Forms of life, forms of mind

Here’s a chapter I wrote for the forthcoming book edited by Igor Adameyko and Gerhard Schlosser – “Open Questions in Developmental Biology” – as a statement for that community of my views on the meaning and future of developmental biology and related fields.

Biophysical Intelligence Between Genotype and Phenotype: the agential material of life from evo-devo to regenerative medicine

Summary:

            Morphogenesis — in which cell groups build or repair a complex organism — occupies a critical position between genotype and phenotype, while being central to fundamental questions in biology, medicine, and philosophy of mind. I argue that morphogenesis is not a mechanical process captured by emergence of complexity, but a fundamentally cognitive one: living matter constitutes an agential material whose multi-scale problem-solving capacities cannot be fully captured by conventional mechanistic or cellular automaton models. Using examples from regeneration, development, and cancer, I show how cell collectives navigate anatomical morphospace using goal-directed strategies amenable to the tools of neural and behavioral sciences. Developmental bioelectricity — networks of ion channels and gap junctions conserved between brain and body — forms a “cognitive glue” that scales individual cellular setpoints into organism-level setpoints in anatomical morphospace (species-specific target morphologies). Endogenous bioelectric patterns are highly tractable entrypoints for regenerative medicine and bioengineering. Moreover, evolution acting on competent, agential materials has distinct dynamics from evolution of passive substrates, with important consequences for evolvability. Finally, novel living constructs such as Xenobots and Anthrobots reveal morphogenetic plasticity toward latent forms outside of prior evolutionary history, underscoring the need for new frameworks to predict and engineer outcomes in natural, artificial, and hybrid embodiments.

1. Introduction

            Each of us has made the remarkable journey from a single cell to a cognitive being capable of understanding their own origin story. Of course, that same kind of journey also takes place on an evolutionary time scale, during the eons in which we moved from our microbial ancestors to complex multicellular beings working in laboratories to understand the mechanisms and the Umwelt of their ancestors. For the purposes of this chapter, I will use “morphogenesis” to refer to the entirety of this process – the ability of a single egg cell to eventually self-assemble into a complex form of anatomy and behavior.

Morphogenesis at the center of the big questions of life and mind

            In a sense, Developmental Biology is the Queen of all the Sciences. Within the morphogenetic process are hidden answers to all of life’s big questions. First, the increase of spatial complexity – self-organizing order on many scales, which connects to questions of emergence, entropy, and collective intelligence. Second, the path to definitive regenerative medicine: here in front of our eyes are processes which regenerate an entire body from one cell (development), repair against drastic injury (split embryos become normal twins), bind individual cell behaviors toward a larger agenda (cancer suppression and reversion), and produce organisms of age 0, not starting at the age of their parent cells or their entire lineage (anti-aging). Third, our path across many sciences: starting off as an oocyte – the province of chemistry and physics – we eventually became subjects of biology and physiology, and then of behavioral science and eventually of psychoanalysis (not to mention friendship, love, etc.). Thus, morphogenesis challenges us not only to understand the mechanisms which enable it, but also to define the lenses through which we see life – the various disciplines appropriate for the study of our different stages of transformation.

            Perhaps the most important lessons morphogenesis teaches are those of gradualism. Yes, we’ve given convenient names to individual stages of development, and it is clear that adults have different capabilities than embryos or egg cells. The standard story, with crisp boundaries between sciences, claims that the final product of morphogenesis is a being with an inner world – having a first person perspective – which the quiescent oocyte does not have (being a “mere blob of chemicals”). But, developmental biology also tells us there is no bright red line – no magic point in which we tick over from being chemico-physical machines to cognitive agents with minds; what it reveals is a gradual process of metamorphosis. Whatever hopes, dreams, preferences, beliefs, and consciousness we may have develop slowly and gradually, as does our body. The process can be slowed down, temporally zoomed in, as far as one wants, to see clearly that “and then the brain develops” and similar stages are extremely lengthy, involved continuous processes. The sharp categories, names, and disciplines we paint onto this process are a feature of our attempts to simplify and understand, not necessarily intrinsic features.

            This aspect is rarely appreciated outside of our field. For example, many books and papers in the philosophy of mind talk about “the human” with its mind, moral sense, responsibility, etc. They do not generally ask about ontogenetic and phylogenetic ancestors, to discuss when and how single cells became something that has these properties. A lack of appreciation of our origins (and of the plasticity of our embodiment) obscures the many unknowns that lie just beneath the convenience of expedient and discrete terms such as “organism”, “machine”, “brain”, “mind”, “intelligence”, etc., all of which give a false sense of understanding of the deepest unanswered questions at their edges. Thus, developmental biology and its sister disciplines bioengineering, synthetic biology, artificial life, etc. are essential backgrounds for progress in many aspects of law, philosophy, and other humanities. All of those communities assume that we have answers that underlie their frameworks; we need to either provide them or communicate the needed changes in those frameworks.

            It is crucial that morphogenesis is not just the study of the origin of the bodily anatomy, but also of the mind. Alan Turing understood this. His seminal paper [1] addressed the question of where patterns come from ab initio, from a chemical soup into structures of embryogenesis.  Why would someone who was so interested in intelligence in unfamiliar implementations, in reprogrammability (plasticity) [2], in the power (and limits) of algorithms, and in the foundations of mathematics, write one of the first papers quantitatively modeling the origin of order from homogeneous media? Because Turing realized the profound symmetries between the origins of the body and those of the mind – he saw that questions of intelligence, of patterns arising (and being detected) from chaos – go beyond specific embodiments, problem spaces, or origin stories. In our quest to understand what it means to be an embodied mind in our universe, we must be prepared to recognize it in unfamiliar guises. Transitional forms, such as those of embryogenesis, provide an ideal training ground in which to improve our conceptual frameworks and our technologies.

An idiosyncratic perspective on morphogenesis: a preview

            I argue that one thing that morphogenesis makes clear is that what we seek are models of transformation. We are not looking for sharp definitions of “life” and of various cognitive terms. The search for clear definitions has held back the field because of a focus on pseudoproblems of classification and because it prevents the transfer of tools across disciplines. Instead, what we must search for are principles of scaling – mechanisms and patterns by which capabilities increase and project into different problem spaces as oocytes become complex organisms [3]. What forces are driving beings across the spectrum from chemical machine to mature mind? What mechanisms scale the tiny setpoints of physiological loops in single cells into the grandiose goal states of cell groups as they implement complex organs or the planetary-scale goals of human minds?

            In this chapter, I discuss what I think are some of the most interesting aspects of morphogenesis, from an admittedly unconventional and interdisciplinary perspective that has guided my group’s work in this field for several decades. Specifically, I propose that morphogenesis, as the layer between the genotype and phenotype, is not merely a complex mapping but a fundamentally cognitive one. Living matter is an agential material, representing a next challenge for engineers past the passive matter of past millennia and the active and computational matter of the last few decades. I also review an emerging field of developmental bioelectricity, which shows that the cognitive glue [4] – the mechanisms that align cells toward large-scale anatomical goals – is highly conserved between the behavior of brains and of bodies. To understand this perspective, a few things must be kept in mind.

            First, what needs to be explained about morphogenesis? It is not merely the reliability (consistency) or the rise in complexity of the outcome of developmental processes. I focus here on examples typically omitted from modern developmental biology classes: phenomena that reveal the context-sensitive problem-solving aspects of the process (a.k.a., intelligence). Second, what is the role of the genome? I discuss recent advances that help understand its role, the software modes of the genetically-specified hardware of life, and what the genome does and doesn’t encode. Third, what are the implications of all this for evolution? I argue that evolving an agential material, vs. a passive one, has many interesting consequences (not the least of which is an agency ratchet that kicks in prior to self-replication and selection).

            What is at stake, in moving from a bottom up view of molecular biology as a privileged layer of explanation [5] to a multi-scale competency architecture model more consistent with recent advances in causal information theory [6-12]? It’s hard to overemphasize the importance of getting this right – moving beyond constraining mechanist and organicist models to understand the deep symmetry at the heart of life and mind. The first impacts will concern the biomedicine of birth defects, traumatic injury, cancer, and aging [13]. Additional impacts will affect the basic understanding of evolution, enabling tools for bioengineering and synthetic morphology (the next generation past synthetic biology) [14], transhumanist efforts to modify our form and cognition beyond what random mutation and selection have given us so far, and the design of AI and robotics built on the deep lessons from the agential material of life.

2. The unreasonable effectiveness of the behavioral sciences in developmental biology

            A famous essay [15] by Eugene Wigner asks why mathematics is so surprisingly effective in explaining and controlling the physical world. But any estimate of whether a formalism is unreasonably (surprisingly) effective requires that we have good priors with respect to which we can form expectations. Progress is inhibited when assumptions about what tools are appropriate for phenomena like morphogenesis are treated as axioms. All such assumptions, such as the ubiquitous but often unspoken claim that formalisms suitable for low-level machines (“mechanisms”) are more facilitative of progress than those suitable for richer agents, are currently being tested.

What formalism is appropriate?

            An important tool for understanding processes like morphogenesis is computational modeling. Computer models require making assumptions explicit, and effectively highlight gaps in knowledge (as exist for example when trying to flesh out the “arrow models”, popular in developmental biology papers, into something that will actually run). Computational models also force a clarity of perspective of how one thinks of the system being modeled: creating an abstraction of cells, signals, or shapes, one has to specify their properties and capabilities.

          Cellular automata (CA) models [16-19] have been popular in developmental biology because they seem to capture the fundamental dynamic: a set of local rules by which cells interact. This seems reasonable at first, because development is so reliable – builds the same thing (nearly) every time, and because CA’s don’t require a central controller – the cells work in parallel. Also, they embody a fundamental assumption of the modern age: that there is no goal, no sense in which the future outcome is represented, just local actions. This is considered a beneficial feature because of Morgan’s Canon [20] – the idea that all systems should be explained without recourse to cognitive concepts where-ever possible. It also matches the pervasive intuition that chemical processes in morphogenetic systems simply can’t know anything – they have to be automata.

            This view has three major problems. First, it is so widely prevalent that workers in the field (especially students raised on this idea) forget that it is only an assumption – not a result, and that estimates of level of intelligence to any system is an empirical question, not a linguistic or philosophical one, which must be settled by experiment. Second, it neglects neuroscience and cybernetics – two modern sciences which have made great progress on natural and artificial systems which literally encode representations of goal states and exert effort to meet those goals (with variable degrees of ingenuity). We now know that it doesn’t take magic to implement goal-seeking systems, and thus such models in developmental biology are not an appeal to mysterianism but rather a natural attempt at extension of rigorous tools used with great profit elsewhere. Third, it appeals to “emergence” to explain the actual complex forms that result. What does “emergence” mean? One thing it signifies is surprise: if the appearance of outcomes from simple rules were not foreseen by a typical scientist presented with the rules, it’s emergent. If a typical modern human can look at the rules and immediately see what will happen, it’s not emergent. The notion of emergence as observer-relative surprise is not particularly useful. What it obscures is the serious problem faced by workers who are certain that chemical material can’t know anything, but (usually) believe that they, and their chemical brain, really do know things. The highly rewarded ability to keep those mentally separate, and to believe that developmental biology doesn’t need to concern itself with cognition, is a major barrier to progress.

            This framework is very limiting for biomedicine and bioengineering, because morphogenesis, as viewed by CA models, is not reversible – it only goes in one direction. Given the rules, one simply runs them forward to see what emerges. But what if, as in engineering, synthetic morphology, or regenerative medicine, one wants to go the other way: given what we want to happen (system-level, large-scale outcomes), what should the rules and starting conditions be? That is, in general, an intractable problem [21]. CA models simply offer no tractable way to design outcomes because the gulf between the parameter knobs (rules) and outcomes is too large and convoluted. There are of course systems (such as the Game of Life classic automaton) where that is the best we can do. But biology offers alternatives. For example, in homeostatic systems (which are ubiquitous in living material), one can shift the encoded setpoint, and the system will do the rest. More advanced versions, such as those in brainy species, have goal states that can be re-set by training or communication, and offer many highly useful competencies for achieving those goals without having to be micromanaged to do so. The question is, what does it take to be a goal-directed system, and do morphogenetic systems have it?

Figure 1: Agentic models are not mysterian. A common assumption holds that only 1st order (feed-forward) models, in which outcomes are “emergent” from the action of subunits where nothing knows the destination of the process, can be rigorous. In contrast, Cybernetics, control theory, and other fields have long produced tools to understand and manage systems with feedback, including a range of progressively-more sophisticated architectures of machines that have goals without magic, ranging from passive matter to human-level metacognition (A) [25]. A developing system moving from oocyte to human being with a brain traverses the types of dynamical control shown in panel A. The “spectrum of persuadability (B) illustrates four waypoints on this continuum, and is organized from a pragmatic, empirically-focused perspective [3]. On this view, all systems (including cellular collectives) must be experimentally probed, not philosophically decreed, to determine what kind of interaction is optimal: hardware rewiring (e.g., genetic engineering and molecular biology), setpoint resetting (editing biophysically encoded homeostatic setpoints), training (positive and negative reinforcement stimuli), or other forms of high-level communication. Panel B by Jeremy Guay of Peregrine Creative.

            I have proposed the idea of a “Spectrum of Persuadability” [3], to adopt a practical, engineering perspective focused on interventions and interactions with a system. It is critical to note that where any given system fits on the Spectrum of Persuadability (Figure 1) is not knowable from a philosophical armchair – we cannot simply skew low (as reductionism and Morgan’s Cannon would have us do), nor can we always skew high (as animist and mysterian approaches often do). We start with plausible estimates and then we do experiments to see how that works out, being ready to revise our estimate in either direction. This works because cognitive claims are interaction protocol claims. When one says that a system, such as an embryo, has a certain level of intelligence, they are making a claim as to what kind of operational toolkits – physical rewiring, cybernetics/control theory, behavior science, psychoanalysis, etc. – will offer the best prediction and control (or more broadly, the most beneficial interaction). It is an iterative, empirical process because having made the claim, everyone can see how it worked out. One can’t simply know the intelligence level of any system by observations of its behavior alone. What one must do, especially to gauge the level of sophistication in goal-directed activity, is perturbative experiments: for example, place a barrier between the system and its hypothesized goals (in whatever space on hypothesizes it to be navigating), and see how much it can implement James’ definition of intelligence: “same goal by variable means” [22]. Feed-forward, cellular automaton models do not capture these kinds of dynamics. Fortunately, if we accept that favoring open-loop (feed-forward, goal-less, know-nothing) mechanisms is only an assumption that we are not bound to, with a whole spectrum of rigorous tools available (Fig. 1A), a number of other options suggest themselves as interesting research agendas that have already begun paying off.

On the application of mentalistic terminology in developmental biology

            The presence of homeostatic loops (basal version of goal-seeking mechanisms) in biology is well-known. It’s seen in physiological regulation [23, 24], problem-solving behavior, and the ability of cancer to evade various therapeutics over time. But homeostasis is the first rung on the ladder of intelligence, implementing a basic ability to store a setpoint (memory), determine current state in order (sensing), and implement that setpoint by exerting effort and minimizing distance to it (navigation of a problem space, goal-seeking behavior) [25]. Subsequent elaborations of this basic scheme include homeodynamics, homeorhesis [26-29], allostasis [30, 31] and more sophisticated mechanisms recognized in cybernetics and behavioral science, which correspond to increasing degrees of competency and ingenuity in which living materials have to reach their goals despite unforeseen challenges and perturbations.  This enables an empirical research program which asks, what set of tools along the spectrum of intelligence is appropriate for any given biological system? Crucially, this is not a philosophical or linguistic project seeking to re-define intelligence as mere complexity. It is an experimental research program which seeks to rigorously determine what problem-solving capacities can be found, and exploited (by bioengineers, other biological systems, and evolution itself), in living materials and beyond. To see why, it’s important to dig into how agentic terminology is normally used.

            The use of “intelligence” and other cognitive terms applied outside of its familiar context of brainy animals immediately raises questions: might not these terms be misused? Are not morphogenetic systems simply following the rules of chemistry – why anthropomorphize them? In the modern age, we must accept that all cognitive systems – we included – exhibit chemistry, not magic, when one drills down to examine the lower levels. Thus, there simply is no special human category of which one can correctly anthropomorphize as somehow being beyond the laws of physics at its base. That word is an anachronism and needs to be retired in favor of an empirically-grounded view, updated with the latest findings in causal information theory [6, 9, 10, 32-35], in which it is perfectly possible for a system to be subject to chemistry and also have additional levels that have causal roles. Indeed, the standard view of modern developmental biology is not really reductionist (in which case it would seek explanations of events in terms of quantum foam), but rather chemo-centric, having picked the level of “chemistry” as an arbitrary gold standard for explanations in the field. Denis Noble [5, 36, 37] and others have sufficiently pointed out that there is no privileged level of causation in biology.

            Fortunately, we have many tools available to us besides the formalisms of chemistry, including those of cybernetics and behavioral science. I focus on two main points in clarifying the use of cognitive terminology (developed in detail in [38]).  First, that almost any paradigm can be rescued by enough epicycles; indeed, after one has produced a new effect or reached a new capability, it is easy to say “that’s just chemistry doing what chemistry does”. The criterion should not be whether some dynamic can be described at the lowest level (chemistry? Quantum mechanics?). Instead, the emphasis should be on novel capabilities, and new research programs facilitated (or suppressed) by a given way to understand a system. I propose that attempts to mine the rich toolbox of behavioral science to exploit capabilities of morphogenetic systems will continue to pay off in many (but likely not all) cases. The existing gains facilitated by this view, and the promised for regenerative medicine, are detailed elsewhere [39, 40].

            A sometimes uncomfortable aspect of my position is that the empirical utility of framings need to be applied fearlessly, and followed where-ever it may lead: its empirical consequences must be taken seriously even when they contradict long-cherished a priori commitments to how non-intelligent a given system “must be”. In other words, if a specific framing, which uses tools normally reserved for brains, results in fruitful new research programs on bacterial biofilms [41-43], plant roots [44-50], the training of gene-regulatory networks [51-54], or developmental/regenerative biology [55], then the scientific approach requires that we consider those systems to be bona fide subjects of that corner of the natural world that is supposed to be described by the behavioral science of a spectrum of minds.

            To cut through philosophical impasses and hew close to testable hypotheses, empirical progress, and research agendas at the bench, I take an operationalist stance with respect to questions like “but is it really cognitive or only metaphorically so?”. In keeping with a fairly standard metaphysics of science, I eschew any use of “really”. All we have in science is metaphors, and an unflinching commitment to testing them and seeing which ones offer the best purchase on the world. Thus, consistent with an engineering approach (formalized in the TAME  framework [3]), I define the perspective of this chapter as: what tools offer the most insight, in terms of not only predictive capacity but in terms of generatively driving new research questions and new capabilities? Specifically, might some of those tools be found within the cognitive and behavioral sciences? It should not be a surprise, because of course the endpoint of development is the construction of a system to which those tools very effectively apply. Do they only kick in after the brain is mature, or might basal versions be found earlier on in the process? We next look at a few specific examples which motivate the use of such models in morphogenetic contexts.

Specific examples: progress made by porting tools from cognitive science

            The perspective I propose is of the following symmetry: much like animal behavior in 3D space is the result of a collective intelligence composed of neurons aligned in networks, navigation of the space of possible anatomies (anatomical morphospace) is also a kind of behavior of the cellular collective. What kind of navigational competencies does morphogenesis offer?

Figure 1: Agentic models are not mysterian. A common assumption holds that only 1st order (feed-forward) models, in which outcomes are “emergent” from the action of subunits where nothing knows the destination of the process, can be rigorous. In contrast, Cybernetics, control theory, and other fields have long produced tools to understand and manage systems with feedback, including a range of progressively-more sophisticated architectures of machines that have goals without magic, ranging from passive matter to human-level metacognition (A) [25]. A developing system moving from oocyte to human being with a brain traverses the types of dynamical control shown in panel A. The “spectrum of persuadability (B) illustrates four waypoints on this continuum, and is organized from a pragmatic, empirically-focused perspective [3]. On this view, all systems (including cellular collectives) must be experimentally probed, not philosophically decreed, to determine what kind of interaction is optimal: hardware rewiring (e.g., genetic engineering and molecular biology), setpoint resetting (editing biophysically encoded homeostatic setpoints), training (positive and negative reinforcement stimuli), or other forms of high-level communication. Panel B by Jeremy Guay of Peregrine Creative.

            The first is the ability to reach the same goal from different starting positions. A salamander’s limb amputated at any position along the proximo-distal axis will regenerate the correct amount of tissue, in the correct structure, and then stop (Figure 2A). This is the most remarkable aspect of regeneration – it stops when a specific outcome is achieved (the “target morphology”). The system reliably navigates from different starting locations in morphospace and continues moving until the delta between target morphology and current position is within acceptable levels of error. Many species’ embryos, when cut into pieces, give rise to monozygotic twins, triplets, etc., not half-bodies. Conversely, multiple separate embryos coalesced into one also form normal embryos (even if carcinoma cells are mixed in) [56, 57]. Scrambled tadpole faces give rise to normal frog faces [58, 59], because craniofacial organs don’t simply move in pre-determined paths during metamorphosis, but rather move as needed (thus accommodating unexpectedly incorrect starting positions) to achieve a normal frog face (Figure 2B). Given that development is basically regeneration of an entire body from one cell, they both share the ability to attempt to find and maintain a specific region of anatomic morphospace. In effect, morphogenesis of development, regeneration, wound healing, and resistance to aging and cancer are all a kind of error minimization driven by feedback loops.

            A closely related concept is that of memory: systems that react to the same inputs differently, after they’ve experienced them. Homeostatic processes have an explicit setpoint to which they minimize distance (or responsively, or even proactively, as in the concept of allostasis [30, 31, 60-62]). Storing that setpoint is a kind of memory; interestingly, that memory is often not hardwired – there is a default, but it’s re-writable.  One example is trophic memory in deer [63, 64]: a wound made at a specific location in a deer antler rack will result in an ectopic branch in subsequent years as the antler rack is shed and regenerated (Figure 2C). Imagine trying to write a typical arrow diagram for a molecular biology model as is standard in developmental biology papers and textbooks: the structure must ascertain the position of the damage, store this information somewhere else in the body for months (while the whole antler rack falls off), and act on it via specific changes to bone, nerve, vasculature, and skin growth at a later timepoint. This is a kind of physiological memory of damage and a response of tissue growth whose future is different than its past (or than its default) because of experiences it has had. Other examples of re-writable pattern memory will be discussed below in the context of bioelectricity in planaria.

            Crucially, living matter implements a multiscale competency architecture, with each level of organization having its own ability to meet goals and implement target states. Life operates in many spaces beyond 3D space of conventional behavior [65]. Anatomical morphospace is one, but others include metabolic space, physiological state space, transcriptional space, etc.; it is plausible that evolution has re-used some of the same mechanisms and algorithms across spaces as it expanded and complexity increased (Figure 3).

Figure 3: Living systems exhibit behavior in many spaces besides conventional 3D space.  Evolution produces systems that adaptively navigate the space of gene expression, physiological states, and anatomical possibilities (A).  Evolution re-used some of the same mechanisms and algorithms, pivoting living systems across metabolic, physiological, transcriptional, morphological, behavioral, and linguistic systems over time (B). Intelligence and evolution form an on-going positive feedback spiral, as autonomous competencies of the living material potentiate evolvability and force evolution of problem-solving agents instead of fixed solutions, which in turn produces systems with higher cognitive capacity (C).  Panels in A used with permission from [249], [250], and [251] respectively. Panel in B by Jeremy Guay of Peregrine Creative.

           Morphogenesis is the result of cooperative (and competitive, see [66]) actions of individual cells. But the collective’s ability to navigate anatomical morphospace is the result of coordinated behaviors of individual cells which are themselves made of competent parts. Gene regulatory networks and chemical pathways can, on their own, exhibit 6 different kinds of learning including habituation, sensitization, and even Pavlovian conditioning; they can even count to small numbers [51, 52]. These were discovered when standard assays of behavioral science were applied to the functionality of GRNs; prior models of GRNs as dumb dynamical systems implied that rewiring (e.g., gene therapy) was the only way to change their behavior. Taking seriously the fact that a need for hardware rewiring was only an assumption, testing behaviorist protocols revealed that even the minimal mathematics of GRNs and pathways (coupled ordinary differential equations) were sufficient to result in a learning system whose future behavior could be altered by prior experience without any changes in the hardware or its parameters (thus opening the door to a myriad applications in drug conditioning and cell training).

            The scale-free competency also applies above the level of individual embryos and the cell-tissue-organ-system levels operating within them. It was recently found that not only do waves of ion flux propagate across groups of embryos, but their communication actually enables large collectives of embryos to resist teratogens better than small collectives or singletons [67]. This CEMA (Cross-Embryo Morphogenetic Assistance) effect shows that groups of embryos solve physiological problems better than singletons. Indeed, such hyperembryos have their own transcriptomes and express different genes than do single embryos on their own.

Flow of control in systems made of competent subsystems

            What is happening in living materials is that each level of organization operates as a virtual governor, which deforms the option space for its parts [68, 69]. The components don’t know that (or why) their problem-solving actions are actually contributing to navigation of a space of which they are unaware. What is “an embryo”, really? What is there one of, and what are we counting, when we look at 10⁶ cells in a blastoderm and call it “one embryo” (Figure 4)? What we are observing is alignment: both physical [70] and teleonomic [71] alignment of cells toward one specific target in anatomical space. Wholes must constantly enforce the alignment of their parts toward specific goals; defects in this process manifest as aging, degeneration, and cancer [72-75].

Figure 4: Collective systems are multiscale agents with causal emergence. An “embryo” is detected when a group of individual cells share a joint homeostatic goal in anatomical space, cooperating to reach a specific outcome (A). Embryonic blastoderms can be cut into pieces (after [252]) where each island will make its own embryo, showing how an excitable medium (the cell field) can give rise to some number (not fixed genetically) of individuals based on the dynamic ability of cells to join into collectives with plastic boundaries (B). Crucially, new tools in information theory show how to quantitatively determine when a whole is more than the sum of its parts – the “embryo” is a macroscopic whole that can exert effects on its cells and chemical pathways and causality does not always flow up from microscopic molecular events. In panel C (taken with permission from [10]), is shown a simple example: a Markov chain with four microscale states that can be encoded into a macroscale chain with two states; the causal structure transforms interventions (functional signaling) into effects. A macro causal model is a form of encoding for the interventions (inputs) and effects (outputs) that can use a greater amount of the capacity of the communication channel, showing how top-down causation is not a philosophical whimsy but a measurable feature of certain systems (biological ones being superb examples [6]). Panel D shows how a network (electrical, or chemical) can coarse-grain information to encode higher-level states: patterned on machine learning architectures [143, 253, 254] and retinal processing [255-258], such as is used in complex image recognition, intermediate layers can encode progressively more generalized information, such as organ-level shape decisions that are transduced by downstream pathways into biochemical states below the single cell level (panel taken with permission from [69]).

            Now, the question of whether these higher levels are real (in the sense of having their own causal power, as in downward causation [76-81]), or are just convenient ways of speaking, used to be a philosophical debate between reductionists and those convinced of a more holistic picture of the world. It is important for developmental biologists to be aware of recent developments in causal information theory, which, remarkably, have moved this debate from the field of philosophy to that of tractable mathematics: there are now tools [6, 52, 82-87] that can be applied to data from living systems to rigorously answer the question of whether higher levels of organization have their own causal power over and above whatever their mechanistic parts are doing. The definitive answer is that sometimes they can (some systems have maximum causation at levels above their smallest components), and biological systems are being quantified with respect to this, which means that for the first time we may soon have good quantitative definitions of what an “organ” and an “embryo” is.

            One of the most fascinating aspects of this multiscale architecture is how control is propagated across scales. Consider the example of what happens when a tail is grafted surgically to the middle flank of an amphibian body: it slowly remodels into a limb [88-90]. What this illustrates is, first, that the collective has a body-scale target morphology toward which it will try to reduce error in contexts beyond repair of injury or embryogenesis (i.e., context-sensitive remodeling to a specific setpoint) [91]. Second, consider the perspective of the tail tip cells: they are sitting in their correct environment – no injury, no damage. Why are they turning into fingers? They don’t know what a finger is, how many there should be, or why the whole thing is remodeling. What happens is that the morphogenetic goals at the organism level are being transduced down into the cell- and molecular-level signals needed to get the tissue to remodel into a new structure (see Figure 4D, [69]). The effect is radically non-local, both in lateral space and in vertical levels of organization.

            This ability of lower-level mechanisms to respond to higher-levels’ goals is a critical part of cognition and highlights one of the most important symmetries between the embodiment of minds and morphogenetic competencies. This is seen in the every-day magic of the mind-body interface during voluntary motion. Upon awakening, a human mind may have very abstract goals – financial, social, scientific, spiritual, etc.  In order for them to execute on those goals via behaviors, these abstract goals must eventually make calcium and potassium ions cross muscle cell membranes in ways they otherwise would not do. In other words, our bodies are a remarkable set of transducers which literally make chemistry obey abstract mental structures. This is implemented by bioelectricity (see section 4 below), but could in theory be driven by many mechanisms. What is important is that just as in cognition, morphogenesis reveals how the behavior at the molecular level is driven by the goal states of higher levels. One simply cannot stay at a single level (e.g., the beloved molecular biology) and have full insight into what is happening or how to control it. Doing so is like giving an accounting of air molecules’ elastic interactions when documenting the results of two mathematicians working on a wonderful new proof: it misses the whole point of what was happening and closes off the steps of discovery of the next big advance.

3. Between genotype and phenotype: morphogenesis as target and driver of evolution

            Mutations affect the genotype; but selection happens on the phenotype. Evolution cannot be understood without knowing how they are linked. This seems obvious on the one hand, and yet, claims about genetic underpinning of anatomical form and function are often made without a serious consideration of their relationship.  A fundamental understanding of evolution, and the role of morphogenesis as its target and its driver, requires a focus on what lies between genotype and phenotype.

Morphogenesis as a problem-solving process that interprets the genome

            The old view of genes encoding specific characters has been replaced by a much more nuanced picture, such as the omnigenic model [92], focusing on emergence and complexity science. And yet, one fundamental thing has remained: the mapping between genes and outcomes may have redundancy, pleiotropy, degeneracy, and other ways to implement a hairball of causality, but it’s always seen as passive. In other words, there are mechanisms linking them, but the unstated, near-universal assumption is that these mechanisms are low on the spectrum of intelligence. Genetic material is thus an input into chemical mechanisms, and even when DNA is described as information (fodder for a cybernetic, or even cognitive, system!), the interpreters of that information are thought to be polymerases and ribosomes, a very low level of description. I argue that it is only a hypothesis that large-scale morphogenesis can be explained as a purely feed-forward emergent consequence of mechanical events. Morphogenesis is actually a problem-solving process, using many strategies well-recognized by behavioral scientists. A potential consideration of more sophisticated dynamics by the morphogenetic machinery opens new ways of thinking about the kind of plasticity that has massive implications for evolvability and for the course of evolution.

            The first thing about morphogenesis is that it offers fascinating reprogrammability. The genetically-specified hardware does the expected work reliably under default outcomes (dogs have puppies and cats have kittens, making it seem like we understand heredity). However, there are a variety of ways in which the default setpoints can be re-written without manipulating the DNA. Much has been said elsewhere about epigenetic inheritance, and this now often refers to chromatin modification mechanisms but there are other important memory media in cells [93], including cytoskeletal and membrane components that, like DNA, offer an unbroken link going all the way back to the last universcal common ancestor.

            What is critical here is the plasticity, or reprogrammability, of living matter and the nano-scale hardware components provided to cells by the DNA. One example is that of cortical inheritance in ciliates [94]. These unicellular organisms have motile hairs (cilia) on their cell surface which can funnel food particles into their mouth opening. If a small square of surface was excised, and rotated 180 degrees, and grafted back in place, all of that cell’s offspring would have such a rotated square in their surface: a permanent line of organisms with a novel morphogenetic feature that is totally invisible to DNA profiling, because that is not where the information is. Moreover, these creatures live on the verge of starvation (because some of the surface is flowing food out of their mouth!) and their perfectly normal genome cannot help them.

Figure 5: Resetting non-local pattern memory in vivo.  Planarian flatworms reliably regenerate their body upon amputation. Typical models explain the ability of each fragment to determine which end should form a head and which end should form a tail using a gradient of some morphogen mapped onto a middle fragment resulting from two cuts (A), which is high at one end and lower at the other (red curve). However, the more interesting comparison is what happens at one cut. Consider the fragments made at level indicated with P1. The cells on either side of that cut have the same positional information because they are adjacent neighbors – there is no significant difference in morphogen level; and yet, they have radically different anatomical fates: one side makes a head and the other makes a tail. This shows that local positional information is not sufficient to set (or explain) anatomical fate. Indeed, the same events happen at position P2, and there is no absolute rule about morphogen level that can correctly explain the fate of each side of the 2 cuts. Instead, it is the relative positional information of the cuts which can tell each one which way it’s facing and what it should do. Reasoning that this is mediated by long-range physiological signaling through gap junctions [95], we inhibited this communication (B), finding not only that 2-headed animals result, but that this change to the target morphology information is permanent, and persists indefinitely through future rounds of amputation (red arrow indicates that a middle fragment can form a 2-headed worm if bioelectric signaling is perturbed; green arrow indicates that a 2-headed worm can be induced back to the 1-head state by inhibition of a proton pump [259]).

           Another example (Figure 5) is found in planarian flatworms [95] (discussed more below in the context of bioelectricity).  Planaria, when cut into pieces, reliably regenerate worms that have exactly one head and one tail. This is a phenomenon that cannot be explained by typical morphogen gradient models because single cuts result in opposite anatomical fates from cells which were adjacent neighbors prior to the cut (having the same positional information). Reasoning that the head vs. tail decision required long-range coordination of wound cells with the rest of the body to know the position and context of the fragment, we inhibited physiological signaling through gap junctions. The result was 2-headed worms. Two-headed planaria were described well over 100 years ago; but no one, to our knowledge, re-cut such animals to see what would happen, until we did it in 2008, presumably because it was obvious: having normal genetics, removal of the ectopic 2^nd head should cause repair to the default target morphology (1 head, 1 tail). We did re-cut them, because we hypothesized that the tissue had a physiological memory that could over-ride the default, and found a remarkable multicellular version of the ciliate example above: 2-headed animals, when cut, continue to produce 2-headed offspring in perpetuity, with no genetic changes being necessary to create a line of animals with a totally different body-plan. Note that what is happening in this example is not just modifying emergent outcomes, but actually modifying the target toward which the cellular processes continue to work in cases of deviation by amputation injury.

            These examples, along with the deer antler trophic memory described above, and others, reveal that the genome facilitates a very powerful kind of substrate: one whose capabilities are not directly determined by the hardware specification, but by the physiological software that it implements.  But the wonder of morphogenesis does not stop at being amenable to changes via external stimuli or perturbations: its most amazing property is intrinsic context-sensitive responses aimed at specific outcomes (problem-solving capacity able to handle unexpected internal and external circumstances).

            One example is illustrated by planarian responses to barium, a non-specific potassium channel blocker. Exposure to barium causes planarian heads to rapidly degenerate, as cells (especially neurons) degrade when unable to participate in potassium flux. Remarkably, when kept in barium, planaria eventually regenerate new heads which are barium-insensitive [96]. Transcriptomic analysis showed that they rapidly identify just a handful of differentially-expressed genes that dealt with their novel stressor – one that they (and likely, their lineage) had never seen before. How did the planarian cells, which don’t turn over rapidly enough for clonal selection, and don’t have time for a generate-and-test search, identify a small number of transcriptional actions in a ~20,000-dimensional space of possibilities that solved the new problem facing them? We don’t know, but the answer to how cells find appropriate responses to novel stressors would do much to advance medical practice. This example takes place in transcriptional and physiological spaces.

Figure 6: Adaptive response to novel scenario: plasticity of form and function.  (A) Newt kidney tubules usually consist of ~8 cells working together. When cells are made much larger, the same size tubule is constructed from fewer cells. When the cells are made enormous, one single cell wraps around itself to make the same large-scale structure, illustrating how in the service of the same final anatomical outcome, different molecular mechanisms (cytoskeletal bending vs. cell-cell communication) can be called up to achieve the goal. (B) Tadpoles with eyes induced on their tail (red arrowhead), but missing primary eyes in the head, have ectopic nerves that do not connect to the brain and stop at the spinal cord (white arrow indicates the ectopic eye; red arrows indicate optic nerve emerging from the ectopic eye); despite this, automated devices that train and test them on visual ques (C) show that they can learn behaviors requiring light sensing and context-sensitive responses (C’) – they can see, despite a completely novel sensory-motor architecture. Panel in A by Jeremy Guay of Peregrine Creative after [97, 98]; panels B-C’ taken with permission from [101, 260, 261].

            The next example of the use of available affordances to solve a new problem (a standard component of IQ tests) is chosen from morphogenetic space. In newts, kidney tubules are created by numerous cells working together (Figure 6A). Polyploid newts can be created, which have extra copies of the genetic material, which illustrate the remarkable multiscale intelligence of morphogenesis [97, 98]. First, the animals are normal and viable despite massive polyploidy. Second, the cells adjust their size accordingly. Third, the animals remain normal size because fewer cells participate in tubule morphogenesis. Finally, and most remarkably, if the cells are made so large that more than one cannot fit in a tubule diameter, then just one cell will bend around itself, leaving a hole in the middle, to produce the same structure. This is, first of all, an example of top down causation as discussed above: cells will adjust as needed in order to achieve a specific large-scale target morphology. Moreover, it’s an example of “same goal by different means” (i.e., problem-solving) – using different molecular mechanisms (cytoskeletal bending instead of cell:cell communication) to achieve the same outcome. Finally, it’s an example of creative problem-solving – using available tools to solve a problem in a new way. Imagine being a newt coming into the world: not only do you face massive uncertainty about the external environment, you can’t even count on your own parts. You don’t know how many copies of your genome you will have, how big your cells will be, how many cells you will have [99, 100], etc. You must  reach the target morphology regardless of those circumstances.

            The final example concerns behavioral endpoints of morphogenesis (Figure 6B-C’). When eye primordia are transplanted to the tail of a Xenopus embryo, normal eyes result (i.e., the eye cells are able to build an eye in abnormal locations). These eyes, as seen in visual behavioral learning assays, are functional – the animals can see [101], despite the fact that the optic nerve connects to the spinal cord, gut, or nowhere – it never connects to the normal location, the brain. The most remarkable aspect is that this radically different sensory-motor architecture does not require new rounds of mutation, selection, and adaptation – it works “right out of the box” in its new configuration. Why?

Beginner’s Mind: problem-solving, not fixed solutions

            The tail-eye tadpole works because the standard Xenopus embryo could never rely on the eye to be in the correct position in the first place. Living matter, unlike most of our computer technology, is a highly unreliable medium: you can never be sure of how many copies of any molecule you have, how long things will last, or what will happen to you. The one thing of which you can definitely be certain is that you will be altered or mutated (depending on the time scale) in ways you can’t predict. The only way to overcome the paradox of change [102, 103] is by committing to plasticity. Evolution mostly doesn’t make solutions to specific problems, it makes problem-solving systems, which under default conditions do the same thing (thus lulling observers like us into models of hardwired characteristics “encoded” by DNA), but due to the intelligence baked in at multiple levels of organization, can also do many other things in a context-sensitive manner. This is most obvious, to us, in the case of brainy animals navigating 3D space via motion, but is seen throughout anatomical, physiological, and transcriptional spaces when we know how to look. On this view [104], the genome is a boundary condition on the cellular hardware, and also a prompt to the intelligence of cell collectives which must interpret and use it as a resource to address their stressors and setpoints (Figure 7). Moreover, the bottleneck or bow-tie node of development [105-108], which squeezes down to an egg between generations, forces creative interpretation of the memories of past instances – an architecture which is essential to cognitive function, as follows.

Figure 7: The Paradox of change requires creative interpretation of information.  At any moment in time, a cognitive being does not have access to the past – what they have access to as memories are biophysical engrams left for them as messages from their past Self (panel A shows this stack of “Selflets”). This means that a cognitive system uses a bowtie architecture where individual experiences in the past are compressed by learning into sparse engrams that discard a lot of details to generalize from specifics. At each moment, a mind must decode those molecular/biophysical states to give them meaning – interpret them (B). It is well known in neuroscience that this process is not passive recall but active reconstruction. The very nature of the compression of learning means that we never automatically know what molecular engrams mean and must actively improvise memories on-the-fly. A time-extended mind must continuously change to remain adaptively engaged. The same paradox of change affects the body; the one thing that a biological system can count on is that nothing will stay the same: not only will the environment change but also the internal parts will eventually be mutated. This means that the system cannot take for granted (in a fixed manner) how its genomically-encoded affordances should be used: the experience of past generations is compressed within the genome but must be decoded by morphogenesis to produce an embryo or some other outcome that fits current circumstances (panel C). Thus, the right side of the bowtie (D) features a non-algorithmic process that must creatively interpret the genome and the environment in the face of missing information. This is what accounts for the remarkable plasticity (Figure 6) and reprogrammability (Figure 5) of living tissue not needing genetic change to produce novel outcomes from the same hardware (such as the exquisitely patterned galls that arise from plant cells normally making flat green leaves, E).  Images A-D by Jeremy Guay of Peregrine Creative. Panel E used with permission from [104].

            Each of us has no access to the past – only to the memory engrams in our brain (or body [109]) that were left as a consequence of past experiences. The interpretation of those engrams, as part of the neurally constructed [110-114] (not just read out) memories, is a continuously active process of self-(re)creation. At each moment, we exist at the bowtie node between the compressed, generalized lessons of the past and the opportunity to interpret our current state for maximum adaptive benefit info the future. This is not necessarily the way past instances of us interpreted the data (although it often is, under normal circumstances); the collective has competencies to do new things with the ancient prompts. Our self-constructing cognitive Self is able to do this because it is a functional pivot of far more ancient skills perfected by our morphogenetic Self [115, 116]. The cellular collective, solving problems in the context of unreliable materials and environment, must also interpret (not just slavishly obey) the memory engram left for it by our past Selves: the genome.

Evolution operating on an agential material

            Morphogenetic strategies across different lineages lie on a spectrum, with respect to how literally they take their genome. On one side exist organisms like the nematode C. elegans – eutelic organisms which all have the same number of cells; development is guided by strict lineage relationships, and each cell can be numbered – each instance  of the organism is identical at the level of cells. Such organisms have minimal plasticity/regenerative capacity, and highly prescribed, stereotyped order. In the middle are amphibians – significantly regenerative and plastic, amenable to all sorts of grafting, cutting, and other perturbations which result in coherent outcomes, not to mention the examples of newt tubules above. On the far end lies a strategy that maximally leans in to uncertainty and the need to reach specific outcomes despite unpredictably changing hardware. A fascinating example is provided by asexual planarian flatworms, which can regenerate every part of the body from even a tiny fragment [117]. Their genome is a mess (they can be mixoploid, like a tumor) [118, 119] because unlike sexually-reproducing animals, they don’t purge somatic mutations at each generation. To reproduce, they split and regenerate, meaning that every mutation that doesn’t kill a neoblast stem cell propagates itself and its altered genome into the two regenerative offspring.

            Is it not shocking, given the standard views of the genome and its role, that the animal with the highest regenerative capacity (morphogenetic stability), cancer resistance, and lack of aging (immortality), is the animal with the messiest, not the cleanest, genome?  Moreover, planaria are perhaps the only model system in which no mutant lines exist (except our 2-headed lines, and those are not genetic – see below), nor has transgenic technology been shown to be possible in these animals. Why? Our simulations [120] have suggested an explanation for these observations, showing how evolution works differently on a competent but unreliable material. In this context, “competent” means (as in the examples above) the ability to reach specific goal states despite changes of internal composition and external microenvironment – some degree of context-sensitive navigation of possibilities, not only hardwired stereotypical activity. Unreliable means that on multiple time scales, the system can make few assumptions as the genetic complement of its cells (and thus the hardware it affords) varies widely. This means that success requires more effort to be put into the creative improvisation of navigating anatomical and physiological spaces with equipment of which one cannot be certain.

            The impact of behavioral intelligence on evolution has been studied extensively [121-126]. But in developmental biology, the consequences for evolution of the role of context-specific problem-solving at all scales can be seen most clearly [127-132]. In our simulations of evolution acting on morphogenetic systems [120, 133], we observed some fascinating dynamics [105, 134, 135]. Imagine a mutation that has multiple consequences in an amphibian, one of which is that the mouth is moved to an aberrant location. In a mosaic (low-competency) organism, starvation and death would result, preventing evolution from exploring the other consequences of that mutation. Evolution would have to wait until they occurred alone, which given the pleiotropic nature of development would take a long time. However, in a plastic, high-competency system, the mouth may move to find its correct location [58, 59, 91], rescuing the animal and thus enabling the other consequences of the mutation to have impact on fitness. Competency of the agential material of life turns many deleterious mutations into neutral ones, greatly facilitating the speed of evolution.

            But it has one other important consequence. When the cellular collective repairs defects (anatomical, physiological, or transcriptional), selection does not see the hardware as clearly: a tadpole with its mouth in the right location could have had great structural genetics, or poor ones that have been made up for by the flexibility of morphogenesis. Selection never really sees genomes (since it always acts on phenotypes) but it’s critical to note that the degree to which information about a genome can be acted upon by selection depends on the degree of active intelligence of the layer mapping genetic information (hardware) into active patterns of form, physiology, and behavior. What we observed is a positive feedback loop: the more competency tissues had, the less time evolution spent optimizing the hardware and the more effort went into creating an algorithm that was effective despite faulty hardware, which in turn hid more information about the genomes from selection. Of course, different species (for different reasons of ecology) will achieve different balance points between making sure the hardware is correct and policies that can get to an adaptive endpoint despite sub-optimal hardware. Planaria are a great example of pushing all the way to the right of that continuum: because of their reproductive mode and unreliable genetics, evolution was forced to find a set of competency algorithms that allow planarian cells to make a successful worm body no matter what it starts from, effectively ignoring its natural mutations just as it ignores scientists’ attempts to introduce transgenes or make mutant lines.

            One final aspect to note is the impact of intelligence of the materials on the course of evolution itself [136-148]. Even mutations random with respect to fitness of the resulting organism, end up working differently due to the competency of the substrate which can bend the option space so that mutations’ effects are better in the future. In a sense, the ability of cells to reach specific outcomes despite novel challenges implements a kind of “chance favors the prepared mind” scenario. It’s not the mutations that have to be smart, nor does there need to be any large-scale plan to the course of its meanderings; nevertheless, the competencies of the material can lend dynamics to the evolutionary process that have simple cybernetic properties [149] themselves, in effect instantiating a process that is neither completely blind and stupid nor requires high-level forward-planning intelligence. The spectrum of the field of diverse intelligence (Figures 1,3) is not limited to the cellular and subcellular levels, but may be applied to the largest scales of space (ecosystems) and time (the evolutionary process itself) – a fascinating research program for the future.

4. Bioelectricity: a highly conserved cognitive glue mechanism

             The previous sections focused heavily on the ability of cell groups to achieve specific goal states in anatomical morphospace, even contrasting that capability with the hardware properties encoded in the genome, as targets that evolution can tune independently.  All hardware (proteins available to cells) have to be encoded genetically; what mechanisms underlie the morphogenetic policies that are in some way distinct? There are probably many, but there is one that is currently the best understood: bioelectrically-mediated cellular networks that implement computational capabilities that do not themselves have to be coded in the genetic hardware because they operate as physiological software and make use of universal information-processing patterns (e.g., as shown in Figure 5B).

Evolutionary context for bioelectricity: a common medium underlying bodies and minds

            The previous strange-sounding statement is just a different way to describe brains. It’s a useful way to do because it allows us to generalize beyond the familiar substrate of neuroscience to better understand its evolutionary history and its implications.  Brains consist of networks of electrically active cells working in networks to process information, perform active inference, store memories, etc. In particular, electrophysiological events in the brain act as a cognitive glue – a mechanism that binds individual neurons into a larger system that has goals, memories, and preferences in problem spaces that no neuron knows anything about (Figure 8). All of us have mental lives pursuing target states in highly abstract social, financial, and other spaces which none of our neurons can access because of this set of policies that binds cells into a network that can operate on larger sets of states. There are other examples of cognitive glue (e.g., stress sharing [150]) that are not bioelectric in nature, and bioelectricity is not unique in its ability to scale competent subunits into collectives that access new problem spaces and have a larger cognitive light cone (size of the setpoints it can work toward). But evolution certainly exploited bioelectric dynamics because they are ideal for implementing feedback (memory) systems that integrate information across both space and time.

            Remarkably, most of the brain’s tricks, including inference of patterns, and mental time travel to the past (memory), to the future (prediction), and to alternate worlds (counterfactual what-if’s) derive from an ancient, well-conserved system of bioelectric networks. These networks consist of cells with ion channels, electrical synapses (gap junctions), and neurotransmitter transducers – many of which go back to the time of bacterial biofilms. Ion channels have long been studied in bacteria [151], and it is now known that even bacterial biofilms already use electrical networks to join into a kind of proto-multicellularity to do things no individual bacterium can do [43, 152, 153]. The ion channels (and pumps) set each cell’s voltage potential, which can propagate to other cells via gap junctions. However, many of the channels and the gap junctions are themselves voltage-sensitive which means the potential for complex feedback loops that establish multicellular regions of different volage that change over time.

Figure 8: Scaling of the cognitive light cone by bioelectric networks.  The cognitive light cone refers to the size, in space and time, of the largest goal state a system is capable of pursuing (i.e., the size of its setpoints). Single cells pursue small setpoints, such as scalar values of physiological parameters like pH, over short ranges of size and memory/anticipation horizon (A). Joined into large-scale tissues, groups of cells can hold much larger and more complex setpoints for homeostatic and allostatic processes, which enables them to reliably pursue enormous construction projects such as rebuilding whole limbs (B).

            Cells in turn react to the bioelectric state of themselves and of their neighbors. Because the same voltage can be reached by very different ion balances, resting potential itself achieves a kind of generalization of stimulus: different past states (whether induced by sodium, potassium, chloride, etc. ions) are interpreted as the same coarse-grained parameter – “voltage potential”, which is a precursor to the formation of generalized categories in brainy organisms during learning. Cell voltage regulates second-messenger systems (and, ultimately, controls gene expression and cell movement) via a set of known transduction mechanisms which include serotonergic machinery, calcium signaling, and others [154]. What cells appear to react to is not absolute voltage states but differences between cells and their neighbors – patterns of bioelectric state across cell sheets, which often set up compartment boundaries during embryogenesis or regeneration [155-161].

            When large-scale patterns of electrical activation move rapidly through the brain, we call it the province of neuroscience. When large-scale patterns of electrical activation move much more slowly through other somatic and embryonic tissues, this is the province of developmental bioelectricity. Just as neurobiologists seek to do “neural decoding” to extract the semantic meaning of thoughts and memories from electrophysiological data of the brain [162-165], developmental physiologists can now extract information from the bioelectric dynamics of patterning tissues [166, 167]. We know that bioelectric patterns in the brain evolved to think about how to move a body in 3D space. What did ancient bioelectric networks, predating nerve and muscle, think about? They thought about moving the organism’s configuration through anatomical morphospace [168]. What evolution did, in evolving nervous systems, was co-opt a navigational problem-solving mechanism, speeding it up from minutes to milliseconds and pivoting it into a new problem space (Figures 3,8).

Developmental electrophysiology and the scaling of cognitive light cones

            A field of cells expressing ion channels and gap junction proteins is in effect an excitable medium in which complex and beautiful patterns can form from spontaneous symmetry breaking and amplification. Developmental biologists are used to Turing patterns and other reaction-diffusion dynamics in chemical media [169-172], but the same kind of phenomenon happens at the physiological level. Crucially, this happens because channels and gap junctions open and close post-translationally. Just as with the action potentials in the brain, which don’t require rapid changes of transcription or translation to implement each signaling spike, cells can change electrical properties with no changes in mRNA or protein. Consequences of this include the facts that the same bioelectric state can be caused by different ion channels’ functions, while cells with identical ion channel complements can be in different bioelectric states (because of past physiological stimuli leading the channels to be open or closed – an inherent historicity and memory in this system). Thus, bioelectric states are not derivable from transcriptomic or proteomic (and of course not genomic) data – physiomics are needed to understand the bioelectric state of cells, just as electrophysiological imaging is needed in neuroscience.

Figure 9: Developmental bioelectricity. In neurons, electrical signals are propagated across the network via ion channels that set resting potential and electrical synapses (gap junctions) which serve as context-sensitive conduits for that voltage to neighboring cells (A).  The exact same system functions in the rest of the body, allowing most kinds of cells to couple into electrical networks (B).   The result of voltage states propagating across a tissue of voltage-sensitive channels and gap junctions is complex self-organization of pattern, such as the voltage map of the early Xenopus laevis embryo anterior neurectoderm (C, shown using voltage-sensitive fluorescent dye), which guide gene expression and cell behavior and often serve as prepatterns or scaffolds for the subsequent anatomy such as the face. (D)  The information in these networks can be effectively modified (re-written) by approaches taken from neuroscience, such as using drugs or optogenetics to target specific ion channels, pumps, and gap junctions. Panels A,B,D by Jeremy Guay of Peregrine Creative. Panel in C taken with permission from [161].

            Individual cells can store tiny setpoints – metabolic, pH, and other ways to represent goals at the level of a single cell. However, joined into large-scale electrical networks, they are able to represent enormous setpoints or bioelectric prepatterns that can scaffold whole organs or bodyplan axes (Figure 9). As brains’ bioelectric patterns store memories for actions in 3D space, the body’s bioelectric network stores pattern memories that guide cell behavior toward morphogenesis of specific forms [173]. Examples include the bioelectric prepattern (Figure 9C) that precedes and functionally dictates the location of craniofacial structures [161], and the planarian voltage gradient that directs the number of heads in flatworms [174]. Three brief examples illustrate important features of this system, especially highlighting what information and tractable control points exist at the level of physiology as distinct from the genetics.

            The first is the case of cancer. Transforming cells rapidly drop gap junctional communication with their neighbors [175, 176]. This renders them incapable of remembering the grandiose morphogenetic setpoints they were working on, and they revert to their ancient unicellular goals of reproduction and survival – both behaviorally (metastasis) and transcriptionally [177, 178]. They are not more selfish than normal cells (as often modeled in game theory simulations), they just have smaller selves, in the sense that the size of an agent’s goal states (the region of spacetime it tries to manage toward specific setpoints – its cognitive light cone) defines the border between the Self and the outside world [71]. Cancer results from a shrinkage of the boundary between a Self and its world (microenvironment), a kind of dissociative identity disorder of the morphogenetic collective intelligence. It reverses the processes of development (and evolution) in which cognitive light cones massively scaled up during the advent of multicellularity. It has now been shown that forcibly re-instating the correct bioelectric state (a kind of physiological integration therapy), despite the presence of strong oncogenic mutations, normalizes cells and prevents tumors in animal models [179-182]. It doesn’t kill cells or fix their genetic defect, but once joined back into the electrical network, they form part of the collective whose goals include morphogenesis and upkeep of whole organs, re-inflating their tiny cognitive light cones to tissue- or organ-level.

The bioelectric interface for large-scale pattern control

            Misexpression of ion channels [183] or their activation via optogenetics [179] can thus regulate the size of the computational unit and its goal states. Other ways to exploit the natural bioelectric system that forms the decision-making medium of cellular collectives is to induce the formation of ectopic organs. The location of narive eyes for example is set by the presence of a specific pattern of resting potential within a set of patterns establishing the gene expression and morphogenesis of the vertebrate face (Figure 9C). A single ion channel misexpression can establish similar pattern in aberrant locations in a frog embryo, resulting in production of ectopic eyes (Figure 10A, [173]); other such voltage patterns can induce brain and other structures [184]. Several key pieces of information are revealed in that experiment. 1) The bioelectric state is instructive – it can be used to create novel coherent structures, not just cause damage. 2) The bioelectric state can encode large-scale structures: in keeping with the notion of bioelectric networks operating with generalizations (primitive concepts) of single-cell states, this particular voltage pattern means “eye” – no individual cell knows what an eye is, but the collective interprets that pattern to stand for a specific organ and executes accordingly (other patterns have been found to indicate other outcomes). We didn’t have to talk to stem cells, or micromanage the structure of an eye, we specified a very high-level prompt to which the competent material reacted (as is the key ability of all embodied cognitive agents – high-level concepts make the chemistry work to implement abstract goals). 3) The material is self-scaling: when only a few cells are targeted with the intervention, they perform a secondary instruction and recruit un-manipulated cells to join in (Figure 10B). 4) The system actually contains a battle of goal states: while the injected cells try to make their neighbors join in to the “eye” project, the surrounding cells, executing a cancer suppression mechanism (resisting rogue voltage states in the microenvironment), try to do the opposite – normalize the injected cells’ voltage to that of the local average. It is an active area of investigation for biomedical applications of bioelectric signaling to understand how to make such bioelectric messages maximally convincing to cell groups, to efficiently re-write setpoints.

Figure 10: Example: organ-level communication with cellular collectives.  When ion channel mRNA is injected into an ectopic region of an embryo, setting up an ectopic prepattern that matches the native eye prepattern from the face (Figure 9C), a new eye is induced (A).  Sectioning reveals that only a few of their cells were actually injected: the rest of the organ is made from cells which joined the morphogenetic task by being secondarily recruited (B). This is a dynamic process, as revealed when early embryos are stained using in situ hybridization for an early eye marker – Rx1 (red arrowheads reveal ectopic eye spots, C). The majority of these ectopic eye spots will in fact not make eyes because surrounding cells normalize their voltage (a cancer suppression mechanism). This reveals the dynamic nature of the instructions mediated by these physiological signals, and the nature of the message – at the level of whole organs, not gene states or cell types – that is being communicated; the specific structure and the molecular events needed, including secondary interactions between cells, are all downstream of the primary voltage pattern trigger, and don’t need to be micromanaged by the bioengineer.

            A final example concerns the repair of birth defects. Notch is an important neurogenesis gene, and when a dominant form of the Notch-ICD mutant is injected into frog embryos, severe defects of the brain result, via disruption of the normal bioelectric prepattern that sets the regionalization of the neural tube. Recently, a computational model of this process was created, which could be asked: what ion channel could be opened or closed to get the pattern back to normal? The model suggested HCN2 – a very interesting channel with the property that it acts as a kind of “sharpen filter” on bioelectric prepatterns [185-187]. Overexpresing HCN2, or opening existing HCN2 channels with drugs, restored the normal shape, gene expression, and even learning rates of Notch mutant-affected animals’ brains, revealing that at least in some cases, hardware defects can be fixed “in software”. And, consistent with the point above about bioelectric states being invisible to conventional omics, note that the number of heads in bioelectrically-induced two-head worms, the absence of tumors in KRAS-injected animals, or the shape of brains in Notch-mutant animals, could not be inferred from their genetics (which point to different, and incorrect predictions).

            The basic theme of this section can be summarized as follows. We now know quite a bit about the mechanisms responsible for cell groups’ ability to reach specific anatomical outcomes – the mapping of genotype (which sets ion channel properties) to phenotype (anatomy) is mediated by the computational properties of bioelectric networks. Bioelectricity is not yet another piece of biophysics that operates in morphogenesis, nor are the regenerative, reparative, and organ-inducing abilities of molecular bioelectric perturbations some sort of external modulation applied to the system. They only work because they exploit the native interface by which cells become an organism. The bioelectric network implements the proto-cognitive medium in tissues that stores the homeodynamic setpoints: the goal states towards which cells will build. This is important in general, because it fleshes out how collective intelligence of morphogenesis is implemented, and specifically because it offers an interface by which regenerative medicine workers and bioengineers can communicate with (not micromanage) the morphogenetic Self.

5. Novel bodies, novel minds, same hardware

            Framing morphogenesis as targeted navigation of anatomical space, with context-sensitive (but finite) capacity to forge new paths to the setpoint when circumstances demand, raises two related questions. What do problem-solving systems do when their standard goals cannot be achieved? And, where do the setpoints originate – what actually sets the target states for these homeodynamic processes? Bioengineering and synthetic morphology [188-192] are natural extensions of developmental biology which extend our understanding of growth and form along those two lines of inquiry.

            The standard answer to the origin of goal states is evolution: selection over eons in specific environments has ensured that the default hardware produced by embryogenesis has very specific anatomical setpoints. Indeed, we know that evolution is very good at managing both internal (morphogenetic) and external (shapes of bird nests etc.) targets of dynamical processes [193-195]. But, much as neuroscientists ask about the plasticity of nervous systems in handling novel circumstances, we can ask: what would be the morphogenetic, physiological, transcriptional, and behavioral properties of novel living beings that have never existed on Earth and have never faced selection at the level of the organism? When cell groups find themselves unable to reach the default target morphology, what will they do?

Figure 11: Limits of current predictive models.  Despite having full access to the genomes of the axolotl and frog, no existing model allows us to predict whether chimeric “frogolotl” embryos will have legs (like axolotls) or not (like frog larvae): the molecular information does not directly reveal decisions made by groups of cells at a larger scale (A). Similarly, in planaria, despite decades of molecular analysis, no model makes a prediction on what kind of a head will form after neoblast stem cells from species of worms with different head shapes are combined into a partially-irradiated host: will one head shape dominate? Will the result be a composite? Or will the head never stop remodeling, as neither set of cells’ complete target morphology is ever reached?  Images in A taken with permission form [262]. Image in B taken with permission from [263].

            Unfortunately, we do not yet have any frameworks in developmental biology to predict this in advance. A more basic limitation (Figure 11A) is observed in the inability to predict outcomes in chimeras [196]. For example, axolotl embryo cells (which create larval legs) can be mixed with frog embryo cells (which do not have larval legs) to make a viable embryo. Despite the availability of both frog and axolotl genomes, we have no way of predicting whether such Frogolotls will have legs, and if they do, whether these legs will consist only of axolotl cells or will recruit frog cells as well. Like in the chimeric neoblast case in planaria (Figure 11B), what is missing despite the high-resolution molecular data is the knowledge of how collective intelligences (groups of cells) make decisions and how those decisions change when the different members of the collective are seeking different setpoints. Novel computational frameworks, using multi-scale models from the collective intelligence field [197], not cellular automata suitable for single-layer systems made of stupid components, are needed to better understand, predict, and control biological outcomes. And a deep question is, what does it mean to understand and predict morphogenetic outcomes as something different than simply run a fast simulation and observe what happens [198].

            But, at least in chimeras, we have two known evolutionarily-established setpoints, even if we don’t know whether the result will be one dominant one, an intermediate case, or an oscillation. What happens when cells find themselves in entirely novel circumstances? This question can likely be studied in many organoids, spheroids, assembloids, embryoids, etc. [199-201], if attention was paid to how they navigate their various problem spaces while remaining still in 3D space.  However, two stark examples of this phenomenon are provided by model systems whose conventional motility makes it much more obvious that their properties as novel proto-organisms need to be studied.

Figure 12: Same genetic hardware, different form and function.  (A) Epithelial cells from a frog embryo combine to form “Xenobots”, which move and can replicate by combining loose epithelial cells into the next generation of Xenobots (B). Xenobot cells express hundreds of genes differently than they do in vivo (C), including a cluster of genes related to sound perception, which enables them (unlike embryos) to respond to sounds (D). Likewise, epithelial cells taken from adult human donors form Anthrobots (E) – motile proto-organisms that express >9000 genes differently than the cells do in vivo, and have capabilities such as healing neural wounds (G; Anthrobots in green, forming a cluster to knit across a wound made in human iPS-derived neurons in red). Images in A,B taken with permission from [203].  In both cases, no genetic editing, scaffolds, synthetic circuits, or other means were used to induce these complex and novel behaviors in living systems that have never existed in this configuration before and were not selected for their numerous new capabilities, specific transcriptomes, etc. Images in C,D taken with permission from [206]. Images in E-G taken with permission from [207, 208].

            The first are “Xenobots” (Figure 12), made from epithelial frog embryo cells. These self-assemble into spherical constructs which move, change direction, and even assemble copies of themselves from loose cells in their environment (kinematic self-replication) [202-205]. They have a unique transcriptome with hundreds of genes differently expressed than the cells do in vivo, including a cluster of hearing-related genes which led to the discovery of their ability to respond to acoustic stimuli in ways that embryos do not [206].  The second are Anthrobots [207-209], made from human adult tracheal epithelial cells. These have four different motility modes with known transition probabilities between behaviors, several discrete morphotypes, over 9000 differentially-expressed genes (about half the genome!), a partially re-set epigenetic aging clock (they are younger than the cells they came from), and the ability to heal neural wounds in their environment. None of these capabilities were predicted in advance prior to their discovery.

            Anthrobots reveal that the remarkable properties of Xenobots are not some unique feature of amphibian or embryonic cells: somatic, 100% Homo sapiens cells from elderly donors take on a novel life-style when liberated from their normal environment. In both types of biobots, no scaffolds, DNA editing, transgenes, or nanomaterials are added. Their novel properties derive from subtraction of instructive cues provided in vivo that mask their intrinsic capabilities in forcing them to boring epithelial fates. Allowing them to reboot their multicellularity shows the plasticity and novel setpoints that the reliability of development obscures. Studies are under way to determine their proto-cognitive capacities (i.e., learning abilities and other problem-solving capacities).

            Crucially, it is not enough, after such living constructs are made, to observe their surprising properties and note that they are compatible with the ciliated nature of their parent tissues etc. Our goal, if we want to understand evolution, collective intelligence, and the roadmap to regenerative medicine, is not merely to record “emergent” outcomes after the fact, but to understand the option space from which such forms and behaviors are drawn, and to be able to predict and guide specific novel setpoints in advance. Moreover, there have never been any Anthrobots or Xenobots, and there’s never been selection for any of their transcriptional, morphological, and behavioral properties. Whence their very specific traits? How much specificity between past environments and the resulting traits should we expect, in an evolutionary origin story? Of course, there are many ways for traits to be established other than from direct selection for them, but it is not enough to say that these capacities were somehow acquired at the same time that human and frog environments drove selection for standard frog and human bodies, gene expression cascades, etc. The next opportunities in this field are to develop rigorous theories of how and when the computational cost for designing good Anthrobot and Xenobot capacities were paid (since they were not paid in the same way as human and frog traits were). More broadly, we must develop heuristics for understanding how the agential materials of life navigate the option space of form and behavior in truly novel circumstances that provided evolutionary guidance at the lowest levels (re-use of cells and molecules that do have a long history) but offer no tight historical information on how to be a multicellular organism with 100% conventional parts but a novel lifestyle.

6. Future impacts of developmental biology

            At one point, not that long ago, the way to control a computer was by physically rewiring it. In other words, one had to interact with it at the lowest level. It is laughable nowadays, to suggest that switching among applications on one’s laptop should require a soldering iron. That is because if the hardware is good enough, it can be controlled with stimuli, and reprogrammed, not micromanaged. This gave rise to the information technology revolution. I argue that the life sciences today, with all of the emphasis on genomic editing, single-molecule approaches, protein engineering, etc. are roughly where computer engineering was in the 1940’s and 1950’s: it’s all about the hardware. The next decades will complement this essential body of knowledge with new tools focused on understanding not only the software of life, but its essential agentic nature. Developmental biology plays a central role in a number of other fields (Figure 13).

Figure 13: A mind-map of developmental biology’s essential contributions to other basic and applied fields. As shown in this mindmap, the study of morphogenesis impacts efforts including diverse intelligence, artificial life, computer science/AI, evolution, engineering, regenerative medicine, and foundational issues in philosophy.

       Living matter is not just a computational material, offering reprogrammability and plasticity of outcome without changing the DNA. It is more than that – it is an agential material, composed of multi-level autonomous subsystems at many scales of organization that have agendas and the ability to behavior-shape other subsystems towards setpoints in diverse spaces. This means that as useful as the software-hardware paradigm is in rising beyond molecular determinism, it is limited; no current machine metaphor is sufficient to interact with life at its fullest. More appropriate formalisms, based on systems which dynamically improvise the meaning of their memories [102] and erase the boundaries between active agents and passive information [210], are being developed.

Biomedicine

            One of the key areas that will be impacted by these advances is regenerative medicine [13]. We currently have increasing control over molecular events, but definitive treatments for birth defects, missing or damaged organs, aging, and cancer elude us. A major limitation is the lack of cures. With the exception of surgery, antibiotics, and single-gene diseases amenable to gene therapy, there are very few therapies that solve a problem permanently: typically it comes back with a vengeance when the drug is stopped (or earlier, when the tissue habituates to it). Massive differences across patients in efficacy and side effects arise, in part, because cells and multicellular collectives have not bought in to the new regime and often resist [211]. New computational models of multiscale autonomous cybernetic systems [54, 212, 213] will enable strategies that do not attempt to force specific molecular states but re-write setpoints, which the cells will pursue autonomously.

            Part of the research program is to develop convincing strategies for effectively manipulating the encoded goal states of physiological, transcriptional, and many other subsystems. Cells, with a long evolutionary history of being subject to hacking and exploitation attempts by other biota, likely have many ways to resist signals that they perceive as coming from outside. Getting the cells’ and tissues’ buy-in, so that they don’t try to contravene the treatment but rather continue to maintain health after the intervention is withdrawn, requires a better understanding of their setpoints, priors (physiological memories), and reprogrammability. One example of this is chemically-induced leg regeneration in amphibia, where a 24-hour intervention leads to 1.5 years of appendage growth with no further manipulations needed [214]. Another example is bioelectric eye induction [173], where cells misexpressing an ion channel actively recruit their neighbors to participate in the novel organogenesis (Figure 10B), or, ignore the exogenous bioelectric state entirely if persuaded by their neighbors to do so (as part of a cancer suppression mechanism).

Bioengineering

            Another field that will be massively impacted by developmental biology is bioengineering. As we move from a focus on life-as-it-is to life-as-it-can-be [215], an understanding of the plasticity and capability of living material at all scales is crucial. Synthetic biology – rewiring the phenotypes of single cells [216-218] – will expand to synthetic morphology [188-192, 219-223], as we learn to communicate larger-scale goal states. Ultimately, this will lead to a functional anatomical compiler [14] – a tool that will allow scientists and engineers to probe and access all possible regions of the latent space of form and function.

            Bioengineering will benefit from a dissolution of artificial walls between the sciences of the body and those of the mind [224] not only because of useful new living machines, but also because of the increasing development of practical tools [225] for mutually-enriching feedback. Indeed, the interoperability of life makes possible a very wide variety of instrumentized ex vivo constructs [226-233] – mergers of life with passive or AI-powered electronic components, which will need the insights of developmental biology and developmental psychology to guide both applications and ethics.

            In many ways, communicating with cells and tissues is analogous to the SETI challenge [234-237]: these are somewhat alien intelligences, who operate in a problem space difficult for us to visualize, and have capabilities and setpoints we do not yet understand. One of the most exciting tools to assist with this challenge is artificial intelligence. The role of AI in this domain is not limited to data mining or analysis [238]. It holds the potential to be used as a translator interface to enable effective communication with these systems. Notice that this is similar to how language-using brain regions allow one human to, albeit indirectly, access and communicate with the private physiological information implemented by neuronal machinery in another human’s brains. The full potential of this technology is when it is used as an “agential translation layer” – a sophisticated problem-solving system motivated to understand the beings on both sides of the equation and improve their interaction.

Cognitive science and philosophy of mind

“Developmental biologists don’t concern themselves with the mind-body problem.”    — anonymous reviewer

            If developmental biologists don’t concern themselves with the mind-body problem, who possibly could?  On the contrary; developmental biology, expanded by modern tools and approaches that transcend traditional disciplinary boundaries, is uniquely placed to provide scientific advances in an area previously left to philosophers: the origin and nature of embodied minds. Mechanisms of scaling of cognition from chemistry to psychology, the shape of the morphospace of possible embodied minds, the role of reductionist strategies in science [239], machine and computational metaphors for mind and life [240], the mechanisms of top-down causation (Figure 14), the relationships between scientists/engineers and their increasingly autonomous materials – these are all crucial areas in which developmental biology can contribute essential ideas, model systems, and best of all, actual data that constrain and enable better philosophy.

Figure 14: Morphogenesis and cognition both feature multiscale control.  (A) One of the most remarkable aspects of developmental biology is that it illustrates how a system slowly progresses not just in complexity, but continuously traverses disciplines that are often taken to treat categorically distinct kinds of subjects. Starting as an unfertilized oocyte that is thought to be suitably handled by chemistry, a being self-assembles and journeys through developmental physiology, behavioral science, and perhaps psychology, with possible detours including oncology and bioengineering (which can allow fragments to persist long after their human donor has become deceased). (B) When a tail is grafted onto the flank of an amphibian, it slowly remodels into a limb. Consider the cells marked in red – tail tip cells, sitting at the end of a tail, which nevertheless become fingers. There is no local damage or injury, and yet their molecular signals and cellular architecture change in accordance with a high-level, to them abstract and unknowable, body-wide target morphology which forces modification toward a large-scale outcome more appropriate for the species. (C) This kind of top-down control of molecular events by higher-level goal states is not just for rare cases of mind-body medicine (such as changes of gene transcription and molecular signaling cascades by meditation or mental expectation of drug activity in placebo experiments [264-269]). It is also precisely what happens every day, when the bioelectric system of the body enables the highly abstract goals of a human being to change the chemistry of ions crossing muscle cell membranes during voluntary motion. The top-down control of chemistry by cognitive constructs is a central feature of nervous systems, but they coopted this basic trick from far more ancient bioelectrical systems first used to control morphogenesis. Panels in A by Jeremy Guay. Panel in B modified after [88]. Panel C by GPT-5.

            It is especially timely now, with the arrival of AI in everyday life [241]. The ubiquitous conversations about the difference between “machines” and “real beings” [242-246] lead to deep questions about future relationships with not only software agents but with much more challenging mergers of human bodies and technology in cyborgs [247] (Figure 15). The vast majority of these discussions are missing basic aspects of developmental biology, and will be greatly enhanced by a wider appreciation of the unique science that addresses the origins of everything that makes us special (in the widest possible sense of “us” [248]).

Figure 15: The future of developmental biology.  (A) Understanding the multi-scale, self-assembling properties of bodies and minds is not only about the continuum of increasing agency (symbolized by the yellow glow) of ontogenetic and phylogenetic change (vertical). It is also essential to understand the forthcoming changes, biological and technological, which will expand the natural living form in ways we can barely imagine.  The early tests of social and ethical frameworks, which must be grounded in real biology, will not be in the relatively easy distinctions between software AI’s and living forms, but by the inevitable appearance of composite beings – cyborgs which defy our pre-scientific, outdated categories of “life” vs. “machine” (panel A’ schematizes futile attempts to figure out if a being is “man” or “machine”, by measuring the precise amount of engineered vs. evolved tissue). Developmental biology, in its essential embrace of the development of mind-ful beings from a drop of chemicals in the oocyte (i.e., an implicit acceptance that we and the inanimate world are part of one continuuum), is uniquely and ideally placed to help humanity through this transition. Images by Jeremy Guay of Peregrine Creative.

Acknowledgements:

            I thank Richard Watson, Scott Gilbert, Lev Beloussov, Katrina Schleisman, Mark Solms, Alexey Tolchinsky, Chris Fields, Tomas Pollak, Pamela Lyon, and many members of the Levin Lab, as well as numerous members of the developmental biology and diverse intelligence communities for many useful discussions. I thank Igor Adameyko and Gerhard Schlosser for helpful comments on the manuscript. The paper is dedicated to Alexander Gurwitsch, an early pioneer of key concepts in this area.

References cited

1.         Turing, A.M., The Chemical Basis of Morphogenesis. Philosophical Transactions of the Royal Society of London Series B-Biological Sciences, 1952. 237(641): p. 37-72.

2.         Turing, A.M., Computing machinery and intelligence. Mind, 1950. 59(236): p. 433-460.

3.         Levin, M., Technological Approach to Mind Everywhere: An Experimentally-Grounded Framework for Understanding Diverse Bodies and Minds. Frontiers in Systems Neuroscience, 2022. 16: p. 768201.

4.         Levin, M., Bioelectric networks: the cognitive glue enabling evolutionary scaling from physiology to mind. Anim Cogn, 2023.

5.         Noble, D., A theory of biological relativity: no privileged level of causation. Interface Focus, 2012. 2(1): p. 55-64.

6.         Hoel, E. and M. Levin, Emergence of informative higher scales in biological systems: a computational toolkit for optimal prediction and control. Commun Integr Biol, 2020. 13(1): p. 108-118.

7.         Klein, B. and E. Hoel Uncertainty and causal emergence in complex networks. arXiv e-prints, 2019.

8.         Albantakis, L., et al., What Caused What? A Quantitative Account of Actual Causation Using Dynamical Causal Networks. Entropy (Basel), 2019. 21(5).

9.         Hoel, E.P., Agent Above, Atom Below: How Agents Causally Emerge from Their Underlying Microphysics, in Wandering Towards a Goal: How Can Mindless Mathematical Laws Give Rise to Aims and Intention?, A. Aguirre, B. Foster, and Z. Merali, Editors. 2018, Springer International Publishing: Cham. p. 63-76.

10.      Hoel, E.P., When the Map Is Better Than the Territory. Entropy, 2017. 19(5).

11.      Hoel, E.P., et al., Can the macro beat the micro? Integrated information across spatiotemporal scales. Neuroscience of Consciousness, 2016. 2016(1): p. niw012.

12.      Hoel, E.P., L. Albantakis, and G. Tononi, Quantifying causal emergence shows that macro can beat micro. Proceedings of the National Academy of Sciences of the United States of America, 2013. 110(49): p. 19790-5.

13.      Levin, M., The Multiscale Wisdom of the Body: Collective Intelligence as a Tractable Interface for Next-Generation Biomedicine. Bioessays, 2024: p. e202400196.

14.      Davies, J. and M. Levin, Synthetic morphology with agential materials. Nature Reviews Bioengineering, 2023. 1: p. 46-59.

15.      Wigner, E.P., The Unreasonable Effectiveness of Mathematics in the Natural Sciences. Communications on Pure and Applied Mathematics, 1960. 13.

16.      Markus, M., D. Bohm, and M. Schmick, Simulation of vessel morphogenesis using cellular automata. Math Biosci, 1999. 156(1-2): p. 191-206.

17.      Lehotzky, D. and G.K.H. Zupanc, Cellular Automata Modeling of Stem-Cell-Driven Development of Tissue in the Nervous System. Dev Neurobiol, 2019. 79(5): p. 497-517.

18.      Hatzikirou, H. and A. Deutsch, Cellular automata as microscopic models of cell migration in heterogeneous environments. Curr Top Dev Biol, 2008. 81: p. 401-34.

19.      Akberdin, I.R., et al., A cellular automaton to model the development of primary shoot meristems of Arabidopsis thaliana. J Bioinform Comput Biol, 2007. 5(2B): p. 641-50.

20.      Morgan, C.L., Other minds than ours, in An Introduction to Comparative Psychology, W. Scott, Editor. 1903. p. 59-.

21.      Lobo, D., et al., A linear-encoding model explains the variability of the target morphology in regeneration. Journal of the Royal Society, Interface / the Royal Society, 2014. 11(92): p. 20130918.

22.      James, W., The principles of psychology. American science–Advanced course. 1890, New York,: H. Holt and company.

23.      Cannon, W.B., The wisdom of the body. Rev. and enl. ed. 1932, New York,: Norton. xviii, 333 p.

24.      Cannon, W.B., ORGANIZATION FOR PHYSIOLOGICAL HOMEOSTASIS. Physiological Reviews, 1929. 9(3): p. 399-431.

25.      Rosenblueth, A., N. Wiener, and J. Bigelow, Behavior, purpose, and teleology. Philosophy of Science, 1943. 10: p. 18-24.

26.      Matsushita, Y. and K. Kaneko, Homeorhesis in Waddington’s landscape by epigenetic feedback regulation. Physical Review Research, 2020. 2(2): p. 023083.

27.      Colditz, I.G., A consideration of physiological regulation from the perspective of Bayesian enactivism. Physiol Behav, 2020. 214: p. 112758.

28.      Webster, M., Morphological homeostasis in the fossil record. Semin Cell Dev Biol, 2019. 88: p. 91-104.

29.      Chuang, J.S., Z. Frentz, and S. Leibler, Homeorhesis and ecological succession quantified in synthetic microbial ecosystems. Proc Natl Acad Sci U S A, 2019. 116(30): p. 14852-14861.

30.      McEwen, B.S., Stress, adaptation, and disease. Allostasis and allostatic load. Ann N Y Acad Sci, 1998. 840: p. 33-44.

31.      Schulkin, J. and P. Sterling, Allostasis: A Brain-Centered, Predictive Mode of Physiological Regulation. Trends Neurosci, 2019. 42(10): p. 740-752.

32.      Moore, D.G., et al., Inform: Efficient Information-Theoretic Analysis of Collective Behaviors. Front Robot AI, 2018. 5(60): p. 60.

33.      Albantakis, L., et al., What caused what? An irreducible account of actual causation. arXiv, 2017. arXiv:1708.06716.

34.      Hoel, E.P., et al., Can the macro beat the micro? Integrated information across spatiotemporal scales. Neurosci Conscious, 2016. 2016(1): p. niw012.

35.      Hoel, E.P., L. Albantakis, and G. Tononi, Quantifying causal emergence shows that macro can beat micro. Proc Natl Acad Sci U S A, 2013. 110(49): p. 19790-5.

36.      Noble, D., The aims of systems biology: between molecules and organisms. Pharmacopsychiatry, 2011. 44 Suppl 1: p. S9-S14.

37.      Noble, D., Biophysics and systems biology. Philosophical transactions. Series A, Mathematical, physical, and engineering sciences, 2010. 368(1914): p. 1125-39.

38.      Levin, M., Technological Approach to Mind Everywhere: An Experimentally-Grounded Framework for Understanding Diverse Bodies and Minds. Front Syst Neurosci, 2022. 16: p. 768201.

39.      Mathews, J., et al., Cellular signaling pathways as plastic, proto-cognitive systems: Implications for biomedicine. Patterns (N Y), 2023. 4(5): p. 100737.

40.      Lagasse, E. and M. Levin, Future medicine: from molecular pathways to the collective intelligence of the body. Trends Mol Med, 2023. 29(9): p. 687-710.

41.      Yang, C.Y., et al., Encoding Membrane-Potential-Based Memory within a Microbial Community. Cell Syst, 2020. 10(5): p. 417-423 e3.

42.      Martinez-Corral, R., et al., Metabolic basis of brain-like electrical signalling in bacterial communities. Philos Trans R Soc Lond B Biol Sci, 2019. 374(1774): p. 20180382.

43.      Prindle, A., et al., Ion channels enable electrical communication in bacterial communities. Nature, 2015. 527(7576): p. 59-63.

44.      Davis, G.V., et al., Toward uncovering an operating system in plant organs. Trends Plant Sci, 2023.

45.      Johnston, I.G. and G.W. Bassel, Identification of a bet-hedging network motif generating noise in hormone concentrations and germination propensity in Arabidopsis. J R Soc Interface, 2018. 15(141).

46.      Bassel, G.W., Information Processing and Distributed Computation in Plant Organs. Trends Plant Sci, 2018. 23(11): p. 994-1005.

47.      Topham, A.T., et al., Temperature variability is integrated by a spatially embedded decision-making center to break dormancy in Arabidopsis seeds. Proc Natl Acad Sci U S A, 2017. 114(25): p. 6629-6634.

48.      Yokawa, K., et al., Anaesthetics stop diverse plant organ movements, affect endocytic vesicle recycling and ROS homeostasis, and block action potentials in Venus flytraps. Ann Bot, 2018. 122(5): p. 747-756.

49.      Ciszak, M., et al., Plant shoots exhibit synchronized oscillatory motions. Commun Integr Biol, 2016. 9(5): p. e1238117.

50.      Calvo, P., F. Baluska, and A. Sims, “Feature Detection” vs. “Predictive Coding” Models of Plant Behavior. Front Psychol, 2016. 7: p. 1505.

51.      Biswas, S., W. Clawson, and M. Levin, Learning in Transcriptional Network Models: Computational Discovery of Pathway-Level Memory and Effective Interventions. Int J Mol Sci, 2022. 24(1).

52.      Biswas, S., et al., Gene Regulatory Networks Exhibit Several Kinds of Memory: Quantification of Memory in Biological and Random Transcriptional Networks. iScience, 2021. 24(3): p. 102131.

53.      Gyurkó, D.M., et al., Adaptation and learning of molecular networks as a description of cancer development at the systems-level: potential use in anti-cancer therapies. Semin Cancer Biol, 2013. 23(4): p. 262-9.

54.      Csermely, P., et al., Learning of Signaling Networks: Molecular Mechanisms. Trends Biochem Sci, 2020. 45(4): p. 284-294.

55.      Levin, M., Bioelectric signaling: Reprogrammable circuits underlying embryogenesis, regeneration, and cancer. Cell, 2021. 184(8): p. 1971-1989.

56.      Mintz, B. and K. Illmensee, Normal genetically mosaic mice produced from malignant teratocarcinoma cells. Proc Natl Acad Sci U S A, 1975. 72(9): p. 3585-9.

57.      Mintz, B., Genetic mosaicism in vivo: development and disease in allophenic mice. Fed Proc, 1971. 30(3): p. 935-43.

58.      Vandenberg, L.N., D.S. Adams, and M. Levin, Normalized shape and location of perturbed craniofacial structures in the Xenopus tadpole reveal an innate ability to achieve correct morphology. Developmental Dynamics, 2012. 241(5): p. 863-78.

59.      Pinet, K., et al., Adaptive correction of craniofacial defects in pre-metamorphic Xenopus laevis tadpoles involves thyroid hormone-independent tissue remodeling. Development, 2019. 146(14).

60.      Harrison, L.A., et al., Resilience phenotypes derived from an active inference account of allostasis. Front Behav Neurosci, 2025. 19: p. 1524722.

61.      Adar, O., J.D. Shakargy, and Y. Ilan, The Constrained Disorder Principle: Beyond Biological Allostasis. Biology (Basel), 2025. 14(4).

62.      Sennesh, E., et al., Interoception as modeling, allostasis as control. Biol Psychol, 2022. 167: p. 108242.

63.      Bubenik, A., Das Geweih. 1966, Hamburg: Paul Parey Verlag.

64.      Bubenik, A.B. and R. Pavlansky, Trophic responses to trauma in growing antlers. J Exp Zool, 1965. 159(3): p. 289-302.

65.      Fields, C. and M. Levin, Competency in Navigating Arbitrary Spaces as an Invariant for Analyzing Cognition in Diverse Embodiments. Entropy (Basel), 2022. 24(6).

66.      Gawne, R., K.Z. McKenna, and M. Levin, Competitive and Coordinative Interactions between Body Parts Produce Adaptive Developmental Outcomes. Bioessays, 2020. 42(8): p. e1900245.

67.      Tung, A., et al., Embryos assist morphogenesis of others through calcium and ATP signaling mechanisms in collective teratogen resistance. Nat Commun, 2024. 15(1): p. 535.

68.      Dewan, E.M., Consciousness as an emergent causal agent in the context of control system theory, G. Globus, G. Maxwell, and I. Savodnik, Editors. 1976, Plenum Press: New York. p. 179-198.

69.      Pezzulo, G. and M. Levin, Top-down models in biology: explanation and control of complex living systems above the molecular level. J R Soc Interface, 2016. 13(124).

70.      Gray, R.S., I. Roszko, and L. Solnica-Krezel, Planar cell polarity: coordinating morphogenetic cell behaviors with embryonic polarity. Dev Cell, 2011. 21(1): p. 120-33.

71.      Levin, M., The Computational Boundary of a “Self”: Developmental Bioelectricity Drives Multicellularity and Scale-Free Cognition. Frontiers in Psychology, 2019. 10(2688): p. 2688.

72.      Rubin, H., Cancer as a dynamic developmental disorder. Cancer Res, 1985. 45(7): p. 2935-42.

73.      Rubin, H., Ordered heterogeneity and its decline in cancer and aging. Advances in cancer research, 2007. 98: p. 117-47.

74.      Rubin, H., What keeps cells in tissues behaving normally in the face of myriad mutations? BioEssays, 2006. 28(5): p. 515-24.

75.      Rubin, H., M. Chow, and A. Yao, Cellular aging, destabilization, and cancer. Proc Natl Acad Sci U S A, 1996. 93(5): p. 1825-30.

76.      Ellis, G.F.R. and C.S. Bloch, Top-Down Causation, Adaptive Selection, and Their Consequences. Human Development, 2011. 54(2): p. 93-100.

77.      Juarrero, A., Top-Down Causation and Autonomy in Complex Systems. Downward Causation and the Neurobiology of Free Will, 2009: p. 83-102.

78.      Ellis, G.F.R., Top-Down Causation and the Human Brain. Downward Causation and the Neurobiology of Free Will, 2009: p. 63-81.

79.      Auletta, G., G.F. Ellis, and L. Jaeger, Top-down causation by information control: from a philosophical problem to a scientific research programme. Journal of the Royal Society, Interface / the Royal Society, 2008. 5(27): p. 1159-72.

80.      Craver, C.F. and W. Bechtel, Top-down causation without top-down causes. Biology & Philosophy, 2007. 22(4): p. 547-563.

81.      Okasha, S., Emergence, hierarchy and top-down causation in evolutionary biology. Interface focus, 2012. 2(1): p. 49-54.

82.      Cliff, O.M., et al., Quantifying Long-Range Interactions and Coherent Structure in Multi-Agent Dynamics. Artif Life, 2017. 23(1): p. 34-57.

83.      Wibral, M., et al., Local active information storage as a tool to understand distributed neural information processing. Front Neuroinform, 2014. 8: p. 1.

84.      Lizier, J.T., JIDT: An Information-Theoretic Toolkit for Studying the Dynamics of Complex Systems. Frontiers in Robotics and AI, 2014. 1(11).

85.      Wang, X.R., et al., Quantifying and tracing information cascades in swarms. PLoS One, 2012. 7(7): p. e40084.

86.      Lizier, J.T., et al., Multivariate information-theoretic measures reveal directed information structure and task relevant changes in fMRI connectivity. J Comput Neurosci, 2011. 30(1): p. 85-107.

87.      Varley, T.F. and E. Hoel, Emergence as the conversion of information: a unifying theory. Philos Trans A Math Phys Eng Sci, 2022. 380(2227): p. 20210150.

88.      Farinella-Ferruzza, N., The transformation of a tail into a limb after xenoplastic transformation. Experientia, 1956. 15: p. 304-305.

89.      Farinella-Ferruzza, N., Risultati di trapianti di bottone codale di urodeli su anuri e vice versa. Rivista di Biologia, 1953. 45: p. 523-527.

90.      Holtfreter, J., Transformation of a Tail into a Limb or Gill-Like Structures. Journal of Experimental Zoology, 1955. 129(3): p. 623-648.

91.      Pinet, K. and K.A. McLaughlin, Mechanisms of physiological tissue remodeling in animals: Manipulating tissue, organ, and organism morphology. Dev Biol, 2019. 451(2): p. 134-145.

92.      Boyle, E.A., Y.I. Li, and J.K. Pritchard, An Expanded View of Complex Traits: From Polygenic to Omnigenic. Cell, 2017. 169(7): p. 1177-1186.

93.      Fields, C. and M. Levin, Multiscale memory and bioelectric error correction in the cytoplasm–cytoskeleton-membrane system. Wiley Interdisciplinary Reviews: Systems Biology and Medicine, 2017. 10(2): p. e1410-n/a.

94.      Beisson, J. and T.M. Sonneborn, Cytoplasmic Inheritance of the Organization of the Cell Cortex in Paramecium Aurelia. Proc Natl Acad Sci U S A, 1965. 53: p. 275-82.

95.      Oviedo, N.J., et al., Long-range neural and gap junction protein-mediated cues control polarity during planarian regeneration. Dev Biol, 2010. 339(1): p. 188-99.

96.      Emmons-Bell, M., et al., Regenerative Adaptation to Electrochemical Perturbation in Planaria: A Molecular Analysis of Physiological Plasticity. iScience, 2019. 22: p. 147-165.

97.      Fankhauser, G., Maintenance of normal structure in heteroploid salamander larvae, through compensation of changes in cell size by adjustment of cell number and cell shape. Journal of Experimental Zoology, 1945. 100(3): p. 445-455.

98.      Fankhauser, G., The Effects of Changes in Chromosome Number on Amphibian Development. The Quarterly Review of Biology, 1945. 20(1): p. 20-78.

99.      Cooke, J., Scale of body pattern adjusts to available cell number in amphibian embryos. Nature, 1981. 290(5809): p. 775-8.

100.    Cooke, J., Cell number in relation to primary pattern formation in the embryo of Xenopus laevis. I:  The cell cycle during new pattern formation in response to implanted organisers. Journal of Embryology and Experimental Morphology, 1979. 51: p. 165-182.

101.    Blackiston, D.J. and M. Levin, Ectopic eyes outside the head in Xenopus tadpoles provide sensory data for light-mediated learning. The Journal of experimental biology, 2013. 216(Pt 6): p. 1031-40.

102.    Levin, M., Self-Improvising Memory: A Perspective on Memories as Agential, Dynamically Reinterpreting Cognitive Glue. Entropy (Basel), 2024. 26(6).

103.    Van Valen, L., A new evolutionary law. Evolutionary Theory, 1973. 1: p. 1-30.

104.    Hartl, B. and M. Levin, What does evolution make? Learning in living lineages and machines. Trends Genet, 2025. 41(6): p. 480-496.

105.    Frank, S.A., The common patterns of nature. J Evol Biol, 2009. 22(8): p. 1563-85.

106.    Frank, S.A., Maladaptation and the paradox of robustness in evolution. PLoS One, 2007. 2(10): p. e1021.

107.    Doyle, J.C. and M. Csete, Architecture, constraints, and behavior. Proceedings of the National Academy of Sciences of the United States of America, 2011. 108 Suppl 3: p. 15624-30.

108.    Csete, M. and J. Doyle, Bow ties, metabolism and disease. Trends Biotechnol, 2004. 22(9): p. 446-50.

109.    Blackiston, D., T. Shomrat, and M. Levin, The Stability of Memories During Brain Remodeling: a Perspective. Communicative & Integrative Biology, 2015. 8(5): p. e1073424.

110.    Davis, D., et al., The Multiple Roles of Emotion in Interpretation and Memory of Sexual Consent. Top Cogn Sci, 2024. 16(4): p. 644-660.

111.    Otgaar, H., et al., Oversimplifications and Misrepresentations in the Repressed Memory Debate: A Reply to Ross. J Child Sex Abus, 2023. 32(1): p. 116-126.

112.    Shao, X., et al., Dynamic changes in neural representations underlie the repetition effect on false memory. Neuroimage, 2022. 259: p. 119442.

113.    Gallo, D.A., False memories and fantastic beliefs: 15 years of the DRM illusion. Mem Cognit, 2010. 38(7): p. 833-48.

114.    Roediger, H.L. and K.B. McDermott, Creating false memories: Remembering words not presented in lists. Journal of experimental psychology, 1995. 21(4): p. 803.

115.    Maturana, H.R. and F.J. Varela, Autopoiesis and Cognition the Realization of the Living, in Boston Studies in the Philosophy of Science,. 1980, Springer Netherlands,: Dordrecht. p. 1 online resource (180 pages).

116.    Rosen, R., Anticipatory systems : philosophical, mathematical, and methodological foundations. 1st ed. IFSR international series on systems science and engineering ; v. 1. 1985, Oxford, England ; New York: Pergamon Press. x, 436.

117.    Saló, E., et al., Planarian regeneration: achievements and future directions after 20 years of research. Int J Dev Biol, 2009. 53(8-10): p. 1317-27.

118.    Guo, L., et al., An adaptable chromosome preparation methodology for use in invertebrate research organisms. BMC Biology, 2018. 16(1): p. 25.

119.    Nishimura, O., et al., Unusually Large Number of Mutations in Asexually Reproducing Clonal Planarian Dugesia japonica. PLoS One, 2015. 10(11): p. e0143525.

120.    Shreesha, L. and M. Levin, Cellular Competency during Development Alters Evolutionary Dynamics in an Artificial Embryogeny Model. Entropy, 2023. 25(1): p. 131.

121.    Thorpe, W.H., Animal Learning and Evolution. Nature, 1945. 156(3950): p. 46-46.

122.    Hinton, G.E. and J. Nowlan, How learning can guide evolution. Complex Systems, 1987. 1(3): p. 495-502.

123.    Dopazo, H., et al., A model for the interaction of learning and evolution. Bulletin of Mathematical Biology, 2001. 63(1): p. 117-34.

124.    Piaget, J., Behaviour and Evolution. 1976.

125.    Baldwin, J.M., A new factor in evolution. The American Naturalist, 1896. 30: p. 441–45.

126.    Odling-Smee, F.J., Niche construction, genetic evolution and cultural change. Behav Processes, 1995. 35(1-3): p. 195-205.

127.    Baluska, F., et al., Sensing, feeling and sentience in unicellular organisms and living cells. Biosystems, 2025. 247: p. 105374.

128.    Reber, A.S., et al., The CBC theory and its entailments : Why current models of the origin of consciousness fail. EMBO Rep, 2024. 25(1): p. 8-12.

129.    Reber, A.S., et al., The sensual cell: Feeling and affect in unicellular species. Biosystems, 2024. 238: p. 105197.

130.    Miller, W.B., Jr., F. Baluska, and A.S. Reber, A revised central dogma for the 21st century:all biology is cognitive information processing. Prog Biophys Mol Biol, 2023. 182: p. 34-48.

131.    Baluska, F., A.S. Reber, and W.B. Miller, Jr., Cellular sentience as the primary source of biological order and evolution. Biosystems, 2022. 218: p. 104694.

132.    Baluska, F., W.B. Miller, and A.S. Reber, Cellular and evolutionary perspectives on organismal cognition: from unicellular to multicellular organisms. Biological Journal of the Linnean Society, 2022.

133.    Hartl, B., S. Risi, and M. Levin, Evolutionary Implications of Self-Assembling Cybernetic Materials with Collective Problem-Solving Intelligence at Multiple Scales. Entropy, 2024. 26(7): p. 532.

134.    Frank, S.A., Evolutionary design of regulatory control. I. A robust control theory analysis of tradeoffs. J Theor Biol, 2019. 463: p. 121-137.

135.    Frank, S.A., Evolutionary design of regulatory control. II. Robust error-correcting feedback increases genetic and phenotypic variability. J Theor Biol, 2019. 468: p. 72-81.

136.    Miller, W.B., Jr., F. Baluska, and J.S. Torday, Cellular senomic measurements in Cognition-Based Evolution. Prog Biophys Mol Biol, 2020. 156: p. 20-33.

137.    Miller, W.B., Jr., J.S. Torday, and F. Baluska, Biological evolution as defense of ‘self’. Prog Biophys Mol Biol, 2018. 142: p. 54-74.

138.    Baluska, F. and W.B. Miller, Jr., Senomic view of the cell: Senome versus Genome. Commun Integr Biol, 2018. 11(3): p. 1-9.

139.    Shapiro, J.A., Engines of innovation: biological origins of genome evolution. Biological Journal of the Linnean Society, 2022.

140.    Shapiro, J.A., Evolution : a view from the 21st century. 2011, Upper Saddle River, N.J.: FT Press Science. xi, 253 p.

141.    Noble, D., Modern physiology vindicates Darwin’s dream. Exp Physiol, 2022. 107(9): p. 1015-1028.

142.    Buckley, C.L., et al., Natural Induction: Spontaneous Adaptive Organisation without Natural Selection. Entropy (Basel), 2024. 26(9).

143.    Watson, R.A., M. Levin, and C.L. Buckley, Design for an Individual: Connectionist Approaches to the Evolutionary Transitions in Individuality. Frontiers in Ecology and Evolution, 2022. 10.

144.    Rago, A., et al., How adaptive plasticity evolves when selected against. PLoS Comput Biol, 2019. 15(3): p. e1006260.

145.    Kouvaris, K., et al., How evolution learns to generalise: Using the principles of learning theory to understand the evolution of developmental organisation. PLoS Comput Biol, 2017. 13(4): p. e1005358.

146.    Watson, R.A. and E. Szathmary, How Can Evolution Learn? Trends Ecol Evol, 2016. 31(2): p. 147-157.

147.    Watson, R.A., et al., Evolutionary Connectionism: Algorithmic Principles Underlying the Evolution of Biological Organisation in Evo-Devo, Evo-Eco and Evolutionary Transitions. Evol Biol, 2016. 43(4): p. 553-581.

148.    Power, D.A., et al., What can ecosystems learn? Expanding evolutionary ecology with learning theory. Biol Direct, 2015. 10: p. 69.

149.    Fields, C. and M. Levin, Does Evolution Have a Target Morphology? Organisms. Journal of Biological Sciences, 2020. 4(1): p. 57-76.

150.    Shreesha, L. and M. Levin, Stress sharing as cognitive glue for collective intelligences: A computational model of stress as a coordinator for morphogenesis. Biochem Biophys Res Commun, 2024. 731: p. 150396.

151.    Koshland, D.E., The Bacterium as a Model Neuron. Trends in Neurosciences, 1983. 6(4): p. 133-137.

152.    Liu, J., et al., Coupling between distant biofilms and emergence of nutrient time-sharing. Science, 2017. 356(6338): p. 638-642.

153.    Lee, D.D., et al., SnapShot: Electrochemical Communication in Biofilms. Cell, 2017. 170(1): p. 214-214 e1.

154.    Levin, M., G. Pezzulo, and J.M. Finkelstein, Endogenous Bioelectric Signaling Networks: Exploiting Voltage Gradients for Control of Growth and Form. Annu Rev Biomed Eng, 2017. 19: p. 353-387.

155.    Woodruff, R.I. and W.H. Telfer, Polarized intercellular bridges in ovarian follicles of the cecropia moth. J Cell Biol, 1973. 58(1): p. 172-88.

156.    Petsakou, A. and N. Perrimon, Bioelectric regulation of intestinal stem cells. Trends Cell Biol, 2023. 33(7): p. 555-567.

157.    Schotthofer, S.K. and J. Bohrmann, Analysing bioelectrical phenomena in the Drosophila ovary with genetic tools: tissue-specific expression of sensors for membrane potential and intracellular pH, and RNAi-knockdown of mechanisms involved in ion exchange. BMC Dev Biol, 2020. 20(1): p. 15.

158.    Weiss, I. and J. Bohrmann, Electrochemical gradients are involved in regulating cytoskeletal patterns during epithelial morphogenesis in the Drosophila ovary. BMC Dev Biol, 2019. 19(1): p. 22.

159.    Emmons-Bell, M. and I.K. Hariharan, Membrane potential regulates Hedgehog signalling in the Drosophila wing imaginal disc. EMBO Rep, 2021. 22(4): p. e51861.

160.    Emmons-Bell, M., R. Yasutomi, and I.K. Hariharan, Membrane potential regulates Hedgehog signaling and compartment boundary maintenance in the Drosophila wing disc. bioRxiv, 2020: p. 2020.06.02.128892.

161.    Vandenberg, L.N., R.D. Morrie, and D.S. Adams, V-ATPase-dependent ectodermal voltage and pH regionalization are required for craniofacial morphogenesis. Dev Dyn, 2011. 240(8): p. 1889-904.

162.    Lu, H.Y., et al., Multi-scale neural decoding and analysis. J Neural Eng, 2021. 18(4).

163.    Huth, A.G., et al., Decoding the Semantic Content of Natural Movies from Human Brain Activity. Front Syst Neurosci, 2016. 10: p. 81.

164.    Nishimoto, S., et al., Reconstructing visual experiences from brain activity evoked by natural movies. Current biology : CB, 2011. 21(19): p. 1641-6.

165.    Naselaris, T., et al., Bayesian reconstruction of natural images from human brain activity. Neuron, 2009. 63(6): p. 902-15.

166.    McMillen, P. and M. Levin, Optical Estimation of Bioelectric Patterns in Living Embryos. Methods Mol Biol, 2024. 2745: p. 91-102.

167.    Adams, D.S. and M. Levin, General principles for measuring resting membrane potential and ion concentration using fluorescent bioelectricity reporters. Cold Spring Harbor protocols, 2012. 2012(4): p. 385-97.

168.    Fields, C., J. Bischof, and M. Levin, Morphological Coordination: A Common Ancestral Function Unifying Neural and Non-Neural Signaling. Physiology, 2020. 35(1): p. 16-30.

169.    Karig, D., et al., Stochastic Turing patterns in a synthetic bacterial population. Proc Natl Acad Sci U S A, 2018. 115(26): p. 6572-6577.

170.    Marcon, L. and J. Sharpe, Turing patterns in development: what about the horse part? Current opinion in genetics & development, 2012. 22(6): p. 578-84.

171.    Nakamasu, A., et al., Interactions between zebrafish pigment cells responsible for the generation of Turing patterns. Proceedings of the National Academy of Sciences of the United States of America, 2009. 106(21): p. 8429-34.

172.    Strier, D.E. and S. Ponce Dawson, Turing patterns inside cells. PloS one, 2007. 2(10): p. e1053.

173.    Pai, V.P., et al., Transmembrane voltage potential controls embryonic eye patterning in Xenopus laevis. Development, 2012. 139(2): p. 313-23.

174.    Durant, F., et al., The Role of Early Bioelectric Signals in the Regeneration of Planarian Anterior/Posterior Polarity. Biophys J, 2019. 116(5): p. 948-961.

175.    Krutovskikh, V. and H. Yamasaki, The role of gap junctional intercellular communication (GJIC) disorders in experimental and human carcinogenesis. Histology and Histopathology, 1997. 12: p. 761-768.

176.    Mesnil, M., et al., Role of blocked gap junctional intercellular communication in non-genotoxic carcinogenesis. Toxicology Letters, 1995. 82-83: p. 701-6.

177.    Davies, P.C.W. and C.H. Lineweaver, Cancer tumors as Metazoa 1.0: tapping genes of ancient ancestors. Physical Biology, 2011. 8(1): p. 015001.

178.    Bussey, K.J., et al., Ancestral gene regulatory networks drive cancer. Proceedings of the National Academy of Sciences of the United States of America, 2017. 114(24): p. 6160-6162.

179.    Chernet, B.T., et al., Use of genetically encoded, light-gated ion translocators to control tumorigenesis. Oncotarget, 2016. 7(15): p. 19575-88.

180.    Chernet, B.T., C. Fields, and M. Levin, Long-range gap junctional signaling controls oncogene-mediated tumorigenesis in Xenopus laevis embryos. Frontiers in physiology, 2015. 5: p. 519.

181.    Chernet, B.T. and M. Levin, Transmembrane voltage potential of somatic cells controls oncogene-mediated tumorigenesis at long-range. Oncotarget, 2014. 5(10): p. 3287-306.

182.    Chernet, B.T. and M. Levin, Endogenous Voltage Potentials and the Microenvironment: Bioelectric Signals that Reveal, Induce and Normalize Cancer. Journal of clinical & experimental oncology, 2013. Suppl 1.

183.    Chernet, B.T. and M. Levin, Transmembrane voltage potential is an essential cellular parameter for the detection and control of tumor development in a Xenopus model. Disease models & mechanisms, 2013. 6(3): p. 595-607.

184.    Blackiston, D.J., K.A. McLaughlin, and M. Levin, Bioelectric controls of cell proliferation: ion channels, membrane voltage and the cell cycle. Cell Cycle, 2009. 8(21): p. 3519-28.

185.    Pai, V.P. and M. Levin, HCN2 channel-induced rescue of brain, eye, heart and gut teratogenesis caused by nicotine, ethanol and aberrant notch signalling. Wound Repair Regen, 2022.

186.    Pai, V.P., et al., HCN2 Channel-Induced Rescue of Brain Teratogenesis via Local and Long-Range Bioelectric Repair. Frontiers in Cellular Neuroscience, 2020. 14(136): p. 136.

187.    Pai, V.P., et al., HCN2 Rescues brain defects by enforcing endogenous voltage pre-patterns. Nature Communications, 2018. 9(1): p. 998.

188.    Doursat, R. and C. Sanchez, Growing fine-grained multicellular robots. Soft Robotics, 2014. 1(2): p. 110-121.

189.    Doursat, R., H. Sayama, and O. Michel, A review of morphogenetic engineering. Natural Computing, 2013. 12(4): p. 517-535.

190.    Sample, M., et al., Multi-cellular engineered living systems: building a community around responsible research on emergence. Biofabrication, 2019. 11(4): p. 043001.

191.    Kamm, R.D., et al., Perspective: The promise of multi-cellular engineered living systems. Apl Bioengineering, 2018. 2(4): p. 040901.

192.    Ebrahimkhani, M.R. and M. Levin, Synthetic living machines: A new window on life. iScience, 2021. 24(5): p. 102505.

193.    Turner, J.S., Termites as models of swarm cognition. Swarm Intelligence, 2011. 5(1): p. 19-43.

194.    Khuong, A., et al., Stigmergic construction and topochemical information shape ant nest architecture. Proceedings of the National Academy of Sciences of the United States of America, 2016. 113(5): p. 1303-1308.

195.    Gottlieb, G., Probabilistic epigenesis and evolution. 1999 Heinz Werner lecture series. 1999, Worcester, Mass.: Clark University Press. vii, 97 p.

196.    Nanos, V. and M. Levin, Multi-scale Chimerism: An experimental window on the algorithms of anatomical control. Cells Dev, 2021. 169: p. 203764.

197.    Pezzulo, G. and M. Levin, Re-membering the body: applications of computational neuroscience to the top-down control of regeneration of limbs and other complex organs. Integr Biol (Camb), 2015. 7(12): p. 1487-517.

198.    Wolpert, L., Do we understand development? Science, 1994. 266(5185): p. 571-2.

199.    Vogt, N., Assembloids. Nat Methods, 2021. 18(1): p. 27.

200.    Panoutsopoulos, A.A., Organoids, Assembloids, and Novel Biotechnology: Steps Forward in Developmental and Disease-Related Neuroscience. Neuroscientist, 2020: p. 1073858420960112.

201.    Haase, K. and B.S. Freedman, Once upon a dish: engineering multicellular systems. Development, 2020. 147(9).

202.    Blackiston, D., et al., Biological Robots: Perspectives on an Emerging Interdisciplinary Field. Soft Robot, 2023.

203.    Kriegman, S., et al., Kinematic self-replication in reconfigurable organisms. Proc Natl Acad Sci U S A, 2021. 118(49).

204.    Blackiston, D., et al., A cellular platform for the development of synthetic living machines. Science Robotics, 2021. 6(52): p. eabf1571.

205.    Kriegman, S., et al., A scalable pipeline for designing reconfigurable organisms. Proc Natl Acad Sci U S A, 2020. 117(4): p. 1853-1859.

206.    Pai, V.P., et al., Basal Xenobot transcriptomics reveals changes and novel control modality in cells freed from organismal influence. Commun Biol, 2025. 8(1): p. 646.

207.    Gumuskaya, G., et al., The Morphological, Behavioral, and Transcriptomic Life Cycle of Anthrobots. Adv Sci (Weinh), 2025: p. e2409330.

208.    Gumuskaya, G., et al., Motile Living Biobots Self-Construct from Adult Human Somatic Progenitor Seed Cells. Adv Sci (Weinh), 2023: p. e2303575.

209.    Bhattaram, D., et al., AggreBots: Configuring CiliaBots through guided, modular tissue aggregation. Sci Adv, 2025. 11(39): p. eadx4176.

210.    Fields, C. and M. Levin, Thoughts and thinkers: On the complementarity between objects and processes. Physics of Life Reviews, 2025.

211.    Lagasse, E. and M. Levin, Future medicine: from molecular pathways to the collective intelligence of the body. Trends Mol Med, 2023.

212.    Keresztes, D., et al., Cancer drug resistance as learning of signaling networks. Biomed Pharmacother, 2025. 183: p. 117880.

213.    Veres, T., et al., Cellular forgetting, desensitisation, stress and ageing in signalling networks. When do cells refuse to learn more? Cell Mol Life Sci, 2024. 81(1): p. 97.

214.    Murugan, N.J., et al., Acute multidrug delivery via a wearable bioreactor facilitates long-term limb regeneration and functional recovery in adult Xenopus laevis. Sci Adv, 2022. 8(4): p. eabj2164.

215.    Langton, C.G., Artificial life : an overview. Complex adaptive systems. 1995, Cambridge, Mass.: MIT Press. xi, 340 , [6] of plates.

216.    McNerney, M.P., et al., Theranostic cells: emerging clinical applications of synthetic biology. Nat Rev Genet, 2021. 22(11): p. 730-746.

217.    Krzyszton, R., et al., Gene-circuit therapy on the horizon: synthetic biology tools for engineered therapeutics. Acta Biochim Pol, 2021. 68(3): p. 377-383.

218.    Sole, R. and J. Macia, Synthetic biology: Biocircuits in synchrony. Nature, 2014. 508(7496): p. 326-7.

219.    Pagan-Diaz, G.J., et al., Engineering geometrical 3-dimensional untethered in vitro neural tissue mimic. Proc Natl Acad Sci U S A, 2019. 116(51): p. 25932-25940.

220.    Aydin, O., et al., Neuromuscular actuation of biohybrid motile bots. Proc Natl Acad Sci U S A, 2019. 116(40): p. 19841-19847.

221.    Ho, C. and L. Morsut, Novel synthetic biology approaches for developmental systems. Stem Cell Reports, 2021. 16(5): p. 1051-1064.

222.    Velazquez, J.J., et al., Programming Morphogenesis through Systems and Synthetic Biology. Trends Biotechnol, 2018. 36(4): p. 415-429.

223.    Davies, J., Using synthetic biology to explore principles of development. Development, 2017. 144(7): p. 1146-1158.

224.    Rouleau, N. and M. Levin, The Multiple Realizability of Sentience in Living Systems and Beyond. eNeuro, 2023. 10(11).

225.    Abramson, C.I. and M. Levin, Behaviorist approaches to investigating memory and learning: A primer for synthetic biology and bioengineering. Commun Integr Biol, 2021. 14(1): p. 230-247.

226.    Selberg, J., M. Gomez, and M. Rolandi, The Potential for Convergence between Synthetic Biology and Bioelectronics. Cell Syst, 2018. 7(3): p. 231-244.

227.    Clawson, W.P. and M. Levin, Endless forms most beautiful 2.0: teleonomy and the bioengineering of chimaeric and synthetic organisms. Biological Journal of the Linnean Society, 2023. 139: p. 457–486.

228.    Ando, N. and R. Kanzaki, Insect-machine hybrid robot. Curr Opin Insect Sci, 2020. 42: p. 61-69.

229.    Tsuda, S., S. Artmann, and K.-P. Zauner, The Phi-Bot: A Robot Controlled by a Slime Mould, in Artificial Life Models in Hardware, A. Adamatzky and M. Komosinski, Editors. 2009, Springer London. p. 213-232.

230.    Bakkum, D.J., et al., MEART: The Semi-Living Artist. Frontiers in neurorobotics, 2007. 1: p. 5.

231.    Potter, A.M., et al. Hybrots: hybrids of living neurons and robots for studying neural computation. in Proceedings of brain inspired cognitive systems. 2004.

232.    Potter, S.M., et al., Long-term bidirectional neuron interfaces for robotic control, and in vitro learning studies. Proceedings of the 25th Annual International Conference of the Ieee Engineering in Medicine and Biology Society, Vols 1-4, 2003. 25: p. 3690-3693.

233.    Warwick, K., Nasuto, S. J., Becerra, V. M., Whalley, B. J., Experiments with an In-Vitro Robot Brain, in Computing with Instinct, Y. Cai, Editor. 1998, Springer.

234.    Grimaldi, C. and G.W. Marcy, Bayesian approach to SETI. Proc Natl Acad Sci U S A, 2018. 115(42): p. E9755-E9764.

235.    Maccone, C., Evo-SETI: A Mathematical Tool for Cladistics, Evolution, and SETI. Life (Basel), 2017. 7(2).

236.    Waldrop, M.M., The search for alien intelligence: SETI is dead–long live SETI. Nature, 2011. 475(7357): p. 442-4.

237.    Maccone, C., A mathematical model for evolution and SETI. Orig Life Evol Biosph, 2011. 41(6): p. 609-19.

238.    Villoutreix, P., What machine learning can do for developmental biology. Development, 2021. 148(1).

239.    Rijssenbeek, J., Synthetic biology: supporting an anti-reductionist view of life. Synthese, 2025. 205(2): p. 52.

240.    Boldt, J., Machine metaphors and ethics in synthetic biology. Life Sci Soc Policy, 2018. 14(1): p. 12.

241.    Rouleau, N. and M. Levin, Discussions of machine versus living intelligence need more clarity. Nature Machine Intelligence, 2024. 6(12): p. 1424-1426.

242.    Levin, M., AI: a Bridge toward Diverse Intelligence and Humanity’s Future. preprint, 2024.

243.    McShea, D.W., Machine wanting. Stud Hist Philos Biol Biomed Sci, 2013. 44(4 Pt B): p. 679-87.

244.    McShea, D.W., Upper-directed systems: a new approach to teleology in biology. Biology & Philosophy, 2012. 27(5): p. 663-684.

245.    Barwich, A.-S. and M.J. Rodriguez, Rage against the what? The machine metaphor in biology. Biology & Philosophy, 2024. 39(4): p. 14.

246.    Nicholson, D.J., Is the cell really a machine? J Theor Biol, 2019. 477: p. 108-126.

247.    Pio-Lopez, L., The rise of the biocyborg: synthetic biology, artificial chimerism and human enhancement. New Genetics and Society, 2021. 40(4): p. 599-619.

248.    Levin, M., Ingressing Minds: Causal Patterns Beyond Genetics and Environment in Natural, Synthetic, and Hybrid Embodiments. preprint, 2025.

249.    Huang, S., I. Ernberg, and S. Kauffman, Cancer attractors: a systems view of tumors from a gene network dynamics and developmental perspective. Semin Cell Dev Biol, 2009. 20(7): p. 869-76.

250.    Marder, E. and J.M. Goaillard, Variability, compensation and homeostasis in neuron and network function. Nature reviews. Neuroscience, 2006. 7(7): p. 563-74.

251.    Cervera, J., M. Levin, and S. Mafe, Morphology changes induced by intercellular gap junction blocking: A reaction-diffusion mechanism. Biosystems, 2021. 209: p. 104511.

252.    Lutz, H., Sur la production experimentale de la polyembryonie et de la monstruosite double chez les oiseaux. Archs Anat. microsc. morph. exp., 1949. 38: p. 79-144.

253.    McClelland, J.L., et al., Letting structure emerge: connectionist and dynamical systems approaches to cognition. Trends Cogn Sci, 2010. 14(8): p. 348-56.

254.    McClelland, J.L. and T.T. Rogers, The parallel distributed processing approach to semantic cognition. Nat Rev Neurosci, 2003. 4(4): p. 310-22.

255.    Grossberg, S., Communication, Memory, and Development, in Progress in Theoretical Biology, R. Rosen and F. Snell, Editors. 1978.

256.    Berry, M.J., 2nd, et al., Anticipation of moving stimuli by the retina. Nature, 1999. 398(6725): p. 334-8.

257.    Marr, D., Vision : a computational investigation into the human representation and processing of visual information. 1982, San Francisco: W.H. Freeman. xvii, 397 p.

258.    Melanitis, N. and K.S. Nikita, Biologically-inspired image processing in computational retina models. Comput Biol Med, 2019. 113: p. 103399.

259.    Beane, W.S., et al., A Chemical genetics approach reveals H,K-ATPase-mediated membrane voltage is required for planarian head regeneration. Chemistry & Biology, 2011. 18(1): p. 77-89.

260.    Blackiston, D.J., K. Vien, and M. Levin, Serotonergic stimulation induces nerve growth and promotes visual learning via posterior eye grafts in a vertebrate model of induced sensory plasticity. npj Regenerative Medicine, 2017. 2(1): p. 8.

261.    Blackiston, D., et al., A second-generation device for automated training and quantitative behavior analyses of molecularly-tractable model organisms. PLoS One, 2010. 5(12): p. e14370.

262.    Zahn, N., M. Levin, and D.S. Adams, The Zahn drawings: new illustrations of Xenopus embryo and tadpole stages for studies of craniofacial development. Development, 2017. 144(15): p. 2708-2713.

263.    Lobo, D., W.S. Beane, and M. Levin, Modeling planarian regeneration: a primer for reverse-engineering the worm. PLoS computational biology, 2012. 8(4): p. e1002481.

264.    Colloca, L. and F. Benedetti, How prior experience shapes placebo analgesia. Pain, 2006. 124(1-2): p. 126-33.

265.    Colloca, L. and F. Benedetti, Placebos and painkillers: is mind as real as matter? Nat Rev Neurosci, 2005. 6(7): p. 545-52.

266.    Lui, F., et al., Neural bases of conditioned placebo analgesia. Pain, 2010. 151(3): p. 816-824.

267.    Black, D.S., G. Christodoulou, and S. Cole, Mindfulness meditation and gene expression: a hypothesis-generating framework. Curr Opin Psychol, 2019. 28: p. 302-306.

268.    Dada, T., et al., Mindfulness Meditation Reduces Intraocular Pressure, Lowers Stress Biomarkers and Modulates Gene Expression in Glaucoma: A Randomized Controlled Trial. J Glaucoma, 2018. 27(12): p. 1061-1067.

269.    Buric, I., et al., What Is the Molecular Signature of Mind-Body Interventions? A Systematic Review of Gene Expression Changes Induced by Meditation and Related Practices. Front Immunol, 2017. 8: p. 670.