Biological Robots: Perspectives on an Emerging Interdisciplinary Field
Advances in science and engineering often reveal the limitations of classical approaches initially used to understand, predict, and control phenomena. With progress, conceptual categories must often be re-evaluated to better track recently discovered invariants across disciplines. It is essential to refine frameworks and resolve conflicting boundaries between disciplines such that they better facilitate, not restrict, experimental approaches and capabilities. In this essay, we address specific questions and critiques which have arisen in response to our research program, which lies at the intersection of developmental biology, computer science, and robotics. In the context of biological machines and robots, we explore changes across concepts and previously distinct fields that are driven by recent advances in materials, information, and life sciences. Herein, each author provides their own perspective on the subject, framed by their own disciplinary training. We argue that as with computation, certain aspects of developmental biology and robotics are not tied to specific materials; rather, the consilience of these fields can help to shed light on issues of multiscale control, self-assembly, and relationships between form and function. We hope new fields can emerge as boundaries arising from technological limitations are overcome, furthering practical applications from regenerative medicine to useful synthetic living machines.
Keywords: embryo, developmental biology, robot, xenobot, computer science, synthetic bioengineering, biorobot, animal cap
Multidisciplinary research programs have the potential to generate important advances both within and between established fields; the integration of existing ideas and techniques among researchers with diverse backgrounds often leads to new approaches and new questions. This can be seen in biological robotics, where living materials are being used to build new kinds of robotic devices. Progress in this nascent field will depend on the development of a shared lexicon, as many terms central to biology and robotics lack operational definitions and can have multiple, sometimes incompatible, interpretations when referenced by different fields. For example, what at its essence is a robot? Machine? Organoid? Organism? How these terms are used is contingent upon many factors, including disciplinary training, journal readership, reviewers, classical terminology, and the target audience.
Debates surrounding biological robotics and its nomenclature can be observed in articles, at conferences, in the popular press, and increasingly on social media. Here, we provide individual perspectives on specific questions and critiques raised across disciplines with respect to our recent work on biological robots. This essay is not meant to serve as a review of biological robots, biorobotics, or artificial intelligence (AI)-designed biology at large, but rather to provide our perspectives on specific key issues that have been raised with regard to our research program and related forthcoming technologies. The individual viewpoints of our interdisciplinary team are sometimes nonoverlapping, yet are integrated toward the overarching goal of improving frameworks for future research, and identifying areas in which dichotomous thinking creates artificial boundaries in understanding.
Biorobotics and materials synthetic biology are relatively young disciplines, with both witnessing a surge in progress across the preceding decade.1–6 Approaches, model systems, and goals are myriad, combining elements of bioengineering, stem cell biology, molecular biology, computer science, engineering, neuroscience, and robotics. My work lies at the intersection of these fascinating disciplines, building self-motile biological machines from the ground up from both user and AI-inspired anatomies, engineered for a specific purpose.7–9 While these results have engendered broad support from the scientific community and general public, several pointed critiques have been raised by members of the developmental biology community across social media platforms,10 which are important when framing the work both within the discipline and to those outside the field. As several of these points are valid, the current author offers here their view on the respective topics, which does not necessarily reflect the view of the other co-authors.
The most common point of discussion is that the system is not a robot nor engineered, it is what developmental biologists traditionally define as an animal cap,11–17 a region of developing frog embryo that gives rise to epidermis and neural tissue in the native system. This position has merit, yet indicates a narrow view of the work and fails to acknowledge the diverse biological machines produced by the design pipeline. The initial research study was comparable to several biohybrid robots, which use a combination of synthetic scaffolds and cardiac muscle-based actuation to generate forward locomotion.18–22 Using these same muscle-based actuators, my designs replaced the synthetic scaffolds with living materials, creating a combination of modified animal cap- and muscle-derived tissues.8
In the follow-up work, the team explored the use of motile cilia-based actuation, small hairlike structures naturally present on the surface of tadpole epidermis.7,9 The simplest of these designs are indeed spherical animal caps, a point acknowledged in the article, which were used to quantify metrics of velocity and life span of ciliated tissues and further advance several computational models. However, the author's view is that this approach represents an advance in materials synthetic biology rather than developmental biology. Tissues and cells were leveraged as materials to create a sliding scale of design, from fully muscle actuated to fully cilia actuated, as well as combinations of the two, and from non-shaped spheroids to carefully designed anatomies with multiple tissue types placed at specific locations (Fig. 1).
Related to this first point is the idea that the locomotive behavior of animal cap-derived tissue is a known phenomenon, a position with which I am in agreement. The fact that ciliated tissues generate sufficient flow to create motion has been reported in the literature for decades,23–25 and the current author would not claim to have discovered something new or unexpected in the system. To the contrary, the results indicate that the cilia and muscle actuators behave like windup toys in a predictable and stereotyped manner, a feature which was leveraged with modeling to simulate how individual, and groups, of biological machines will behave in vitro.
It is precisely through manipulation of these known and stereotyped behaviors that we were able to generate kinematic self-replication,9 a fragile process requiring careful titration of several input variables from the investigator. Thus, the innovation comes not from novel developmental insight, but rather using these known systems as living materials, enhancing, or changing movement and behavior through anatomical design by placing actuators at specific locations, or by manually layering tissues into novel configurations. A future goal remains better control of the design process to generate motion and behaviors, which would not be predicted in traditional animal caps.
Similarly, there has been controversy when using the name xenobots to describe the biological machines, as it represents a rebranding of the established animal cap nomenclature. This author avoids the term xenobot as it is a nontechnical descriptor, and one that was not used in the first article, nor the third following community feedback. However, an umbrella term remains necessary to capture the full design space of the research program, which includes cilia and muscle-based actuation, various tissue geometries, sizes, and in the future, cells derived from diverse taxa. Internally, team members disagree regarding the correct vocabulary to describe the system. These terms include: reconfigurable organisms, computer-designed organisms, biobots, xenobots, engineered living systems, biological robots, and biomachines, with the latter being the present author's preferred term as it captures the full range of design space.
Additionally, this term should only be applied to the system where motility is rationally designed or used to complete human specified work, and an unmodified animal cap, or the differentiated mucociliary organoid,24,26–29 should be labeled as such where applicable. Finally, while reconfigurable organism is the specific nomenclature used in the articles, this term is also problematic as biologists currently have no operational definition of the word organism,30,31 with some arguing the term is without technical meaning (the current author does not consider the designs organisms by any definition).32
There has also been dispute when using the terms “machine” and “robot” to describe the designs. Here, I must respectfully disagree with some members of the developmental biology community. Across this multidisciplinary research process, the author has come to appreciate that robots are defined by their rational design, autonomy, and ability to do human specified work, rather than by the fabrication materials. Indeed, many robots are constructed from nonintuitive components including cardiac powered biohybrid designs,20,22,33–35 pinecone and oat seed robots driven by hygromorphic actuators,36,37 liquid droplets,38–41 and a light-driven Caenorhabditis elegans RoboWorm,42 to name but a few.
In this framing, there is little difference between a biohybrid robot constructed from a synthetic soft-scaffold and muscle tissue and one constructed from a living soft-scaffold and muscle tissue. Both are designed for a human purpose, fabricated, and evaluated for performance, and both function through predictable open-loop control systems. As progress continues toward allowing our designs to sense and respond to their environment, their definition will begin to skew from biomachine toward biobot proper. These principles and continued exposure to the robotics field have been rewarding and inspiring for the author, who hopes members of other communities will likewise have an open mind when discussing the discipline.
Beyond the immediate science, an important point of discourse surrounding this research relates to the larger issue of science communication, in its many forms and venues.43 How, and where, should one promote one's own findings to both the scientific community and general public in an increasingly digital age? Our current research has been fortunate to receive significant visibility from the popular press, in both written and video formats. Engaging with these venues remains up to the investigator and there exists potential for both scientific benefit and risk with either course of action. Sensationalism, especially in headlines, drives traffic and revenue for periodicals, and a resistance to commentary with the popular press removes an important check to accurately reporting one's results.
Specifically, several periodicals have implied, or state definitively, that the system contains various degrees of cognition, or that the cells spontaneously generate novel functionality and behaviors rather than being designed by an investigator. However, as the scientist who envisioned and constructed all of the in vitro designs to date, the data do not support these speculations. Correcting these statements post hoc becomes challenging as the articles are often managed by a website publisher, meaning the original journalist no longer has editorial oversight of the piece. In other cases, publishers have been unwilling to address concerns, and the present author has removed their name from several articles, which misrepresent the research findings.
This is not a solution, however, as the incorrect message remains public facing in perpetuity. Further, close engagement with popular press represents a potential conflict of interest, as the investigator is both producing and selling the product, effectively creating a non-reliable narrator and eroding trust in the discipline at large.44 While the current author maintains caution when engaging with the press, it is also a reality that these venues generate value as privatized funding sources become increasingly common.
Finally, peer review, both during publication and through subsequent dissemination of the work remains essential to the research process. Presently, we have witnessed this review moving from the realm of grants, articles, and conferences into the digital world in the form of social media. The benefits here are many; investigators can rapidly reach a large community when sharing ideas, troubleshooting methods, creating a professional network, and providing scientific critique. However, there is also danger when scientific critique is built on an emotional response toward an individual rather than rational analysis of a result.45
As a community, we should continue to assume the best in our colleagues. Should a particular result or finding appear misplaced, one's goal should always be to improve the science rather than tear down an individual, and mentorship remains essential to members at all levels of academic positions. However, it is also necessary that scientists separate speculation from data, and we must absolutely hold our colleagues to the highest standards if we are to maintain public trust in scientific results. The current author has had to learn scientific communication on the fly, falling short on several occasions, and continues to appreciate the community feedback as he develops in this space.
In Jewish folklore and Greek mythology, it's relatively easy to create an intelligent robot: Simply combine mud (Golem) or sculpt ivory (Galatea) into the right shape. Without divine intervention, it's much harder to bring inanimate objects to life. Transforming containers of fluids and elastomers into a wiggling soft robot requires careful design and precise manufacturing.46 To get such a robot to do something else besides wiggling in place is exceedingly difficult.
So, why not start with living materials instead? In the “xenobots” project,7–9 that's exactly what we did. Creating a self-powered, self-driving, self-repairing, self-replicating robot (Table 1) was not trivial—doing so required careful planning, design, and construction—but it was, in essence, as simple as combining and sculpting the right material into the right shape. In our case, the right material was Xenopus cells and the right shape was one that, in computer simulations, maximized the likelihood of generating an interesting nonrandom behavior, such as forward locomotion.
This raises a tricky question: At what point do living tissues become a robot? Indeed, there are many different pieces of a developing animal that, when isolated from the host, can move and sense on their own. Parts of animals are not robots. However, once we artificially combine and shape them to render desired behaviors, they are no longer merely animal parts, they become artifacts: artificial yet fully biological robots.
If removed from early frog embryos, a small piece of ectoderm called the animal cap will mature to form a ball of skin covered in motile cilia, which can propel the ball forward through water. Early tadpoles glide around the same way, at the same speed as this explanted tissue. However, there are several ways to deflect (or “program”) the system away from its natural behavior. In our work to date, we have shaped these tissues into quadrupeds, bipeds, pyramids, toruses, semi-toroids, and various other nonspherical body shapes (Fig. 1). In some designs, we removed cilia entirely and relied instead on localized patches of cardiac tissue to produce volumetric actuation. Different body shapes, and different admixtures of tissue types, resulted in different (and thus artificial) behaviors.
The resulting “xenobots” are autonomous: they are able to maintain their structure (and thus function) without human intervention and exhibit a diverse array of behavioral repertoires. They are also adaptive: xenobots can repair their structure after significant damage. But are they intelligent?
The xenobots reported to date have no known mechanism of sensor–motor coordination. All the cells within a xenobot can sense and act and talk to their neighbors, but xenobots as a whole do not exhibit perceptive (sense-guided) behavior: Their movement is consistent with models of blindly actuating bundles of motors (open-loop control). Their healing is consistent with models of blindly adhering bundles of sticky particles. Sometimes xenobots appear to behave perceptively: They suddenly “decide” to turn around, or they become “interested” in an object and “examine” it repeatedly. However, the very same behaviors are manifested by quasi-stable bundles of motors when tipped into a new orientation.47 This is the problem of ascribing a capacity to an agent solely on the basis of its behavior. But this does not preclude future xenobots from genetic modifications that give rise to sense-guided behavior, action alternatives, memory, learning, and other increasingly cognitive behaviors.
Formally, we have referred to xenobots as reconfigurable organisms, a nod to reconfigurable modular robots48,49 in which robot modules (like the cells within a frog embryo) can be attached, detached, and rearranged to form new structures (configurations). As biorobotics researchers begin to reconfigure other organisms, the term “xenobot” might be reserved to describe a particular subset of reconfigurable organisms. If biological robots are one day built out of cells harvested from the African elephant (Loxodonta) instead of the African clawed frog (Xenopus laevis), then perhaps they will be called “loxobots” instead of “xenobots.”
(This is not to say the choice of “frog modules” was entirely arbitrary. Developing frog eggs, and xenobots derived from them, survive in fresh water without food; mammalian-derived robots such as loxobots would require special media with growth factors and nutrients and precisely tuned CO2 levels.) There is also the possibility of chimeras (mixtures of species) and biohybrids (mixtures of living and artificial materials). Any one of the creatures in this menagerie is just as likely to invoke the literal interpretation of xenobots: “strange robots.”
Whether we call them robots or organisms, and whether or not they become chimeras or cyborgs, designing and optimizing such systems is not only strange but also extremely nonintuitive. To make the design problem tractable for human minds, the solution space must be winnowed down to a vanishingly small subset of possible forms and functions. Breaking free from design constraints imposed by human cognitive limits would greatly widen our search for useful technologies and new knowledge, but it will by definition require nonhuman assistance. Computational tools (old8,50–54 and new9,55–57) are poised to help.
We used computers to rapidly generate a diversity of buildable xenobot forms that maximized a desired behavior and maintained that behavior across a range of simulated conditions (Fig. 2). This led us into parts of design space human engineers typically do not wander. There, we not only found creative xenobot designs (fractals, strange asymmetries, and porous structures) but also new biological design principles (e.g., how the right morphology can cohere the noisy actions of its parts).
Exploring this space in simulation allowed us to filter out billions of bad designs before attempting to build them in reality. This greatly improved efficiency in the wetlab and reduced biowaste. But this process can be energy-intensive and its carbon footprint grows with each second of simulation time. This is arguably the most pressing ethical concern of xenobot research. Future work should thus strive to encode xenobot design space such that, instead of meandering through random design variants,58 gradients can be efficiently followed toward optimal solutions.59 If successful, this move from trial-and-error to gradient descent could reduce processing time, and thus carbon footprint, by several orders of magnitude.
Biological robots, however efficiently designed, may never become a viable technology. This fact can get lost in the rhetoric surrounding xenobots. When engaging with the popular press and funding agencies (or when trying to explain your research to your grandma), it is helpful to articulate a compelling purpose. We have communicated several visions of the future in which xenobots clean up the environment, monitor industrial processes, deliver medicine, and even fight cancer. It is important to remember that xenobots are not yet useful: they cannot solve any problems at all.