paper
active
2021
ⓒ 99
paper:doi-10-3389-fevo-2021-650726

Living Things Are Not (20th Century) Machines: Updating Mechanism Metaphors in Light of the Modern Science of Machine Behavior

ByJoshua Bongard·Michael LevinⓘAllen Discovery Center At Tufts University, Allen Discovery Center, Tufts University + 11 more
DOI 10.3389/fevo.2021.650726OpenAlex W3137009159

TL;DR

Bongard and Levin argue that the longstanding debate over whether living things are machines has been conducted against a 20th-century, static definition of 'machine' that modern engineering has already surpassed, making the debate largely obsolete. Seven classical properties routinely invoked to distinguish machines from life—independence, predictability, human design, linear modularity, cognitive absence, reductionist tractability, and clear hardware/software distinction—each fail when tested against contemporary systems: evolutionary algorithms have produced jet engines (Yu et al., 2019), metamaterials (Zhang et al., 2020), and computer-designed Xenopus-cell organisms (Kriegman et al., 2020) without direct human specification of outcomes; backpropagation-trained deep networks resist exactly the reductionist decomposition Nicholson (2019) listed as a necessary machine feature; and planarian flatworms harbor re-writable bioelectric voltage patterns in non-neural cells that function as latent morphogenetic memory editable without touching the genome (Durant et al., 2017). The framework the paper introduces is the multi-axis continuum of 'machine behavior'—a 2D option space spanning degree of design vs. evolution and degree of autonomy, applicable independently at each level of biological organization (cell, organism, swarm)—drawn from the emerging interdisciplinary field Rahwan et al. (2019) named 'machine behavior.' Bongard and Levin argue this implies that the correct response is not to abandon the machine metaphor but to update it: biology and computer science are branches of a single information science, sharp boundaries between evolved and designed systems will not persist, and a conceptual framework that treats agency, programmability, and autonomy as continuous variables across all substrates is both necessary and sufficient to guide synthetic bioengineering, regenerative medicine, and machine design in the coming decades.

What to take away

  1. 1. Bongard and Levin identify seven classical machine properties—independence, predictability, human design, linear modularity, cognitive absence, reductionist tractability, and hardware/software separability—and argue each has been empirically undermined by advances in Machine Behavior, artificial life, and synthetic bioengineering.
  2. 2. An evolutionary algorithm operating on a two-cell-type (Xenopus laevis epithelial and cardiac muscle) construction kit produced computer-designed organisms whose locomotion arose entirely from novel shape and tissue distribution rather than neural control or altered DNA (Kriegman et al., 2020), inverting the assumption that machines require human design.
  3. 3. Planarian flatworms contain voltage-pattern memories in non-neural cells that function as re-writable regenerative targets: editing these bioelectric gradients causes worms to permanently generate two-headed forms despite retaining completely wild-type genomic sequence (Durant et al., 2017), demonstrating a sharp software/hardware inversion in biology.
  4. 4. Nicholson (2019) defined machines by four necessary features—specificity, constraint, efficiency, and non-continuity (halt-disassemble-repair-restart)—and the paper demonstrates that 21st-century machines, including deep neural networks and swarms, fail all four criteria, making those features contingent rather than essential.
  5. 5. The paper introduces the multi-axis continuum framework—a 2D option space whose axes are degree of design vs. evolution and degree of autonomy—applied independently at each nested biological scale (organelle, cell, organism, swarm), as a replacement for binary life/machine categories.
  6. 6. Artificially evolved neural networks (Clune et al., 2013) and morphologically evolved robots (Bernatskiy and Bongard, 2017) lack modularity unless it is explicitly selected for, and many exhibit inefficiencies from evolutionarily duplicated sub-structures, erasing efficiency as a machine-distinguishing trait.
  7. 7. Deep reinforcement learning—the primary training method for autonomous cars and drones—is biologically inspired by behaviorism (Mnih et al., 2015), while convolutional neural networks are inspired by hierarchical receptive fields in primary visual cortex (Krizhevsky et al., 2017), illustrating that the most successful AI architectures already blur the life/machine boundary.
  8. 8. A soft robot recovered faster from unexpected physical damage by contorting its body into a new shape (a hardware change) than by learning a compensating gait (a software change) (Kriegman et al., 2019), empirically inverting the standard assumption that hardware is harder to alter than software.
  9. 9. The paper raises the open question of whether any essential natural kinds or major transitions exist within the continuum of fused biological-technological systems, or whether all apparent discontinuities are artifacts of the history of science rather than deep ontological boundaries.
  10. 10. As a replicable methodology, the paper operationalizes 'degree of roboticism' as the amount of energy or intervention effort required to produce a given magnitude of behavioral change in a system, providing a continuous, substrate-independent metric applicable to prosthetics, autonomous vehicles, and hybrid biotic-electronic organisms alike.

Peer brief — for seminar discussion

Bongard and Levin's 2021 hypothesis-and-theory paper, published in Frontiers in Ecology and Evolution, takes the centuries-old debate over whether living things are machines and reframes it as a definitional problem: opponents of the machine metaphor have been arguing against a 19th- and 20th-century conception of machines that contemporary engineering has already discarded. The paper systematically works through seven classical distinguishing properties—independence, predictability, human design, linear modularity, cognitive uniqueness, reductionist tractability, and hardware/software separability—and in each case marshals recent empirical and theoretical work to show the distinction has collapsed. The load-bearing argument is not philosophical but evidential: evolutionary algorithms have produced jet engines (Yu et al., 2019), graphene metamaterials (Zhang et al., 2020), and Xenopus laevis-based computer-designed organisms whose locomotion depends entirely on novel shape and tissue distribution rather than genome or neural control (Kriegman et al., 2020); planarian flatworms harbor re-writable bioelectric memory in non-neural cells that directs regeneration independently of DNA sequence, and editing those voltage patterns produces permanently two-headed worms from wild-type genomes (Durant et al., 2017); and Nicholson's (2019) four-feature machine definition—specificity, constraint, efficiency, non-continuity—fails empirically against deep networks, swarms, and soft robots. The method the paper introduces is a multi-axis continuum framework: a 2D option space whose axes are degree of design vs. evolution and degree of autonomy, applied independently at each level of biological organization (cell, organism, swarm), drawn from the emerging field Rahwan et al. (2019) named 'machine behavior.' An alternative approach the paper could have used is formal complexity-theoretic analysis—measuring autonomy via information-theoretic empowerment or causal closure indices—rather than the discursive taxonomy it employs. The central implication is that biology and computer science are both branches of information science operating in different media, and that a framework treating agency, programmability, and autonomy as continuous variables is necessary to guide regenerative medicine, synthetic bioengineering, and regulation of hybrid systems. The paper also advances a prediction: near-future systems will be composed of mixed biological and artificial components at multiple organizational scales simultaneously, making binary life/machine classification not merely imprecise but operationally useless. A critical reader would push back on the paper's scope: it is entirely a conceptual review with no new experimental data, and its core move—replacing a bad definition with a better one—sidesteps the harder question of whether the proposed continuum axes are measurable with sufficient precision to do real scientific or regulatory work. The 2D option space is intuitive but unquantified, and the claim that 'degree of autonomy' and 'degree of design vs. evolution' are the right principal components is asserted rather than derived; alternative axes such as metabolic self-maintenance, error-correction capacity, or open-ended evolvability might carve the space differently and yield different conclusions about where particular hybrid systems fall.

Findings (8)

Claims (30)

Hypotheses (3)

Questions (7)

Original abstract (expand)

One of the most useful metaphors for driving scientific and engineering progress has been that of the “machine.” Much controversy exists about the applicability of this concept in the life sciences. Advances in molecular biology have revealed numerous design principles that can be harnessed to understand cells from an engineering perspective, and build novel devices to rationally exploit the laws of chemistry, physics, and computation. At the same time, organicists point to the many unique features of life, especially at larger scales of organization, which have resisted decomposition analysis and artificial implementation. Here, we argue that much of this debate has focused on inessential aspects of machines – classical properties which have been surpassed by advances in modern Machine Behavior and no longer apply. This emerging multidisciplinary field, at the interface of artificial life, machine learning, and synthetic bioengineering, is highlighting the inadequacy of existing definitions. Key terms such as machine, robot, program, software, evolved, designed, etc., need to be revised in light of technological and theoretical advances that have moved past the dated philosophical conceptions that have limited our understanding of both evolved and designed systems. Moving beyond contingent aspects of historical and current machines will enable conceptual tools that embrace inevitable advances in synthetic and hybrid bioengineering and computer science, toward a framework that identifies essential distinctions between fundamental concepts of devices and living agents. Progress in both theory and practical applications requires the establishment of a novel conception of “machines as they could be,” based on the profound lessons of biology at all scales. We sketch a perspective that acknowledges the remarkable, unique aspects of life to help re-define key terms, and identify deep, essential features of concepts for a future in which sharp boundaries between evolved and designed systems will not exist.

Related work— refs + corpus + external arXiv

Cited / in-corpus / arXiv badges show which signals surfaced each row. Multi-source rows weighted higher.

+28 more

Similar preprints — Semantic Scholar

Cited by (5)

Cross-corpus bridges (12)

same_concept_as · Nomic cosine

External markdown files that talk about the same concept as this entity.

  • aboutblank_kb
    Can synthetic living machines exhibit genuine cognition and intentional behavior without evolved neural systems?questions/can-synthetic-living-machines-exhibit-genuine-cognition-and.md0.853
  • aboutblank_kb
    Can artificial systems designed to be living achieve comparable cognition and adaptability to natural organisms?questions/can-artificial-systems-designed-to-be-living-achieve.md0.834
  • aboutblank_kb
    When we combine living material with electronics and computer software, where should we draw the line between 'machine' and 'organism'?questions/when-we-combine-living-material-with-electronics-and.md0.826
  • aboutblank_kb
    How can we identify and evaluate the sentience of beings created through bioengineering, synthetic morphology, and AI that offer no familiar touchstones?questions/how-can-we-identify-and-evaluate-the-sentience.md0.819
  • aboutblank_kb
    How do we explain the ubiquitous presence of novel capabilities in living things for which no direct selection forces are plausible?questions/how-do-we-explain-the-ubiquitous-presence-of.md0.819
  • aboutblank_kb
    Where do the goals, preferences, and competencies of novel beings come from when they have never existed in evolution?questions/where-do-the-goals-preferences-and-competencies-of.md0.814
  • aboutblank_kb
    Can novel body forms implemented in silico and synthesized in vivo achieve goal-directed behavior without evolutionary design?questions/can-novel-body-forms-implemented-in-silico-and.md0.814
  • aboutblank_kb
    What role do competing interpretations play in enabling biological innovation and evolutionary flexibility?questions/what-role-do-competing-interpretations-play-in-enabling.md0.807
  • aboutblank_kb
    Where do goals, preferences, and behaviors of novel artificial beings come from if they've never been subject to evolutionary selection?questions/where-do-goals-preferences-and-behaviors-of-novel.md0.805
  • aboutblank_kb
    How can bioethics adapt to recognize consciousness in radically different architectures?questions/how-can-bioethics-adapt-to-recognize-consciousness-in.md0.805
  • aboutblank_kb
    How can cognitive science accommodate beings whose origin is de novo design rather than evolutionary history?questions/how-can-cognitive-science-accommodate-beings-whose-origin.md0.805
  • aboutblank_kb
    How should we define and identify genuine artificial life and intelligence in synthetic organisms and biological robots?questions/how-should-we-define-and-identify-genuine-artificial.md0.803