The Multiscale Wisdom of the Body: Collective Intelligence as a Tractable Interface for Next-Generation Biomedicine

Could some of these patterns within the cellular collective intelligence be addressed in the same way as harmful, persistent thoughts are treated within the neural cognitive system of patients [53, 228, 229, 230]? Some diseases may be due to perceptual illusions, sensory or attention deficits, mistaken beliefs, excessive insecurity, or harmful self‐models formed in tissue as a result of prior experiences and could be addressed by the powerful emerging tools of computational psychiatry and cognitive behavioral therapies, but aimed instead at the non‐neural intelligence in tissues and implemented in physiological and transcriptional spaces.

Conversely, not all disease may be caused by the addition of unwanted information patterns; we have previously proposed that aging is due to the degradation of important endogenous (bioelectric) pattern memories that are required throughout the lifespan to maintain tissue order [146, 147]. It is known that removing instructive signals, for example by denervation, can cause disorganization of mature tissues such as tongue papillae and, in general, render tissues much more susceptible to disorders of morphostasis such as cancer [231, 232, 233, 234]. Morphogenesis does not end when the body is complete—the Ship of Theseus is a good analogy for the body, consisting of the replacement policies in the somatic intelligence of the body that needs to make context‐sensitive repairs to a tight target specification. It is likely that solutions to the aging problem need to not only include rejuvenating signals at the cell level (e.g., Yamanaka factors) but also target the information scaffold needed to know what the new cells should do and where they should do it [235]. Some of that information could be imposed directly (by optogenetic or pharmacological stimuli), but some of it may be best implemented by facilitating the communication between the right kind of modules—letting cells talk to other cells, biobots, or next‐generation living “bandages.” In the case of cancer, 1st‐order treatments would be bioelectrical stimuli that force a hyperpolarized state [175, 181]; 2nd‐order treatments would facilitate gap junctional connections among cells or exposure to active morphogenetic cues (including non‐bioelectric ones) that are known to induce normalization [236, 237, 238].

Third‐order treatments may target the stress perception machinery in cells [239] to counteract the eventual shrinkage of the border between self and world that occurs in agents with continuous exposure to danger signals in their social milieu [150]. We can also go well beyond biomedicine in vivo, applying these same principles to bioengineering efforts to recreate organs and other needed living structures for transplantation or other purposes.

Complementing the efforts to influence higher levels of biology (writing information into the system), the field must develop methods to query the insights that these higher levels have about their own function (reading information, at levels across the neuroscience spectrum ranging from physiological recording, behavioral analysis, and conversation).

Exploiting the problem‐solving capacities of cells and tissues is essential to unlock the promise of other, conventional technologies. For example, even when CRISPR becomes 100% reliable and specific for single‐gene diseases, genomic editing will still need to navigate a further barrier in order to develop applications for the control of growth and form: which genes must be edited to achieve a desired complex anatomical change? Planaria rapidly discern which of their tens of thousands of genes hold the answer to a novel stressor (barium) [66]. While it is not yet known how they do it, there exists a native mechanism for identifying the molecular affordances that can be activated to solve even unforeseen stressors [66]. Likewise, salamander cells call upon the right cell‐biological mechanism to make a kidney tubule out of many or just one cell [71, 100]. It should be possible to use imaging and computational interpretation filters to get cells (in situ or in bioengineered avatars) to tell us what steps they would take for a given situation and use it to guide gene therapy and pharmacological interventions. The technology that must be developed for this is simply the neural decoding [240, 241, 242, 243] system applied to non‐neural tissues.

More broadly, tools (such as AI applied to non‐invasive, multi‐modal physiological profiling across scales, Figure 6D) can provide a communications channel to biological sensors. In top‐down diagnostics, we might not try to read and interpret the status of specific markers but instead ask cells and tissues their perception of their neighbors and overall conditions. Not just surrogate site diagnostics at a distance [122], but using living components as the final layer of a classifier neural network to help interpret complex biological states as inputs. We could track stress levels and other aspects of behavior of cells, organoids, and biobots as they are exposed to patient tissues, to benefit from their built‐in ability to coarse‐grain and react to myriad physiological and biochemical parameters of their microenvironment. The combination of language models, living information filters, and physiomic profiling raises an intriguing possibility. If molecular states can be controlled through the linguistic interface in psychodermatology [244, 245, 246], perhaps similar techniques can be used to extract insight into health and disease processes. Spontaneous cases of novel, undiagnosed disease states being brought to clinician's attention via language [247], suggest that it may be possible to create tools that penetrate the normally tight virtualization (abstraction layers) used in biology and communicate bi‐directionally with organs, tissues, and cells.

A fundamental question about information in biology, especially setpoints, is “where is it encoded?” Physically, bodies consist of a functional hierarchy in which genes and the laws of chemistry determine the properties of molecular interaction, which in turn, with the laws of computation, determine the resulting behavior of cells, tissues, and organs as they self‐construct and repair toward complex anatomical outcomes. Philosophically, one can hold that the final result is encoded anywhere along that range of scales—from the basic laws of physics underlying everything that happens in the universe to the pattern memories that finally indicate a specific shape for a given organ. Practically, what matters is the distance between a given level of organization and the actionable information specifying outcomes one seeks to control. The distance between dynamic anatomy and the genetic sequence of proteins is enormous, due to the many active processes of morphogenesis that lie between them. The distance between the prepatterns (encoded for example in bioelectric states) and the anatomy is smaller. Crossing levels is very difficult, which is why we typically do not program our computers by tweaking the properties of silicon and copper. It is critical to identify the most proximal layer of description to the phenotypes in question and discover a set of tools to optimally manipulate the system at that level.

The major hypothesis discussed herein is a fundamental symmetry between mind and body, in terms of mechanisms and the functional causal architecture they implement. The implication of this perspective is that tools from behavioral neuroscience—ranging from electrophysiology to psychiatry—could be applicable outside the brain and its control of conventional behavior. Some of these tools have already been used to implement novel capabilities to control morphology in birth defects, regeneration, and cancer. Others remain speculative proposals whose value will be tested in forthcoming experimental approaches.

The key issue in any application in biomedicine, bioengineering, and physiological health/disease is: what is the optimal level of interaction (Figure 1A)? Training animals is more efficient than micromanaging nerves and muscles because it exploits the native top‐down control system that the body uses to manage low‐level molecular and cellular events toward whole‐body adaptive goals. Learning writes information into the medium much more efficiently than our clumsy interventions, and it behooves us to understand which layers offer what affordances to read and write information toward desired outcomes. In some cases, the most efficient targets will be molecular components. In others, it will be the higher‐order memories, goals, and self‐models of the cells and tissues. A crucial component of this strategy is an organicist perspective that treats the biology not as a simple machine and not just as a source of emergent complexity but as a multi‐scale society of agents with agendas. In psychiatry, the best predictor of success is the belief‐congruent approach in which the patient respects and feels aligned with the therapist. The somatic version of this is interventions that are perceived by cells and tissues as something they wanted to do in the first place—not micromanagement of symptoms but instead targeting deep control structures that represent the system's own goal states and memories, resulting in a “therapeutic alliance” [248]. The path to effective regenerative medicine is to take seriously the teleonomy that pervades living systems and harness it.

Albert Mason developed a remarkable practice of hypnodermatology [244, 249]—using commands filtered through the language interface to modify cellular behavior (now known as mind‐body medicine [250]). He eventually changed course and became a psychotherapist because he noted that his patients’ skin conditions would clear up, but they would develop problems elsewhere in their lives because, albeit at a higher level, he was still treating the manifestation of disease, not the underlying cause. Perhaps a similar path will play out in this field. Rewriting bioelectric patterns [116] and exploiting drug conditioning [54, 251, 252, 253] are higher‐level interventions than direct modulation of transcription factors and signaling proteins, but it may still be just a temporary crutch for an even higher‐level semantic interface. Top‐level executive mental constructs can cause ion flows in muscle (voluntary motion in the pursuit of career goals, for example). If bioelectricity transduces mental content into changes in biochemical events, and bioelectrical signals can induce regeneration [113, 254] and cancer normalization [148], is it possible that eventually, we will be able to go directly from mental states to anatomical outcomes, skipping the intermediate components described above? This, of course, has been suggested by alternative health practitioners for a very long time, in terms of their emphasis on the power of the mind for healing. But moving beyond theoretical claims toward reliable universal regenerative effects requires a lot of rigorous research to flesh out the connection across therapeutic levels via available technologies.

Future medicine may look a lot more like psychiatry than it does like chemistry, because in the end, the real mind‐body medicine may have to target the many minds operating within the body, not just the mind of the “patient” that normally commandeers our attention. At stake are truly transformative applications ranging from repair of birth defects and injury to cancer reprogramming, bioengineered organs, and the freedom of embodiment offered by effective rational control over growth and form.

My laboratory has sponsored research agreements with Morphoceuticals and Astonishing Labs, two companies that operate in a space relevant to this paper.