi, 2004), bone (Turner et al., 2002; Spencer and Genever, 2003), and pancreas (Goel and Mehta, 2013). It is predicted that using rewards and punishments (with nutrients/endorphins and shock), not micromanagement of pathway hardware, could be a path to anatomical control in clinical settings, whether for morphology or for gene expression (Biswas et al., 2021). This would have massive implications for regenerative medicine, because the complexity barrier prevents advances such as genomic editing from impacting e.g., limb regeneration in the foreseeable future. The same reasons for which we would train a rat for a specific behavior, rather than control all of the relevant neurons to force it to do it like a puppet, explain why the direct control of molecular hardware is a far more difficult biomedical path than understanding the sets of stimuli that could motivate tissues to build a specific desired structure.

The key lesson of computer science has been that even with hardware we understand (if we built it ourselves), it is much more efficient and powerful to understand the software and evince desired outcomes by the appropriate stimulation and signaling, not physical rewiring. If the hardware is reprogrammable (and it is here argued that much of the biological hardware meets this transition), one can offload much of the complexity onto the system itself, taking advantage of whatever competence the sub-modules have. Indeed, neuroscience itself may benefit from cracking a simpler version of the problem, in the sense of neural decoding, done first in non-neural tissues.

The hardware of the brain consists of ion channels which set the cells’ electrical state, and controllable synapses (e.g., gap junctions) which can propagate those states across the network. This machinery, including the neurotransmitters that eventually transduce these computations into transcriptional and other cell behaviors, is in fact highly conserved and present in all cells, from the time of fertilization (Figures 7C,D). A major difference between neural and non-neural bioelectricity is the time constant with which it acts [brains speed up the system into millisecond scales, while developmental voltage changes occur in minutes or hours (Harris, 2021; Levin, 2021a)]. Key aspects of this system in any tissue that enable it to support flexible software include the fact that both ion channels and gap junctions are themselves voltage sensitive—in effect, they are transistors (voltage-gated current conductances). This enables evolution to exploit the laws of physics to rapidly generate very complex circuits with positive (memory) and negative (robustness) feedback (Law and Levin, 2015; Cervera et al., 2018, 2019a,b,2020a). The fact that a transient voltage state passing through a cell can set off a cycle of progressive depolarization (like an action potential) or gap junctional (GJ) closure means that such circuits readily form dynamical systems memories which can store different information and change their computational behavior without changing the hardware (i.e., not requiring new channels or gap junctions) (Pietak and Levin, 2017); this is obvious in the action potential propagations in neural networks but is rarely thought about in development.

It should be noted that there are many additional biophysical modalities, such as parahormones, volume conduction, biomechanics (strain and other forces), cytoskeletal dynamics, and perhaps even quantum coherence events that could likewise play interesting roles. These are not discussed here only due to length limitations; instead, we are focusing on the bioelectric mechanisms as one particularly illustrative example of how evolution exploits physics for computation and cognition.

Consistent with its proposed role, slowly-changing resting potentials serve as instructive patterns guiding embryogenesis, regeneration, and cancer suppression (Bates, 2015; Levin et al., 2017; McLaughlin and Levin, 2018). In addition to the pattern memories encoded electrically in planaria (discussed above), bioelectric prepatterns have also been shown to dictate the morphogenesis of the face, limbs, and brain, and function in determining primary body axes, size, and organ identity [reviewed in Levin and Martyniuk (2018)]. One of the most interesting aspects of developmental bioelectricity is its modular nature: very simple voltage states trigger complex, downstream patterning cascades. As in the brain, modularity goes hand-in-hand with pattern completion: the ability of such networks to provide entire behaviors from partial inputs. For example, Figure 7F shows how a few cells transduced with an ion channel that sets them into a “make the eye here” trigger recruit their neighbors, in any region of the body, to fulfill the purpose of the subroutine call and create an eye. Such modularity makes it very easy for evolution to develop novel patterns by re-using powerful triggers. Moreover, as do brains, tissues use bioelectric circuits to implement pattern memories that set the target morphology for anatomical homeostasis (as seen in the planarian examples above). This reveals the non-neural material substrate that stores the information in cellular collectives, which is a distributed, dynamic, re-writable form of storage that parallels recent discoveries of how group knowledge is stored in larger-scale agents such as animal swarms (Thierry et al., 1995; Couzin et al., 2002).

Finally, bioelectric domains (Pai et al., 2017, 2018; Pitcairn et al., 2017; McNamara et al., 2019, 2020) set the borders for groups of cells that are going to complete a specific morphogenetic outcome—a system-level process like “make an eye.” They define the spatio-temporal borders of the modular activity, and suggest a powerful model for how Selves scale in general.

Gap junctional connections between cells provide an interesting case study for how the borders of the Self can expand or contract, in the case of a morphogenetic collective intelligence (Figure 8). Crucially, gap junctions [and gap junctions extended by tunneling nanotubes (Wang et al., 2010; Ariazi et al., 2017)] enable a kind of cellular parabiosis—a regulated fusion between cells that enables lateral inheritance of physiological information, which speeds up processing in the same way that lateral gene inheritance potentiates change on evolutionary timescales. The following is a case study hypothesizing one way in which evolution solves the many-into-one problem (how competent smaller Selves bind into an emergent higher Self), and how this process can break down leading to a reversal (shrinking) of the Self boundary (summarized in Table 3).

Single cells (e.g., the protozoan Lacrymaria olor) are very competent in handling morphological, physiological, and behavioral goals on the scale of one cell. When connected to each other via gap junctions, as in metazoan embryos, several things happen (much of which is familiar to neuroscientists and workers in machine learning in terms of the benefits of neural networks) that lead to the creation of a Self with a new, larger cognitive boundary. First, when cells join into an electrochemical network, they can now sense events, and act, on a much larger physical “radius of concern” than a single cell. Moreover, the network can now integrate information coming from spatially disparate regions in complex ways that result in activity in other spatial regions. Second, the network has much more computational power than any of its individual cells (nodes), providing an IQ boost for the newly formed Self. In such networks, Hebbian dynamics on the electrical synapse (GJ) can provide association between action in one location and reward in another, which enables the system to support credit assignment at the level of the larger individual.

The third consequence of GJ connectivity is the partial dissolution of informational boundaries between the subunits. GJ-mediated signals are unique because they give each cell immediate access to the internal milieu of other cells. A conventional secreted biochemical signal arrives from the outside, and when it triggers cell receptors on the surface, the cell clearly knows that this information originated externally (and can be attended to, ignored, etc.)—it is easy to maintain boundary between Self and world. However, imagine a signal like a calcium spike originating in a cell due to some damage stimulus for example. When that calcium propagates onto the GJ-coupled neighbor, there are no metadata on that signal marking its origin; the recipient cell only knows that a calcium transient occurred, and cannot tell that this information does not belong to it. The downstream effects of this second messenger event are a kind of false memory for the recipient cell, but a true memory for the collective network of the stimulus that occurred in one part of the individual. This wiping of ownership information for GJ signals as they propagate through the network is critical to enabling a partial “mind meld” between the cells: keeping identity (in terms of distinct individual history of physiological states—memory) becomes very difficult, as small informational molecules propagate and mix within the network. Thus, this property of GJ coupling promotes the creation of a larger Self by partially erasing the mnemic boundaries between the parts which might impair their ability to work toward a common goal.

This is a key part of the scaling of the Self by enlarging toward common goals—not by micromanagement, but by bringing multiple subunits into the same goal-directed loop by tightly coupling the sensing, memory, and action steps in a syncytium where all activity is bound toward a system-level teleonomic process. When individual identities are blurred in favor of longer time-scale, larger computations in tissues, small-horizon (myopic) action in individual cells (e.g., cancer cells’ temporary gains followed by maladaptive death of the host) leads to a more adaptive longer-term future as a healthy organism. In effect, this builds up long-term collective rationality from the action of short-sighted irrational agents (Sasaki and Biro, 2017; Berdahl et al., 2018).

It is important to note that part of this story has already been empirically tested in assays that reveal the shrinking as well as the expansion of the Self boundary (Figure 9). One implication of these hypotheses is that the binding process can break down. Indeed this occurs in cancer, where oncogene expression and carcinogen exposure leads to a closure of GJs (Vine and Bertram, 2002; Leithe et al., 2006). The consequence of this is transformation to cancer, where cells revert to their ancient unicellular selves (Levin, 2021b)—shrinking their computational boundaries and treating the rest of the body as external environment. The cells migrate at will and proliferate as much as they can, fulfilling their cell-level goals—metastasis [but also sometimes attempting to, poorly, reboot their multicellularity and make tumors (Egeblad et al., 2010)]. The model implies that this phenotype can be reverted by artificially managing the bioelectric connections between a cell and its neighbors. Indeed, recent data show that managing this connectivity can override default genetically-determined states, inducing metastatic melanoma in a perfectly wild-type background (Blackiston et al., 2011) or suppressing tumorigenesis induced by strong oncogenes like p53 or KRAS mutations (Chernet and Levin, 2013a,b). The focus on physiological connectivity (information dynamics)—the software—is consistent with the observed facts that genetic alterations (hardware) are not necessary to either induce or revert cancer [reviewed in Chernet and Levin (2013a)].

All these dynamics lead to a few interesting consequences. GJ-mediated communications are not merely conversations (in the way that external signaling is)—they are binding, in the sense that once a GJ is open, a cell is subject to whatever comes in from the neighbor. In the same sense, having a synapse makes one vulnerable to the state of neighbors. GJs spread (dilute) the pain of depolarization, but at the same time give a cell’s neighbors the power to change its state. Compatible with the proposal that the magnitude of a Self is the scale and complexity of states by which it can be stressed, connections by tunable, dynamic GJs greatly expand the spatial, temporal, and complexity of things that can irritate cells; complex events from far away can now percolate into a cell via non-linear GJ paths through the network, and enabling the drive to minimize such events now necessarily involves homeostatic activity of goal states, sensing, and activity on a much larger scale. Stress signals, propagating through such networks, incentivize other regions of the tissue to act cooperatively in response to distant events by harnessing their selfish drive to reduce their own stress. This facilitates the coherent, system-level response to perturbations beyond their local consequences, and gives rise to larger Selves able to react coherently to stressful departures from more complex, spatially-distributed allostatic setpoints. For example, whereas a solitary cell might be stressed (and react to) abnormal local pH, cells that are part of a transplanted salamander limb will be induced to a more grandiose activity: they will change the number of fingers they produce to be correct relative to the limb’s new position in the host’s body (Ruud, 1929), a decision that involves large-scale sensing, decision-making, and control.

A fourth consequence of the coupling is that cooperation in general is greatly enhanced. In the game theory sense, it is impossible to cheat against your neighbor if you are physiologically coupled. Any positive or negative effects of a cell’s actions toward the neighboring cell are immediately propagated back to it, in effect producing one individual in which the parts cannot “defect” against each other. This dynamic suggests an interesting twist on Prisoners’ Dilemma models in which the number of agents is not fixed, because they have the options of Cooperate, Defect, Merge, and Split (we are currently analyzing such models). Specifically, merging with another agent creates an important dimensionality reduction (because defection is no longer an option); this not only changes the calculus of game theory as applied to biological interactions, but also the action space itself. These dynamics take place on a developmental timescale, complementing the rich existing literature on game theory in evolution (Maynard Smith and Szathmáry, 1995; Maynard Smith, 1999; McEvoy, 2009; Pacheco et al., 2014).

Indeed, the smaller and larger agents’ traversal of their various spaces provides a way to think about how smaller agents’ (cell-level) simple homeostatic loops can scale up into large, organ-level anatomical homeostatic loops. Prentner recently showed how agents build up spatial models of their worlds by taking actions that nullify changes in their experience (Prentner, 2019). Working to nullify changes to one’s state that would otherwise be induced by the vagaries of external environment (and other agents) is the core of homeostasis—the action loops that seek to preserve important states against intervention and entropy. This is not only for physical movement (which results in a creature perceiving itself to be situated in spacetime) but also for other states in which actuation takes place via turning on/off specific genes, remodeling an anatomy, or opening/closing ion channels to change physiological state. An agent can notice patterns in what actions it had to take to keep in homeostasis despite various perturbations that occur, and based on that refine an internal model of some space within which it is acting. This is closely related to the surprise minimization framework (Friston, 2013; Friston et al., 2013; Friston K. et al., 2014), and suggests a straightforward sense in which larger Selves scale up to models of their world and themselves from evolutionary primitives such as metabolic homeostasis. Bioelectricity provides examples where cell-level physiological homeostats form networks that implement much larger-scale pattern memories as attractors, akin to Hopfield networks (Figure 10; Hopfield, 1982; Inoue, 2008; Pietak and Levin, 2017; Cervera et al., 2018, 2019a,b,2020a). This enables all tissues to participate in the kind of pattern completion seen in neural networks—a critical capability for regenerative and developmental repair (anatomical homeostasis).

With these pieces in place, it is now possible to mechanistically visualize one specific aspect of the progressive scaling that expands the cognitive light cone. Cells with a chemical receptor can engage in predictive coding to manage their sensory experience (Friston K. et al., 2014; Friston K. J. et al., 2014; Thornton, 2017). Similarly, individual cells homeostatically maintain Vmem (cell membrane resting potential voltage) levels. However, cells can electrically couple via gap junctions to create bioelectric networks that work as a kind of virtual governor—coupled oscillators possess emergent dynamics that now maintain large, spatial patterns of Vmem against perturbation with greater stability (Pietak and Levin, 2017, 2018; Cervera et al., 2018, 2019a,b,2020a,b; Pai et al., 2018; Manicka and Levin, 2019a). These spatial patterns serve as instructive pattern memories guiding the activity of a cell collective through anatomical morphospace toward the correct target morphology (Sullivan et al., 2016; Levin, 2021a; Pezzulo et al., 2021).

Voltage is especially interesting because each Vmem level in a single cell is a coarse-grained parameter, subsuming many distinct combinations of sodium, potassium, chloride, etc., levels, and many distinct open/closed states of particular ion channel proteins, which all result in the same resting potential. This can be seen as the minimal version of generalization—cells learning to respond to classes of events by transducing not specific ion levels or channel protein activity states but the macrovariable “voltage.” This in turn enables them to repurpose existing responses for novel combinations of stimuli (e.g., familiar depolarization events caused by novel ion dynamics).

Moreover, gap junctions propagate voltage states across tissue, allowing cells to respond to events that are not local in nature (larger-scale) and to respond en masse. More generally, this means that the input to any group of cells is produced by the output of groups of cells—sub-networks, which can be complex and highly processed over time (not instantaneous), enabling predictive coding to manage complex states (at a distance in space and time) and not only purely local, immediate sensory data. It also means that the system is extremely modular, readily connecting diverse upstream and downstream events to each other via the universal adapter of bioelectric states. When this is applied to the homeostatic TOTE (test-operate-exit) loop, allowing its measurement, comparison, and action modules to be independently scaled up (across space, time, and complexity metrics), this inherently expands the cognitive light cone of a homeostatic agent to enable progressively more grandiose goals.

Crucially, all of the above-mentioned aspects of the role of generic bioelectric networks underlying the scaling of Selves are not only the products of the evolutionary process, but have many functional implications for evolution itself (forming a positive feedback loop in which rising multiscale agency potentiates the evolution of increasingly more complex versions).

Developmental bioelectricity works alongside other modalities such as gene-regulatory networks, biomechanics, and biochemical systems. The TAME framework emphasizes that what makes it special is that it’s not just another micro-mechanism that developmental biologists need to track. First, developmental bioelectrics is a unique computational layer that provides a tractable entrypoint into the informational architecture and content of the collective intelligence of morphogenesis. Second, bioelectric circuits show examples of modularity, memory, spatial integration, and generalization (abstraction over ion channel microstates)—critical aspects of understanding basal origins of cognition. Developmental bioelectricity provides a bridge between the early problem-solving of body anatomy and the more recent complexity of behavioral sophistication via brains. This unification of two disciplines suggests a number of hypotheses about the evolutionary path that pivoted morphogenetic control mechanisms into the cognitive capacities of behavior, and thus sheds light on how Selves arise and expand.

Developmental bioelectrics is an ancient precursor to nervous systems. Analog bioelectrical dynamics generate patterns in homogenous cell sheets and coordinate information that regulates transcription and cell behaviors. Evolution first exploited this to enable cell groups to position the body configuration in developmental morphospace, long before some cells specialized to use very fast, digital spiking as neural networks for control of behavior as movement in 3-dimensional space (Fields et al., 2020). The function of nervous systems as spatial organizers operating on data from the external world (Keijzer et al., 2013) is an adaptation built upon the prior activity of bioelectric circuits in organizing the internal morphology by processing data from the internal milieu. While neural tissues electrically encode spatial information to guide movement (e.g., memory of a maze in a rat brain) by controlling muscles, bioelectric prepatterns guide the behaviors of other cell types, on slower timescales, during development, regeneration, and remodeling toward invariant, robust anatomical configurations.

Developmental bioelectricity illustrates clearly the continuous nature of properties thought to be important for cognition, and the lack of a clean line separating brainy creatures from others. On a single-cell level, even defining a “neuron” is not trivial, as most cells possess the bioelectrical machinery and a large percentage of neuronal genes are also expressed in non-neural cells (Bucher and Anderson, 2015), while neural molecular components are found in cytonemes (Huang et al., 2019). Many channel families were likely already present in the most recent unicellular ancestor (Liebeskind et al., 2015). The phylogeny of ion channels is ancient, and the appearance of context-sensitive channels (enabling new kinds of bioelectrical feedback loops) tracks well with the appearance of complex body plans at the emergence of metazoa (Moran et al., 2015), revealing the remarkable evolutionary continuum that leads from membrane excitability in single cells to cognitive functions in advanced organisms, by way of somatic pattern control (Cook et al., 2014).

Fascinating work on bacteria has shown that prokaryotes also utilize bioelectric state for proliferation control (Stratford et al., 2019); and, paralleling the developmental data discussed above, bioelectric phenomena in bacteria scale easily from single-cell properties (Kralj et al., 2011) to the emergence of proto-bodies as bacterial biofilms. Bacterial communities use brain-like bioelectric dynamics to organize tissue-level distribution of metabolites and second messenger molecules, illustrating many of the phenomena observed in complex morphogenetic contexts, such as encoding stable information in membrane potential patterns, bistability, and spatial integration (Humphries et al., 2017; Liu et al., 2017; Larkin et al., 2018; Martinez-Corral et al., 2018; Yang et al., 2020). Not only animal lineages, but plants (Baluska and Mancuso, 2012; Volkov et al., 2019; Serre et al., 2021) use bioelectricity, as evolution frequently exploits the fact that bioelectric dynamics are a powerful and convenient medium for the computations needed to solve problems in a variety of spaces not limited to movement in 3D space.