Collective intelligence: A unifying concept for integrating biology across scales and substrates

A defining feature of biology is the use of a multiscale architecture, ranging from molecular networks to cells, tissues, organs, whole bodies, and swarms. Crucially however, biology is not only nested structurally, but also functionally: each level is able to solve problems in distinct problem spaces, such as physiological, morphological, and behavioral state space. Percolating adaptive functionality from one level of competent subunits to a higher functional level of organization requires collective dynamics: multiple components must work together to achieve specific outcomes. Here we overview a number of biological examples at different scales which highlight the ability of cellular material to make decisions that implement cooperation toward specific homeodynamic endpoints, and implement collective intelligence by solving problems at the cell, tissue, and whole-organism levels. We explore the hypothesis that collective intelligence is not only the province of groups of animals, and that an important symmetry exists between the behavioral science of swarms and the competencies of cells and other biological systems at different scales. We then briefly outline the implications of this approach, and the possible impact of tools from the field of diverse intelligence for regenerative medicine and synthetic bioengineering.

One defining feature of complex life, making it distinct from our current engineered artifacts, is its multiscale nature: there is order in biology across levels of organization, from molecules to cells, tissues, organs, whole organisms, and societies/swarms1,2. Crucially, however, this goes well beyond structural nesting: it is in fact a multiscale competency architecture3,4 because each level solves problems in its own relevant domains (Fig. 1). As evolution facilitated the increase of complexity, living things became composed of layers that cooperate and compete to solve problems in metabolic, physiological, anatomical, and behavioral state spaces (reviewed in refs. 5,6). Biology’s robustness, open-endedness, evolvability, and unique complexity likely depend on the fact that evolution works with an agential material – a substrate with competencies, computational abilities, and homeodynamic setpoints5,7 that strongly influence the structure and function of multicellular forms. Adaptive behavior in new problem spaces3,4 can arise because higher levels of organization can deform the energy landscape for the subunits8, while benefitting from their ability to navigate those landscapes autonomously and without micromanagement.

Understanding how the behavior of subunits percolates up toward adaptive processes at higher levels (Fig. 1a–e), and how higher levels of organization constrain and facilitate the behavior of their parts9–15, is critical not only to basic evolutionary biology but also to the control of system-level outcomes in biomedicine16,17 and to the design of novel engineered systems18–24. We have previously proposed that this research program can be advanced by exploiting collective intelligence as a crucial symmetry across levels, which enables the tools of behavioral science to be brought to bear on novel unconventional substrates16,17,25,26, especially the capabilities of cell groups in transcriptional, physiological, and anatomical spaces (Fig. 1c, e). Specifically, we have argued that regulative morphogenesis is a kind of behavior of cellular collectives traversing anatomical morphospace (Fig. 1f, g)27–32, and others have argued that immune systems33,34, bacterial biofilms35–38, and many other unconventional substrates39–42 can be effectively understood and rationally controlled by using techniques from behavioral and cognitive science43.

Here, we explore a number of phenomena in biology which illustrate this approach, specifically focusing on two aspects that feature prominently in behavioral science. One is intelligence, in William James’44 sense of a degree of ability to reach the same goal by different means (i.e., problem-solving in changing or novel circumstances). The other is collective decision-making, as studied in the rapidly advancing study of group behavior among swarms45–50. This parallel has been explored previously51–54, and we extend those ideas here with specific references to more recent data revealed by advances in non-invasive imaging and functional cell modulation technology. We emphasize organizational principles that enable not just emergent complexity, but adaptive proto-cognitive systems (problem-solving with respect to adaptive goals and novel circumstances) to appear3,55. A central claim of the emerging field of diverse intelligence is that cognitive capacities (Box. 1) exist on a spectrum: that tools, concepts, and approaches from behavioral sciences can be productively applied to understand and control systems far beyond familiar animals with central nervous systems (without the necessity to attribute advanced, human-level metacognitive traits). We extend James’ definition of intelligence to collectives by considering the perceptual field of an agent: the area in space and time that the agent can survey to find alternative paths to a goal (Fig. 2a, b). As the size of a collective increases its perceptual field increases, improving its ability to find variant paths. (Fig. 2c). It should be noted that while we here focus on animal development, there are also fascinating data of this kind in plants53,56–61.

The most familiar examples of collective/swarm intelligence are beehives, ant and termite colonies62,63, and flocks of birds and fish64–66. However, it is important to keep in mind that there is no sharp distinction between these collective minds46 and putative centralized ones like those possessed by complex animals and ourselves67 – instead, the biosphere offers a spectrum of architectures including familiar solid brains where the neurons do not move much relative to each other (except in exceptional cases such as metamorphosis68) and so-called “liquid brains” – constructs in which the subunits can implement fluid interactions63,69. Fundamentally, typical brains are a collective of neurons, and provide an experience and functional unity of memories, goals, and preferences because of their interaction dynamics. Thus, one way to view cognitive science is as the study of the collective Intelligence of neurons and other cell types. Understanding how collectives ensure cooperation toward adaptive ends in diverse problem spaces is as much a part of understanding ourselves as of understanding ant colonies. Indeed it has been hypothesized that the remarkable ability of neurons to unify toward a centralized self – the emergent agent that is the subject of memories, preferences, and goals which are not assignable to any of the individual components – is an evolutionary pivot of far earlier cell communication strategies that first solved problems in navigating another domain that requires information processing above the single cell level: anatomical morphospace70. By exploring possible scale-free dynamics in diverse systems, such as viewing the processes of morphogenesis as a kind of behavior of cellular swarms in anatomical space, we may enrich both behavioral neuroscience and developmental/regenerative biology by an influx of new ways of looking at the data28,71,72.

Crucially, despite the clear parallels to the neuroscience of cognition, we here do not make any claims about first-person experience of unconventional collectives73–76, nor are we saying that the phenomena we describe are of the same degree as familiar human-level cognitive capacities. Instead, we aim to take developmental biology and evolution seriously, and investigate the plesiomorphic, necessarily much more minimal, versions of decision-making and other proto-cognitive functions in multicellular contexts such as regeneration, development, and cancer. We view this as an important step to unify the recent progress in studies of single cell perception/action loops77–80 with the work on active inference and perceptual control theory currently being developed in neuroscience, robotics, and artificial life81–87. Specifically, it is essential to move beyond low-level models of information processing, memory, and anticipation in chemical pathways88–90 or in single cells91–94, to understanding the higher-level perceptual landscape of multicellular collectives95,96. This knowledge is essential to improve our ability to explain, control, and re-engineer complex morphological and functional outcomes that today are still outside of our reach25,26. Here, we review interesting examples of the early, simple precursors to the processes that could underlie complex cognitive architectures, and we explore the hypothesis that the porting of tools across disciplines (dissolving artificial barriers between fields) may facilitate further research.

The use of the word 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? This is a crucial question. First, in the modern age, we must accept that all cognitive systems – ourselves included – exhibit chemistry, not magic, when one drills down to examine the lower levels. Thus, there simply is no special human category which one can correctly anthropomorphize as somehow being beyond the laws of physics at its base. We argue that this word is an anachronism and needs to be retired in favor of an empirically-grounded view, updated with the latest findings in causal information theory205–210,233, in which it is perfectly possible (in fact, unavoidable) for a system to both, be subject to chemistry, and also to possess additional levels of description and control whose recognition affords novel benefits. We offer two points in clarifying our use of this terminology (developed in detail in3).

An uncontroversial aspect of our view is that claims of intelligence (and other cognitive terms), like all others, must be based on rigorous experiment, supported or ruled out by the degree of objective benefit that a given framing affords in terms of a) prediction and control, and b) future discovery (and new research programs) it suggests. The latter is most significant, because almost any paradigm can be rescued by enough epicycles; indeed, after one has discovered a new effect or reached a new capability, it is easy to drill down to the chemistry and – looking backwards – claim that there is no intelligence here because it mechanically follows the laws of physics. The same is true for any act of a complex human brain-body system – if one insists on a view from the level of particles, it will always be there. The key question is: does that level of perspective provide the most interesting platform from which to make the next discovery or develop the most effective control policy. The emphasis should be on novel capabilities, and new research programs facilitated (or suppressed) by a given perspective. Thus, we propose that attempts to mine the rich toolbox of behavioral science to understand and exploit capabilities of morphogenetic systems will continue to pay off in many (but no doubt, not all) cases. We have fleshed out the prior gains facilitated by this view, and the promises for regenerative medicine, elsewhere16,17.

The less conventional, and sometimes uncomfortable, aspect of our position is that the empirical utility of framings needs to be applied fearlessly, and followed wherever 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 biofilms35,36,38, plant roots56,57,59–61,234,235, the training of gene-regulatory networks225,226,236,237, or developmental/regenerative biology117, 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.

A most fundamental example of collectivity is observed during embryogenesis (Fig. 3a). When we observe a blastoderm, we call it an embryo. What precisely are we counting when we say it is 1 embryo? One answer is that the system consists of subunits all cooperating toward a specific path in morphospace: the cells are committed to making a specific functional anatomy corresponding to 1 individual. In effect, we are noticing alignment – both physically, in the sense of planar polarity of cell orientation in a collective97–99, and functionally, as is seen in regulative development: if perturbed, the processes of anatomical homeostasis28,100 will attempt to correct and compensate, toward a specific outcome considered normal, across a range of circumstances that includes, but is not limited to, standard development101,102.

Indeed, the question of how many individuals are present in an embryonic blastoderm is not fixed at 1 by the genetics, because temporary introduction of breaks in the blastoderm (leading to informational isolation of islands of cell masses) results in twins, triplets, etc103. (Fig. 3b–d), showing that the blastoderm is a dynamical excitable medium in which multiple coherent embryos can self-organize. The same is true of organogenesis, such as when an induced large eye-field fragments into a number of individual eyes instead of one large eye (Fig. 3e). Thus, the physiological process that leads to the emergence of integrated collectives, which scientists and conspecifics recognize as discrete individuals is fundamentally dependent on the geometry of interactions (and signaling barriers) present during the early establishment of individuality and the setting of borders between Self and outside world (since every cell is some other cell’s adjacent neighbor).

One major question about the origin of higher-level individuals from active components (cells, which are themselves not passive agents9,21), is how the behaviors of the higher levels depend on those of the lower-level components. The most obvious scaling mode is linear: the collective does what its individual cells are doing. But a more interesting aspect is that the collective often displays new behaviors or preferences. One example of this in the same space concerns cell migration. In an electric field, keratocytes migrate to the cathode, but fragments of keratocytes migrate to the anode104. Remarkably, individual fragments have the opposite direction of taxis to that of a collective of those fragments (an intact cell). The behaviors of a collective can, even in relatively minimal systems, be a complex and hard-to-predict function of the tendencies of the components. This is a microcosm of the larger issue of competition between wholes and their parts6,105, and of the more general feature of multiscale organization in which collective agents bend the energy landscape for their components to exploit their mechanisms towards distinct ends.

One failure mode of collective behavior in vivo is cancer106,107. When cells become isolated from the information structure of the tissue, they revert back to an ancient, unicellular transcriptional108 and behavioral phenotype109,110. Exciting work focusing on the biochemical nature of the microenvironment has shown the ability of non-cell-autonomous cues to normalize cancer111–116. However, more recent work has focused on bioelectric cues that normally orchestrate multicellular anatomical outcomes117, and the consequences of their disruption (Fig. 4).

In the tadpole model, it was shown that normal melanocytes could be driven into a melanoma-like converted phenotype: they over-proliferated, migrated inappropriately to regions normally devoid of melanocytes, invaded the blood vessels and brain, and changed shape into a highly arborized, invasive form118. This could be achieved in the absence of classical carcinogens, oncogenes, or DNA damage, by brief exposure to chemical (chloride ion channel activator drug) or molecular-genetic (GlyCl mutant) targeting of the bioelectric state of a specific cell population: instructor cells which normally keep the melanocytes in their healthy state via serotonergic signaling118. Indeed, targeting only a handful of instructor cells in a region away from the source of melanocyte populations (as confirmed by lineage label) was sufficient to turn the whole tadpole into a hyperpigmented phenotype strongly resembling metastatic melanoma: all of the melanocytes converted, even the ones not close to the GlyCl-activated cells.

The most remarkable thing was that this phenotype is an all-or-none phenomenon. Using different reagents could induce different incidences of hyperpigmentation (conversion) in a cohort of animals, but this was a population-level phenotype: for example, 70% of the animals could be converted, but any given animal was either fully converted or fully normal (Fig. 4a, b). A computational model of the known signaling steps was designed and parametrized (Fig. 4c) to reproduce this all-or-none behavior and fit the experimentally-observed incidence percentages across different perturbations119. The model illustrated how cells navigate biochemical state space (Fig. 4c’) and face specific decision-points at regions of that landscape. The benefit of such a model of collective decision-making is that it can be used to infer interventions. Specifically, the model was used to predict an intervention that would break the concordance of melanocytes within single animals. It suggested two drugs and a dominant negative construct – an experiment that had never been done before - which were then experimentally confirmed to produce the first partially-converted animals (Fig. 4d) seen in almost a decade of experiments in this system120.

In addition to the 3D and transcriptional/physiological spaces discussed above, one of the most interesting aspects of collective intelligence is the navigation of anatomical morphospace. Cell groups need to make specific decisions about which organ will be built and what shape they must make. This is a fundamentally different problem than identifying gene regulatory networks and differentiation signals. For example (Fig. 5), planarian cells can rebuild a complete worm from any kind of cut or fragment (Fig. 5a). Typical treatments of this problem focus on a fragment within a morphogen gradient, that offers distinct concentrations of an instructive chemical signal that can confer head/tail fate decisions locally to each wound (Fig. 5b). However, the more interesting and fundamental issue is seen when considering just one cut: the cells on either side of the cut will create a head and tail respectively, but they were adjacent neighbors before the cut and located at the same positional information value. In other words, it is actually impossible for an anatomical decision like this to be made locally – the cells of the wound must coordinate with the remaining fragment to get information about where they are located, which way they are facing, and what other structures exist121,122, in order to make adaptive decisions about large-scale growth and form that enable regeneration of normal worms.

More generally, numerous excellent papers have studied planarian neoblasts and their control networks, as well as the gradients of morphogens that pattern the anterior-posterior, dorso-ventral, and medio-lateral axes123–127. Despite these advances, there is very little understanding of how cells build specific head shapes or how they know when to stop mitosis and morphogenesis when the correct head shape has been achieved. Specifically, for example, no existing model makes a prediction on what will happen if 50% of the neoblasts of a given planarian are replaced with those of a different species and the head is cut off (Fig. 5c). Whether the head will be of the right shape for one of the two species (dominant), or an in-between hybrid form, or in fact continuously cycle between shapes (as each set of neoblasts works to remodel toward the shape they normally make with great fidelity), cannot yet be derived from the properties of single cell regulatory pathways – it is a collective decision about navigating the space of possible head shapes128,129 (Fig. 5d).

Indeed, modification of cell:cell communication during regeneration can cause genetically-normal fragments to produce heads appropriate to other species of planaria128,129 – visiting attractors in morphospace normally reserved for other genetic lineages. More specifically, several perturbations targeting the bioelectric control circuit (Fig. 5e) have shown randomization of outcome: such Cryptic planaria are destabilized, and fragments (even from the same parent worm) will form 1-head and 2-head forms at a set frequency of ~1:2130 (Fig. 5f). This phenomenon highlights collective decision-making because this randomization is at the level of the population: each individual animal has clear heads and tails, not tissue speckled with cells of different identity. In other words, the randomization of bioelectric state131,132 and the downstream morphogen gradients is interpreted with respect to anterior-posterior organ identity by collectives, not by individual cells.

The left-right axis in vertebrates shows a similar phenomenon (Fig. 6). Consistent asymmetries across the midline first show up in the chick embryo around the primitive streak and Hensen’s node133. A number of treatments, including targeting of the bioelectric134–138 or downstream biochemical139–142 pathways, result in randomization of molecular and anatomical consequences of symmetry breakage and orientation143–148. The animals display, in addition to the normal L:R identity, double-right or double-left (isomerism), or reversed (situs inversus) patterns of lateral identity markers, followed by heterotaxy of the heart and viscera. Remarkably, while many of these treatments randomize outcomes, the randomization is once again above the level of the individual: any given embryo has a consistent identity on the L and R side, and all of the cells agree. In all of the many studies on perturbation of the LR pathway, we are aware of only one that actually breaks the concordance: disruption of the planar polarity pathway by down-regulating VANGL signaling149 leads to a speckling on both sides of the midline, consistent with individual cells within a single lateral domain disagreeing on whether they should have L or R identity (Fig. 6a–d).

A number of collective decisions are mediated by bioelectric signaling, which coordinates cells in the body as a likely precursor to its role in coordinating neurons in the brain toward the emergence of a coherent, problem-solving Self27,70. One example of this at the organ level concerns the induction of whole ectopic eyes in the frog embryo by misexpression of ion channels150 whose activity sets up a voltage gradient similar to that of the eye spot which normally determines their location in the head151. As with previous examples, this is a signal to the collective, setting organ-level identity, not micromanaging the differentiation of the many cells which need to be produced and placed with exquisite precision to make a normal vertebrate eye. Interestingly, this signaling has another built-in competency: recruitment. If very few cells are injected with the channel (attached), they will often recruit their neighbors (Fig. 6e, f) to help them complete the task. This is a kind of secondary instruction, where we instruct a group of cells to make an eye, and they recruit the others (which were never directly manipulated), including all of the necessary downstream morphogenetic steps. This recruitment of individuals to accomplish a high-level goal is seen in other collective systems like ant colonies152,153, which often call in helpers when a task is large. The ability to recruit participants to complete tasks may be a central competency of collective intelligence that works across scales, from cells to swarms of entire organisms7.

The neural crest is a cell population that arises between the neural plate and the non-neural ectoderm before migrating throughout the body to produce a constellation of cell types including head mesenchyme, peripheral nervous system and melanocytes154. Neural crest cells (NCCs) must successfully traverse the complex and rapidly changing embryonic body before identifying their target location and integrating with nearby tissue. The energy landscape that they traverse is more complex than it initially appears, however. While each cell navigates a fairly simple dorsal-ventral cartesian space towards a goal destination, the neural crest collective is navigating morphospace to create properly spaced, symmetrical facial structures. When the cell-level navigation of cartesian space is put in opposition to the collective-level navigation of morphospace, the collective supersedes the behavior of the individual to achieve the organism-level morphogenetic target of forming a functional, symmetrical face as we describe in the examples below.

Grafting and ablation experiments underscore the collective ability of the neural crest to accomplish its morphogenetic goals despite some novel circumstances. Axolotls regulate the number of cells, compensating for too few or too many155. Neural crest cells (and in some cases neural tube cells156,157) regulatively adapt their migratory behavior to compensate for the loss of NCCs in nearby or contralateral branchial arches158,159. This re-routed migration suggests that individual cells can leverage the perceptual field of the neural crest cell collective to determine the movements that they should take to contribute to proper system level morphogenesis (Fig. 7). In an especially striking example of the NCCs’ intelligent capability to achieve their ontogenic goals in challenging environments, mouse NCCs grafted into chicken embryos will successfully navigate the forming embryonic face and form teeth160.

Individual cells transposed from one anterior-posterior axial domain to another will change their gene expression to match their neighbors159,161,162. In contrast, groups of cells transposed along the anterior-posterior axis maintain their original gene expression, thus resisting the inductive effects of the surrounding tissue159,161,162. Within the context of our perceptual field model, the increased positional memory and resistance to neighbor effects suggests that cell collectives have an expanded perceptual cone in the posterior time (history) dimension (Fig. 7). While individual cells rapidly lose their memory of past inductive cues, collectives are better able to maintain a consistent identity in a noisy developmental environment159,161,162. As with the re-routing post arch ablation example above, this example suggests that the cell-level behavior is subordinate to the collective-level behavior. When collective-level behaviors are put in opposition by grafting of collectives these is no clear hierarchy, and original fate is maintained.

Another well-studied example of collective intelligence is the vertebrate segmentation clock comprising coordinated oscillations of Notch pathway target genes to establish segmental boundaries of the early vertebrate embryo163. The segmentation clock exhibits functional robustness to interventions, consistent with James’ definition of intelligence, because of the ability to take different paths through morphospace to correctly partition tissue into uniform, correctly sized segments. Clever experiments radically altering the geometry of the tissue test its intelligence by forcing it to explore a variant morphospatial landscape.

In whole embryos, coordinated oscillations will re-emerge following chemically induced disruption and resume producing properly spaced segments164, though the complexity of the developing organism make it challenging to determine if this re-emergence is intrinsic to the segmenting tissue or imposed upon it by other tissues. Recent work with paraxial mesoderm explants165 further emphasizes the remarkable intelligence of this system. In the context of James’ framework for intelligence, the goal states are 1) coordinated oscillatory gene expression and 2) morphological segmentation, and the obstacle is the severe geometric transformation from 3D tube to 2D sheet. In the 2D geometry, oscillatory gene expression waves manifest as outwardly propagating rings that successfully effect segmentation of the outer edge of the explant. While these segments manifest as serially repeated spheres in the 3D in vivo environment, 2D cultures form segments circumscribing the explant’s circumference (Fig. 8a). Segmentation can be re-capitulated from embryonic stem cells in culture by production of trunk-like organoids termed gastruloids166,167, which arrive at a segmented target morphology despite a very different ontogenic history than normal trunk cells.

The ability of the segmentation clock to intelligently navigate its morphospatial landscape has also been tested by forcing it into a state that it would never normally adopt by grafting out of phase cells into oscillating tissue. The segmentation clock functions to coordinate a collection of cells to organize into a large super-cellular structure. Consistent with this function, collectivity is necessary for the segmentation clock to function. When wild type cells are grafted into mutant non-cycling fish, they express the normally oscillatory gene her1, but it does not cycle168. Similarly, pre-somitic mesoderm cells do not oscillate when cultured independently, but will resume oscillating when cultured collectively169. This loss of oscillation can be partially rescued by addition of external FGF, potentially mimicking the effects of high cell density169 and implicating collectivity in stem cell maintenance. Most directly, cells hetero-grafted from tissue in one phase of the clock into a group of cells in a different phase will synchronize to the phase of their lateral neighbors170, (Fig. 8b). In the context of our perceptual field model, the grafted cells benefit from the expanded memory and predictive power of their neighbors to determine their correct position in the clock (Fig. 8b). These cells then adjust their intrinsic oscillatory dynamics to entrain to their neighbors, thus completing their task despite an internal configuration with novel hardware components which do not have an evolutionary history of living together in a single organism.

Though bacteria are unicellular, they often form into large biofilms that exhibit fascinating physiological and morphological collective properties171–174. Interestingly, much as bioelectric networks are used in metazoan systems to bind individual cells together to large-scale morphogenetic projects, bacteria likewise exploit electrical signaling across space and time to coordinate175,176. Cells within the biofilm (and even between biofilms) exhibit bioelectrically-coordinated oscillatory growth patterns that favor the health of the collective at the expense of their own individual fitness177,178, and bioelectric signals coordinate metabolism among distant cells within the biofilm38. These bioelectric signals help recruit new bacteria to the biofilm, even across species179, and can be optogenetically controlled to evoke long-lasting changes on bacterial behavior – a collective memory35.

Exciting recent work has identified a mechanism similar to the vertebrate segmentation clock in bacterial biofilms responding to nitrogen stress mediated by a negative feedback loop180. The similarities between this system and the vertebrate segmentation clock point to further roles for this phenomenon in collective intelligence, and the manifestation of similar molecular logic circuits in distant clades suggests that such collective intelligence is a much more widespread phenomenon than is currently appreciated. Furthermore, the parameter space is neither cartesian space nor morphospace as in our previous examples, but physiological space. The bacterial cells intelligently adjust their individual physiologies to achieve an optimal collective physiology. The capacity of such simple organisms to collectively navigate physiology space using paradigms recapitulated in multicellular organisms highlights the deep cruciality of such navigation to the emergence of complex tissue and points to the necessity of understanding how such navigation occurs during animal development and pathology.

Cell and developmental biology offer very rich fodder for the emerging field of diverse intelligence: discovering a vast spectrum of problem-solving capacities in novel substrates and at unconventional spatiotemporal scales. Because of life’s multi-scale competency architecture, a fundamental aspect of intelligence is collective behavior: all intelligences appear to be made of parts, connected by mechanisms implementing policies that bind the competent components into a cooperative (and competitive6) computational medium that solves problems in new spaces and at higher scales. The harnessing of individual cell behaviors toward regulative morphogenesis (navigating anatomical morphospace), and system-level physiological robustness (traversing physiological space) are especially interesting examples. Indeed, it could be argued that a unique signature of Life is a causal architecture in which the problem-solving competency of the whole is greater than that of its parts). Evolution seems to be particularly good at finding ways to scale the cognitive light cone of cells3–5,181 to achieve spectacular capabilities for gracefully and adaptively handling complexity, novelty, and noise at large scale.

A key aspect of collective intelligence of cell groups is binding subunits’ activities to the same target morphology – a kind of discrete (e.g., head vs. tail) outcome whereas the components have states that range over many continuous quantities. In axial patterning (left-right, anterior-posterior), collective decision-making enables large numbers of cells in a compartment to agree on an organ-scale anatomical fate despite stochastic influences upstream. And it is seen that a decision with respect to morphogenetic outcome, and harnessing cells to the same decision, are orthogonal functions with distinct mechanisms that can be experimentally dissociated.