Endless forms most beautiful 2.0: teleonomy and the bioengineering of chimaeric and synthetic organisms

Abstract

The rich variety of biological forms and behaviours results from one evolutionary history on Earth, via frozen accidents and selection in specific environments. This ubiquitous baggage in natural, familiar model species obscures the plasticity and swarm intelligence of cellular collectives. Significant gaps exist in our understanding of the origin of anatomical novelty, of the relationship between genome and form, and of strategies for control of large-scale structure and function in regenerative medicine and bioengineering. Analysis of living forms that have never existed before is necessary to reveal deep design principles of life as it can be. We briefly review existing examples of chimaeras, cyborgs, hybrots and other beings along the spectrum containing evolved and designed systems. To drive experimental progress in multicellular synthetic morphology, we propose teleonomic (goal-seeking, problem-solving) behaviour in diverse problem spaces as a powerful invariant across possible beings regardless of composition or origin. Cybernetic perspectives on chimaeric morphogenesis erase artificial distinctions established by past limitations of technology and imagination. We suggest that a multi-scale competency architecture facilitates evolution of robust problem-solving, living machines. Creation and analysis of novel living forms will be an essential testbed for the emerging field of diverse intelligence, with numerous implications across regenerative medicine, robotics and ethics.

Full text

1
Biological Journal of the Linnean Society, 2022, XX, 1–30. With 8 figures.
Endless forms most beautiful 2.0: teleonomy and the
bioengineering of chimaeric and synthetic organisms
WESLEYP.CLAWSON1 and MICHAELLEVIN1,2,*,
1Allen Discovery Center at Tufts University, 200 Boston Avenue, Suite 4600, Medford, MA 02155, USA
2Wyss Institute for Biologically Inspired Engineering at Harvard University, Center for Life Science
Building, 3 Blackfan Circle, Boston, MA 02115, USA
Received 15 March 2022; revised 31 May 2022; accepted for publication 4 June 2022
The rich variety of biological forms and behaviours results from one evolutionary history on Earth, via frozen
accidents and selection in specific environments. This ubiquitous baggage in natural, familiar model species obscures
the plasticity and swarm intelligence of cellular collectives. Significant gaps exist in our understanding of the
origin of anatomical novelty, of the relationship between genome and form, and of strategies for control of large-
scale structure and function in regenerative medicine and bioengineering. Analysis of living forms that have never
existed before is necessary to reveal deep design principles of life as it can be. We briefly review existing examples of
chimaeras, cyborgs, hybrots and other beings along the spectrum containing evolved and designed systems. To drive
experimental progress in multicellular synthetic morphology, we propose teleonomic (goal-seeking, problem-solving)
behaviour in diverse problem spaces as a powerful invariant across possible beings regardless of composition or origin.
Cybernetic perspectives on chimaeric morphogenesis erase artificial distinctions established by past limitations of
technology and imagination. We suggest that a multi-scale competency architecture facilitates evolution of robust
problem-solving, living machines. Creation and analysis of novel living forms will be an essential testbed for the
emerging field of diverse intelligence, with numerous implications across regenerative medicine, robotics and ethics.
ADDITIONAL KEYWORDS: artificial life – basal cognition – chimaeras – evolution – hybrids – synthetic
morphology.
INTRODUCTION: OPEN PROBLEMS AND
KNOWLEDGEGAPS
Progress in molecular biology and genetics has led
to great strides in understanding the micro-scale
hardware of cells (the protein machinery encoded
by the genome). However, as shown clearly by the
trajectory of the information sciences, this is only the
beginning; the next frontier is the software of life:
developing a mature science of prediction and control
over system-level phenotypes. Despite a deluge of
big data on the molecular mechanisms necessary
for specific functionalities, important capability and
knowledge gaps remain with respect to the dynamics
that are sufficient for the remarkable robustness and
plasticity we observe in the livingworld.
A fertilized egg produces a cellular swarm that
reliably self-assembles into a highly complex organism.
Importantly, this process is not hardwired: mammalian
embryos cut in half produce normal monozygotic twins,
because despite this damage each side can grow what
is missing. Some organisms maintain this regenerative
capacity throughout their lifetime; for example,
salamanders regenerate limbs, jaws, eyes, tails and
ovaries (McCusker & Gardiner, 2011). Asalamander
limb can be amputated at any level and will produce
precisely the missing parts and then stop when a correct
salamander limb is complete (when the distance from
the correct target morphology is sufficiently reduced).
The ability to handle novelty in the form of external
damage is not the only aspect of the robust plasticity of
life; this ability also extends to unexpected changes in
the internal building blocks of the organism.
Tadpoles generated to have no primary eyes, but
an ectopic eye on their tail, can see reasonably well
*Corresponding author. E-mail: michael.levin@tufts.edu
© 2022 The Linnean Society of London.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://
creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any
medium, provided the original work is properly cited.
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

2 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
(Blackiston & Levin, 2013); in their aberrant new
location, eye primordia cells form a proper eye and
often connect the optic nerve to the spinal cord.
Abrain that has evolved for millions of years to expect
visual input to a specific location immediately adjusts
its behavioural programmes to operate with signals
coming from its tail. The same plasticity has been
observed in adult humans provided with novel senses
and effector organs (such as prosthetics with novel
degrees of freedom; Bach-y-Rita, 1967; Bach-y-Rita
etal., 1969; Danilov & Tyler, 2005; Nagel etal., 2005;
Shull & Damian, 2015).
Perhaps even more impressive are examples of
structural robustness to change. Kidney tubules in
newts normally form from the interactions of eight to
ten cells in cross-section, working together to make a
lumen of a specific size. Cell size can be made larger
by artificially increasing the ploidy (chromosome
number) of embryonic cells; that this results in a
viable embryo is, in itself, amazing. Such animals
are of a normal size and proportion, because fewer
and fewer cells participate in the tubulogenesis as
the cell size increases. Most remarkably, when cells
are truly enormous, a single cell will wrap around
itself to make a proper tubule (Fankhauser, 1945a,
b). In this case, instead of the normal cell-to-cell
communication, cytoskeletal bending is used to
achieve the same morphological goal. Thus, diverse
molecular mechanisms are triggered in service of
a higher-level anatomical specification (a tubule of
specific cross-section).
These changes in the quantity of genetic material
and cell size are dealt with dynamically. Our robotics
technology does not even begin to approach this kind
of capacity, and any engineered swarm that could
adjust to this type of novelty (perturbation in size
and information content of its components) would be
hailed as a milestone in artificial intelligence. The
connection to intelligence is not accidental: William
James defined it as the ability to reach the same goal
by different means (James, 1890). The proficiency of
living systems in this respect is best revealed not by
the reliable normal development of standard model
species (which obscures the true capacities of cellular
swarms), but by the type of engineered, novel beings
of which the above altered examples are only the
beginning.
This type of plasticity, in the context of epigenetic
controls and responses to the environment, is
familiar to biologists with respect to changes of the
environment and epigenetic control (West-Eberhard,
1998, 2005a, b) and extends to all scales, from the
variability of the traditional environment of the
whole animal to the microenvironment (internal
properties) of organs and cells. Life is massively
inter-operable, enabling functional chimaeras and
hybrids at the molecular, cellular, tissue, organ
and even organism levels (Nanos & Levin, 2022).
What computations, algorithms or dynamics enable
cellular collectives to respond adaptively, reaching
the same form and function despite radical induced
changes of circumstances? Importantly, all these
phenomena show adaptations to novelty that exists
on the time scale of an individual, not requiring
aeons of evolutionary search. This suggests that life
exploits an architecture that provides much more
efficiency than could be expected from a blind process
that always chooses short-term gains (selection that
adapts to a specific environment).
The answer to ‘what determines the shape of an
organism?’ is often said to be ‘the genome’, but many
deep questions remain about the relationship between
the genome and anatomy. In addition to artificially
produced chimaeric organisms, in which diverse
genomes can live together and generate large-scale
form and function (Nanos & Levin, 2022), some animals
are natural chimaeras. Some species of planarians
reproduce largely by fission and regeneration; this
avoids the segregation of the germline in a type of
somatic inheritance: any mutation that does not
kill a stem cell takes it into the next generation to
proliferate in the lineage (Fields & Levin, 2018; Levin
etal., 2019). For hundreds of millions of years, these
animals have accumulated mutations and are even
mixoploid; different cells within one animal can have
different numbers of chromosomes. And despite this
messy genome, they are champions of regeneration,
building the correct body from even small fragments
with very high anatomical fidelity (Saló etal., 2009);
each piece of a cut planarian produces a perfect
little worm. We have no models in developmental
genetics that would predict that the highest fidelity
of anatomical outcomes would be associated with
genetic diversity that rivals any tumour. Indeed,
planarian lines can even be made permanently two
headed by manipulating the bioelectric circuit that
stores head number (Oviedo etal., 2010; Durant
etal., 2017), resulting in permanent ‘strains’ of
animals whose cells continue to build worms with
a different anatomical body plan from the genomic
default. What is the relationship between the genome
and anatomy, and what mechanisms allow biology
to exhibit robustness and plasticity simultaneously,
enabling adaptive, coherent organisms to arise in
novel circumstances?
Moreover, the highly competent decision-making
of cellular collectives in anatomical morphospace
reveals a fascinating commonality between
problems of cognition (mind) and problems of body
(morphogenesis). This fundamental link was well
understood by early workers in developmental biology,
such as Hans Spemann (Spemann, 1967), and those in
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

ENGINEERING NOVEL BODIES AND NOVEL MINDS 3
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
computer science, such as Alan Turing (Turing, 1952);
it is only now beginning to be fleshed out (Grossberg,
1978; Friston etal., 2015; Pezzulo & Levin, 2015).
Ontogeny recapitulates phylogeny, in that all of us
have made the journey across the Cartesian cut (from
the ‘mere physics’ of molecular networks in a quiescent
oocyte to a thinking being) not only during evolution
but also during our own lifetimes. What aspects of
biological structure enabled mind gradually to develop
andexpand?
These knowledge gaps are not obscure issues in
evolutionary biology and philosophy of mind. The
whole of regenerative medicine hangs on the question
of how to induce collectives of cells to build one
structure rather than another. Birth defects, traumatic
injury, cancer and degenerative disease would all
be solved if we had a mature science of making
predictions and deriving rational interventions into
the morphogenetic process (Levin, 2011). Advances in
molecular medicine and genomic editing will not have
an impact on biomedicine unless we know what to edit
or which pathways to target to achieve system-level
goals, such as ‘make a new arm’. It is no accident that
current medical interventions that solve problems in
the long term exist only in the realms of infectious
disease and surgery. Transformative regenerative
medicine awaits a mature understanding of how to
induce collections of cells to make desired anatomical
features.
THE NEED TO GO BEYOND STANDARD
MODELSPECIES
Chimaeras and bioengineered organisms challenge
us to make predictions and spotlight areas in which
genomics have driven unwarranted complacency. For
example, when we make a frogolotl (a hybrid frog–
axolotl embryo), will it have legs, like a larval axolotl,
or not, like a tadpole? If it has legs, will they be made of
frog cells or only of axolotl cells? We have no models in
biology to make predictions about such cases, despite
having full access to the genomes of both species.
Crucially, this inability to predict or control outcomes
is not a special feature of rare ‘corner cases’; it lays bare
the often-neglected fact that even for a single species,
knowing its genome enables us to say almost nothing
about the form or function of the organism it ‘encodes’
(except when we cheat by comparing the genome with
that of organisms whose anatomy we know already).
This is because, although the genetics specify the cell-
level hardware of the system (proteins), the outcome
is the product of physiological software dynamics that
are not predicted easily from the hardware level (Lobo
etal., 2014; Pezzulo & Levin, 2016).
If we ‘zoomed in’ to observe developmental events
at the cellular level, seeing all of the stochastic cell
behaviour and signalling noise, would we be able to
predict that all of that activity would reliably give rise
to a fish or mouse, if we did not already know about
development and the fact that it is highly reliable
in a range of conditions? Managing the reliability
of outcomes in novel circumstances challenges us to
develop a science of predicting stable outcomes at
large scales (for a similar issue in neuroscience, see
Jonas & Kording, 2017). Knowing how to detect and
characterize specific goal states (in the cybernetic
sense) of collectives, such as cell groups, is a crucial
part of understanding systems. Novel instantiations
of multicellular life are a crucial dataset on which to
train and improve the conceptual tools of scientists
and learning machines and must complement the
developmental biology of standard, evolved model
systems.
Beyond life on Earth, how would we recognize
novel forms of life? What is the appropriate
scale of observation for detecting the behaviour
and appropriate problem spaces in which life
operates? Bioengineering provides a crucial inroad
for exobiology; a stepping-stone for enabling
generalization of biology such that we can detect
truly alien forms of life if and when we encounter
them. Regardless of natural life elsewhere, exploring
the option space of beings enables us to improve
the terms and categories we use to understand life,
manipulating all the components in a schema such as
the traditional brain–body–environment framework
to unfamiliar components, in order to see how our
existing approaches break down when faced with
unfamiliar implementations. Like the very successful
strategy of looking for symmetries in physics, in
biology we must ask which of our concepts are deep
invariants. Which of them remain when contingent
details of implementation (cell type, genetics, origin,
etc.) are changed?
The need to expand beyond familiar life forms goes
deeper than Feynman’s dictum, that we do not really
understand a thing until we can make one ourselves.
The future of biomedicine and engineering depends on
being able to offload much of the complexity we wish to
manage onto the system itself; for example, triggering
a body to regenerate a limb in the right location
instead of trying to micromanage its construction
from the molecular level. This, in turn, relies on
understanding the modules, decision-making and
information processing of which the system is capable.
We have argued in the past that an important aspect
of future medicine will involve guiding the large-
scale behaviour of cellular collectives with incentives,
stimuli and set point rewriting (Pezzulo & Levin,
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

4 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
2015, 2016; Mathews & Levin, 2018), rather than
micromanaging the structure of their parts. This means
that we need to understand the algorithms guiding
these systems at multiple scales and, in particular,
their basal intelligence (i.e. their ability to navigate
physiological, transcriptional and morphological
spaces competently despite novel circumstances and
perturbations). However, such intelligence is rarely
apparent in ‘normal’ circumstances, when a system
appears to be doing the same thing every time; this is
what gives rise to a view of most biological systems as
clockwork mechanisms, full of complexity but without
intelligence. In order to uncover, understand, control
and, eventually, cooperate with the true intelligence of
biology at all scales, it is essential to confront biological
systems with novelty, both inside and out, and to study
the context-specific problem-solving capacities and
plasticity (Braun, 2015) that are revealed in response
to that novelty.
Chimaerism (mixing biological components) was
a popular concept in the ancient world (Fig. 1A–C),
but the story of Adam naming a discrete set of
animals in the Garden of Eden (Fig. 1D) suggests
a different and much more limiting picture: a type
of essentialism that suggests sharp categories
(distinctions of biological form) and natural
kinds that do not, in fact, exist. We now have the
opportunity to extend this story and ‘name the
animals’ in a much deeper way, by understanding
the design principles of biology that transcend
extant evolutionary examples. The implications of
embracing the space of possible beings will extend
to terminology, conceptual frameworks, research
programmes in several fields, andethics.
Here, we initially review some examples of
existing technologies that promise to expand our
understanding of life radically. Going beyond
classical chimaeras, we describe the mergers of
diverse products of evolution and human design
to sketch the dimensions of the space of possible
beings. Given that origin, composition and familiar
phylogenetic position will not be reliable guides
to properties of living forms in this space, we then
suggest an approach to the search for invariants:
what do all such forms have in common that can
be used to compare them directly and understand
them? We suggest that goal-directed behaviour,
in diverse spaces, is a central concept that fulfils
the role of a framework for driving experimental
approaches. Towards a unification of the sciences of
the mind and body, we then introduce an expansion
of neuro- and behavioural science outside the brain.
We discuss the fields of basal cognition and diverse
intelligence, in order to begin to generalize the
idea of goal-directed activity beyond the function of
complex brains. We discuss teleonomy as a guiding
framework for understanding diverse aspects of
biology, suggesting that a deep principle of biology
is nested goal directedness at multiple levels. We
next explore some implications of such a multi-scale
competency for evolution. Finally, we conclude with
a sketch of a research programme, enabled by these
ideas, which spans regenerative medicine, robotics
and ethics.
Figure 1. Chimaeras and natural kinds. A, a representation
of the Devourer, who waited to eat the hearts of sinners in
the afterlife’s Hall of Judgment. Papyrus of Ani, ~1275BC;
photograph from British Museum. B, Matsya (fish) Avatar of
Vishnu. Nineteenth century lithograph. From Wikipedia. C,
Oannes, a Mesopotamian mythological being who brought
civilization to mankind. Curious creatures in zoology, by
John Ashton (1890), p.209. Available at: https://commons.
wikimedia.org/wiki/File:Curious_creatures_in_zoology_
(15565912981).jpg D, ‘Adam names the animals in the
Garden of Eden’ illustrates a pre-scientific (but implicitly,
still widely prevalent notion) that current animals and
plants represent discrete natural kinds with sharp
separations, especially with respect to humans. Available
at: https://wellcomecollection.org/works/q6hw2nrg/items
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

ENGINEERING NOVEL BODIES AND NOVEL MINDS 5
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
SYNTHETIC BEINGS: EXPANDING THROUGH
THE SPACE OF POSSIBLE BODIES
ANDMINDS
My son, if you would devote yourself to combining
Holy Names, still greater things would happen to
you.… And now, my son, admit that you are unable
to bear not combining. Give half to this and half to
that, that is do combinations half of the night, and
permutations half of thenight.
Abraham ben Samuel Abulafia
Examples of novel ‘life as it could be’ (Fig. 2) are
highly diverse, and current technologies and proof-of-
concept results clearly indicate the coming diversity
of life in the near future (Doursat et al., 2013;
Doursat & Sanchez, 2014; Kamm & Bashir, 2014;
Ebrahimkhani & Levin, 2021). These novel life forms
result from the recombination of existing evolved
components at multiple scales (DNA, cells, tissues and
organs) (Nanos & Levin, 2022) and the incorporation
of designed components, such as nanomaterials,
electronic/chemical/optical interfaces and software
algorithms; replacements can be made at every level
of organization, with parts that occupy some position
with respect to functional sophistication and origin
(Fig. 3). Table 1 is an overview of some key examples
(although sharp boundaries between these categories
cannot be drawn), and we describe only a few in detail
(Fig. 4).
Bioengineering and evolution work in the same
living medium (Wagner & Rosen, 2014; Ollé-Vila etal.,
2016); thus, we first consider an example of an entirely
natural evolved functional chimaerism, where two
kingdoms of life come together to form a new hybrid
system (Fig. 4A–D). Ophiocordyceps unilateralis
s.l. is a fungal parasite that creates ‘zombie ants’ of
Camponotus castaneus, a carpenter ant (Andersen
etal., 2009). This fungus infects an ant, freely flowing
through the circulatory system of the host. After it
expands throughout the organism, takes over the
body of the host ant. Under control of the fungus,
the host ant navigates to a plant located near the
semi-permanent food trails of the colony, climbs up
to an optimal growth zone and bites onto the plant,
allowing the fungal spores to disperse onto uninfected
ants wandering the nearby trails. Most interestingly,
the brain of the host ant is left entirely untouched
(Fredericksen etal., 2017). This poses the question of
how a ‘simple’ fungus can manipulate the behaviour of
the host ant in such a reproducible way. To achieve its
goal of reproduction, the fungus must adapt to sensory
signals from the environment; this involves processing
the incoming signals to determine when it has come
in contact with a suitable host and can thus begin
targeted growth by navigating inside the host body.
Importantly, this process needs to be precise enough
to take control of the host ant without altering the
behaviour of the host too strongly, because ant colonies
will reject infected members of the colony. However, as
the fungus infects the host ant further, what incoming
signals does the fungus receive and how does it process
these signals now that it is in a body not its own? Is it
cut off from the external world and does it ‘see’ only the
inside of the ant, or does the fungus hijack the sensory
system of the host, thus acquiring new ways to sense
and act inthe world? Although these are unanswered
questions, the study by Fredericksen etal. (2017)
demonstrates that large, complex fungal networks
invade the muscle fibres in the host, potentially
allowing for precise body-wide control without need
for the brain-to-central nervous system axis. However,
agent-based modelling shows that the fungus might
hijack and repurpose the ‘food searching’ behavioural
regimen of the ant into an algorithm to find sites for
fungal dispersion (Imirzian & Hughes, 2021). This
would imply that O.unilateraliss.l. has developed
a way evolutionarily to exhibit control of a dynamic
system (the nervous system of the host ant) without
needing to understand how the nervous system of
its host functions. It is clear that there is much to be
understood in chimaeric ‘life as it is’ studies, because
researchers can study existing evolutionary accidents
to reveal common relationships between brain, body
and behaviour across species and systems.
The field of artificial chimaeras has been realized
most notably in brain–computer interfaces (Bonifazi
etal., 2013; Buccelli etal., 2019; Degenhart etal.,
2020). Willett etal. (2021) carried out a remarkable
study that merges existing biological systems, here the
motor cortex in a human subject, with technology to
provide a way for a patient to communicate with the
outside world even years after the onset of paralysis
(Fig. 4E–G). They instructed a patient to ‘to “attempt”
to write as if his hand were not paralyzed, while
imagining that he was holding a pen on a piece of
ruled paper’, while recording neural activity from the
precentral gyrus. Using a variety of methods, neural
activity was decoded and associated with each letter,
allowing a computer program to produce written
sentences for the patient in real time at a speed much
higher than other approaches (Willett etal., 2021).
In many ways, this artificial chimaera works in the
same way as the parasitic fungus; experimenters do
not need to know how certain neural dynamics arise
owing to imaginary handwriting, merely that these
signals can be used to drive goal-directed behaviour.
In addition, this particular patient was not paralysed
at birth; writing had been learned before the paralysis,
and the hypothesis is that this aided in the stability
of the decoded neural dynamics. This is similar to the
case of the zombie ants; the fungi were not necessarily
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

6 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
Figure 2. The option space of unconventional novel agents. The interoperability of the multi-scale competency architecture
of life enables novel agents to be formed in any combination of evolved material, designed (engineered) material and
software. This forms an immense option space of hybrots, chimaeras, cyborgs and many other kinds of novel creatures never
before seen on Earth and having no clear relationship to the existing phylogenetic lineage. Image courtesy of Jeremy Guay
of Peregrine Creative.
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

ENGINEERING NOVEL BODIES AND NOVEL MINDS 7
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
Figure 3. Multi-scale chimaerism. Bioengineering now allows every layer of a tiered living system to be replaced with
components from some position in a plane of orthogonal metrics of how much cognition it has and how much design/
evolution resulted in its creation. Evolutionary techniques in machine design and tools for synthetic morphology are erasing
the artificial lines that used to exist between evolved, living forms and engineered machines with teleonomic capacity.
Image courtesy of Jeremy Guay of Peregrine Creative, taken with permission from Bongard & Levin (2021).
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

8 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
creating a new behaviour in their hosts but were
instead exploiting existing behaviours towards a goal.
In the case of Willett etal. (2021), the chimaerism
arises artificially; the computer software is built
to sense the ‘environment’ of motor cortex neural
dynamics to produce action in a virtual world, action
that previously was not possible. It is important to note
that the neural signals were recorded from a cortical
region of the brain associated with motor movement.
Indeed, it is assumed that specific cortices evolved
to control specific behaviour, and although neural
representations of movement can change owing to
prosthetics (Kieliba etal., 2021), the question is raised:
can neurons from other modalities, such as vision, be
trained in external motor control? How ‘frozen’ is the
behaviour of cortical regions?
Hybrots and animats are systems composed of neural
cultures, coupled via closed loop techniques to physical
robots (in the former) and virtual animals (in the latter),
and are ideal platforms for tackling such questions
(Fig. 4H, I). Early examples were developed in 2001,
when DeMarse and co-workers dissociated cortical
tissue from rats and cultured it as a two-dimensional
plane of neural tissue on a microelectrode array; this
array was capable of recording the electrical activity of
the tissue while delivering stimuli via electrical pulses
(DeMarse etal., 2001). The spiking activity of the
neurons drove a virtual animal, the animat, in a virtual
maze by associating unique spatiotemporal spiking
patterns with directional movement. Feedback based
on that directional movement and on the distance to
obstacles in the maze was then returned to the culture.
This was designed to mimic how neural systems have
evolved to take input and interact meaningfully with
the world, with the goal of allowing the culture to learn
relationships between its own activity and incoming
‘sensory stimuli’.
This paradigm was later used to build MEART
(multi-electrode array art), a hybrot that was built
to create art through neural control of robotic arms
attached to drawing utensils (Bakkum etal., 2007b).
This study focused on examining the stimulus
(patterned training stimulus) that was applied to the
neural culture, based on the present drawing and the
desired ‘goal drawing’, a black square in the middle of
the canvas. Here, they reported that although there
were shifts in synaptic plasticity in the network, if
the patterned training stimulus was not updated at
regular intervals, the overall behaviour did not reflect
any signs of learning. This closed-loop chimaeric
approach allows for investigation of the plasticity
with which evolved biological components learn to
function with novel bodies and environments. It is
not known how neural systems adapt their dynamics
as new behaviours, whether goal directed or not, are
learned, with hypotheses typically revolving around
Table 1. Examples of novel life configurations
Type of life form Properties References
Embryoids, organoids
and assembloids
Ex vivo cultured cells and tissues
with emergent morphogenesis
Simunovic & Brivanlou (2017); Vogt (2021)
Cyborgs Tissues of animals and plants tightly
integrated with engineered inor-
ganic interfaces, often with closed-
loop controls enabling the cells to
control and be controlled by ma-
chines and their microenvironment
Cohen-Karni etal. (2012); Giselbrecht etal.
(2013); Warwick (2014); Gershlak etal.
(2017); Aaser etal. (2017); Ricotti etal.
(2017); Ding etal. (2018); Mehrali etal.
(2018); Anderson etal. (2020); Merritt etal.
(2020); Orive etal. (2020); Li etal. (2021); Pio-
Lopez (2021)
Biorobotics Computer-controlled animals Ando & Kanzaki (2020); Saha etal. (2020);
Dong etal. (2021)
Biobots Synthetic living machines with pre-
dictable behaviour
Park etal. (2016); Aydin etal. (2019); Pagan-
Diaz etal. (2019); Kriegman etal. (2020,
2021); Blackiston etal. (2021)
Biocomponents Repurposed biological structures as
components of machines
Whiting etal. (2016); Adamatzky (2018)
Neuroprosthetics and
sensory augmentation
Interfaces enabling patients to con-
trol novel effectors or use novel
sensors
Rothschild (2010); Lebedev & Nicolelis (2011);
van den Brand etal. (2015); Adewole etal.
(2016); Turner (2016); Wright etal. (2016)
Hybrots Living brain tissue instrumentized to
control artificial new bodies, such
as vehicles
Warwick (1998); DeMarse etal. (2001); Potter
etal. (2003); Madhavan etal. (2006); Bakkum
etal. (2007a, b); Tsuda etal. (2009); Ando &
Kanzaki (2020)
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

ENGINEERING NOVEL BODIES AND NOVEL MINDS 9
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
Figure 4. Three examples of novel organisms. A, a photograph of a host Camponotus atriceps ant in the death-bite position
induced by infection from Ophiocordyceps unilateralis s.l. B, C, immunofluorescence images of neural synapses (green) in an
uninfected ant (B) with no host tracheae (red) alongside an image taken from infected ant (C); note how there are no hyphal
bodies (red) inside the brain (green). D, however, through three-dimensional reconstruction techniques, Fredericksen etal.
(2017) found hyphal bodies (yellow) around adductor muscles (red) in the ant, a part of the complex fungal network throughout
the host body. A, used with permission from Hughes & Libersat (2019). B-D used with permission from Fredericksen etal.
(2017). E, in the experimental set-up from the study by Willett etal. (2021), the patient imagines writing a given letter using
his hand and a pen when he sees a ‘GO’ signal, while an implanted microelectrode array (MEA) records electrical activity of
a region of the motor cortex. F, principal components analysis reveals neural behaviour that explains most variance in the
neural dynamics for the letters ‘d’, ‘e’ and ‘m’ over 27 repetitions. G, these data are used to calculate computer reconstructions
of the written letters, which are then processed further to type words and sentences on a computer screen. E–G, used with
permission from Willett etal. (2021). H, the experimental designs for both the animat and MEART. The animat is controlled
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

10 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
changes in synaptic weight. However, there are many
ways in which underlying neuronal circuitry might
change. Hybrots are poised to serve as a general-
purpose platform for understanding the basic rules of
engagement for neural systems, because they lack the
evolutionary complexity of complete nervous systems
that might confound interpretation, yet they still
possess goal-directed capabilities.
The many diverse viable combinations of evolved
and engineered components reveal a rich medium in
which to understand truly general principles of self-
organization of structure and function in unexpected
circumstances. The ability to make novel living beings
(Nanos & Levin, 2022) is significant in several ways. It
frees us from the contingencies and frozen accidents
of the trajectory of evolution on Earth and shows us
life as it can be, enabling us to expand our thinking
and generalize to learn lessons of biology in its general
form. It allows us to probe the intelligence of life at
all levels by exploring the degree of goal-directed
behaviour, competency and failure modes of living
modules in highly diverse environments. However,
novelty is not only about new external circumstances.
Some of the most informative aspects of perturbation
are changes of inner composition; changes to the
parts of the living system itself that are enabled by
chimaerism and bioengineering. The introduction
of synthetic DNA constructs, nanomaterials and
other components into living systems reveals how
multicellular collectives handle rapid changes in the
properties of their parts. Response to a wide variety of
new challenges is the cornerstone of the study of the
intelligence of any system, including that of evolution
(generalizing adaptive behaviour beyond previously
encountered challenges).
A hallmark of chimaeric and bioengineered
organisms is that they are functional; they operate in
physiological and behavioural spaces and often exhibit
the same robustness as evolved beings with respect
to being able to reach specific states or behaviours
despite perturbations. Figure 5 shows an example
in which the traditional brain–body–environment
schema is maintained while setting all of its individual
components to be as unusual as possible. The study of
this class of systems provides ways to uncover novel
goal-directed activities that are masked by default
developmental constraints and standard ecological
scenarios. But most of all, the diversity of composition
and provenance in this rich space of possible life forms
requires us to look for a deep invariant: a parameter
that can help us to recognize, compare and relate to
intelligences in novel embodiments, when the familiar
phylogenetic tree offers no convenient classification.
BASAL COGNITION AND DIVERSE
INTELLIGENCES: TELEONOMY AS
ACOMMONTHREAD
Intelligence is the ability to reach the same goal
through differentmeans.
William James
One important promise of synthetic life forms concerns
what they can tell us about the relationship of mind and
body (i.e. cognition in novel media). Importantly, the
ability to solve problems in various spaces, with diverse
degrees of competency, is known to extend well beyond
brains. The emerging field of basal cognition seeks to
understand the roots of intelligence in ancient, pre-
neural forms and unfamiliar guises (Jennings, 1906;
Lyon, 2006; Balazsi etal., 2011; Keijzer etal., 2013;
Lyon, 2015; Baluška & Levin, 2016; Baluška etal.,
2021). Recently, it has been argued that the origins of
problem-solving in novel circumstances (behavioural
intelligence) lie in ancient capacities that long pre-
date central nervous system development (Fields
etal., 2020). The implications of this view include the
idea that the tools of neuroscience can be applied far
beyond neurons, to understand how all types of cells
join into collectives that work to achieve large-scale
objectives (Levin, 2019, 2022). Importantly however,
we must look beyond behaviour in three-dimensional
space (movement) as the arena in which intelligence
can be observed. Life solves problems in many
different spaces, and chimaeric organisms help us to
widen our criteria and begin to recognize intelligence
in unexpected guises. The ability to rearrange neural
structures in novel configurations also reminds us that
as collections of neurons, beings with brains are, in an
important sense, also collective intelligences.
Morphogenesis by cell groups is a natural example
that is only now beginning to be understood as a
collective intelligence problem and to be investigated
using the same tools and conceptual paradigms as
those used in neuroscience (Fig. 6 ). For example,
by a two-dimensional neural culture in an MEA, and behaviour exists solely in a virtual world. For MEART, a two-dimensional
culture controls the arms of MEART by controlling air compressors for hydraulic motion. Adapted from DeMarse etal.
(2001) and Bakkum etal. (2007b). I, data showing that learning occurs in both systems but is stronger in the animat with
adaptive training. Circles show degrees of movement, with colours indicating probability of a given direction of movement
after training. Black arcs indicate the trained direction. Used with permission from Bakkum etal. (2007b).
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

ENGINEERING NOVEL BODIES AND NOVEL MINDS 11
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
Figure 5. An extreme example of an unconventional agent. To get a real feel for the novel kinds of agents that are possible,
imagine taking the traditional brain–body–environment schema (A) and making its modular components truly diverse.
B, the brain of this new agent is a cultured mammalian brain chimaerized with insect neurons. It acts in the real world
because its output neuronal activity is detected by electronic interfaces and used to control a body. B′, the body consists of
a robotic swarm acting in an arena, in which they can pick up glucose and other fuels and deliver those to the bioreactor
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

12 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
tadpoles with craniofacial organs placed in abnormal
positions can become frogs with normal faces,
because the components of the head will move around
as needed to produce a correct face (Vandenberg
etal., 2012; Pinet etal., 2019). This ability to find
the right region of morphospace corresponding to
a normal frog anatomy, despite components of that
anatomy starting off in the wrong position, reveals an
important fact visible in many examples of regulative
development, regeneration and remodelling (such
as those described in the Introduction). Evolution
does not simply make hardwired machines that
execute a predetermined set of steps (such as the
default movements of tadpole eyes and jaws during
metamorphosis). Instead, it produces hardware that
can execute error minimization, traversing novel
paths in morphological and transcriptional spaces
(Elgart etal., 2015; Schreier etal., 2017; Emmons-
Bell etal., 2019) to achieve their target morphologies.
Models of morphogenesis as collective intelligence
are driving empirical work to understand what such
systems measure, how they store set points for their
homeostatic activity, how these set points arise and
how they can be edited by biomedical approaches
(Mathews & Levin, 2018). They also suggest a
research programme to understand how homeostatic
loops scale and pivot across problem spaces during
evolution (Levin, 2019, 2022).
A focus on the actions of systems in arbitrary spaces,
rather than on their anatomical or molecular–genetic
composition, requires us to identify invariants that
can serve as a parameter by which to organize and
compare highly diverse types of agents in unfamiliar
embodiments. We suggest that one important and
interesting thing that all agents, no matter their
composition or origin, have in common is goal-directed
behaviour (at some level of competency) (Rosenblueth
etal., 1943).
Goal-directed behaviour is, at the very least,
uncontroversial in human animals. It is thought that
this capacity is enabled by collectives of neurons
(brains) exhibiting memory, error minimization
capacity and second-order metacognition that
enables us to think about those goals (and perhaps
re-set them) in addition to executing them. However,
brains evolved from much more ancient bioelectric
networks that are formed by all cells in the body and
are as old as bacterial biofilms (Prindle etal., 2015;
Fields etal., 2020; Yang etal., 2020). These networks
readily form circuits with memory that enables basal
homeostatic function (Pietak & Levin, 2017; Cervera
etal., 2018, 2019, 2020). The remarkable capacity to
exhibit both robustness and novelty in morphogenesis
reveals the central role of the scaling of goals as an
explanatory, facilitating concept for new basic research
and biomedical applications (Levin, 2019). What is
essential is to understand and tame the gradual
changes in information processing that occur in the
slow transition from egg cell to complex goal-driven
cognitive agent with intrinsic purposiveness.
Cybernetic approaches are substrate independent
and remind us that no specific materials (cytoplasm,
neurons, etc.) are required for the key capacity of proto-
cognitive systems (and perhaps all life): homeostatic
and allostatic loops that expend energy to attain
specific preferred states with respect to information,
prediction error, metabolic conditions and anatomical
configurations (Allen & Friston, 2018; Constant etal.,
2018; Badcock etal., 2019; Ramstead et al., 2019).
This view emphasizes a central invariant that unifies
all attempts to predict, control, recognize, create and
relate to such systems: teleonomy (i.e. goal-directed
behaviour that achieves specific observable states, by
different means and with varying degrees of reliability
and competency; Rosenblueth etal., 1943; Bertalanffy,
1951; Varela & Maturana, 1972; Varela etal., 1974,
to feed the brain. The collective needs to work together to power themselves and the central controller. C, importantly, the
environment is not only the arena, but also includes other sentient agents. The arena is watched over by an audience of
human observers, who express their degree of approval of the antics of the robots (or explicitly pay for additional glucose)
via real-time social media posts on their networked hand-held devices. An artificial intelligence (AI) language processor
scrapes the social media posts, converts the text into specific tokens and feeds it to the brain as input to its sensory neurons.
Much like our cells, the robots also have a degree of their own on-board AI, and the behaviour of the whole system is a very
complex interplay of input, learning, noise, unreliability of components (including those of the observers), etc. This thought
experiment is designed to shock us out of our typical assumptions about what a functional brain, body and environment
must look like, in order to illustrate the immense variety of different implementations of the central components of an
environmentally embedded cognitive agent. All the individual pieces of this construct are possible with current science
(Ebrahimkhani & Levin, 2021; Nanos & Levin, 2022; Pio-Lopez, 2021). How can we understand, predict, communicate with
and relate to such alien creatures? New bioengineering technologies are leading to an inevitable one-way journey out of a
biology limited to a discrete set of forms that happened to have evolved on one planet. When origin story and composition
cease to be good guides to how one should relate to a system, teleonomic capacity is one central concept that will survive the
coming decades and drive advances in next-generation research and ethics. B taken with permission from Ebrahimkhani &
Levin (2021). A, B’, C images courtesy of Jeremy Guay of Peregrine Creative.
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

ENGINEERING NOVEL BODIES AND NOVEL MINDS 13
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
Figure 6. Mechanisms of cognition outside the brain. A, the schematic diagram of conventional cognition can be represented
by a software–hardware duality. Anetwork of neurons (the hardware) enables real-time dynamics of computation via a set
of electrical processes (the software). The result of these dynamics is a set of instructions to muscles, in order to move the
creature through three-dimensional space; what we recognize as ‘behaviour’, often with some degree of problem-solving
ability. The commitment of neuroscience is that the information content of the collective intelligence of neurons can be read
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

14 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
1991; Maturana & Varela, 1980; Lander, 2004;
McShea, 2016; Turner, 2017). This independence from
implementation details, and a willingness to recognize
navigation towards preferred states in any problem
space, removes traditional but limiting opacities from
the lens through which we view ‘agents’ that exhibit
teleonomic behaviour; such self-imposed filters, like
a type of mind-blindness, have restricted research,
because our perceptual systems are tuned to recognize
only familiar types of agency (i.e. that of medium-sized
beings navigating three-dimensionalspace).
Teleonomy here is proposed as a conceptual tool
to drive the creation and analysis of novel synthetic
beings. Rather than (as is sometimes claimed) being a
tool to distinguish living beings from ‘mere machines’,
it provides a unifying framework to understand the
whole multidimensional spectrum of possible agents.
Cybernetics (Rosenblueth etal., 1943) is an ideal
framework for life as it can be. The dissolution of
shallow, contingent boundaries between evolved and
engineered systems was foreseen long ago in the title
of Wiener’s foundational work, ‘Cybernetics, or control
and communication in the animal and the machine’
(Wiener, 1961). It gives us a mature framework for
understanding goal-directed behaviour without resort
to mysterianism; dynamical systems and control
theories offer rigorous formalisms for understanding
causes of long-term behaviour as types of navigation
policies in problem space (Pfeifer etal., 2007). Attractors
(regions of the state space not occupied by the system
at a given moment) serve as empirically useful views
of causes of system-level behaviour as it navigates the
space, guided by its non-local topology (Manicka &
Levin, 2019). In keeping with the practical engineering
focus of our approach, we eschew questions of teleology
(philosophical wrangling over real, objective purpose
or agency) and focus on goal-seeking and problem-
solving behaviour that is apparent, or relative, to an
observer (usually, scientists, but also to the system
itself, in the case of sufficiently sophisticated agents).
This approach has already been emphasized by key
early figures in cybernetics: ‘organization is partly in
the eye of the beholder’ (Ashby, 1952). Thus, our ability
to detect, understand and manage teleonomy says as
much about our own intelligence as it does about the
system being studied (Sims & Pezzulo, 2021).
We propose a specific hypothesis about the
fundamental origin of the plasticity, robustness and
intelligence in diverse embodiments: that life exploits
a multi-scale competency architecture that allows the
products of evolution to thrive in the face of novelty. It
is obvious that biological systems are hierarchical in
terms of structure. More recently, studies of molecular
genetics and developmental biology have revealed the
modular nature of functions in the body. What is only
now beginning to be appreciated is that the subsystems
making up living bodies exhibit multi-scale goal
directedness: each subunit has an agenda (a goal of
some scale) in its various spaces. Classical workers
in developmental biology recognized this ‘struggle of
the parts’ (Heams, 2012) and their ability to reach
goals (coarse-graining of molecular microstates into
meaningful anatomical set points) by different means.
The ability of parts to cooperate and compete (Gawne
etal., 2020) within and across levels of scale and
organization has fascinating implications. We next
consider the implications of multi-scale competency
for natural evolution and for synthetic bioengineering
and robotics.
MULTI-SCALE COMPETENCY POTENTIATES
EVOLUTION
Expanding our capacity for bioengineering novel
life forms towards complex outcomes requires
learning to work with agential materials; that is, not
micromanaging outcomes at every level, but guiding
self-assembly by inducing components with agendas
and competencies to change their default behaviours
(guided self-assembly). Evolution learned to work
in this medium first (Vane-Wright, 2014; Watson
& Szathmary, 2016), and we can benefit greatly
from understanding how it manages (and benefits
out by a process of neural decoding. B, evolution discovered the importance of electrical networks for information processing
and binding into collective selves long before brains appeared. An ancient function for bioelectric networks existed in all
somatic cells, which work in a manner isomorphic to that of familiar neuroscience content. All cells have ion channels and
gap junctions, resulting in bioelectric dynamics that also solve problems; they control cell behaviours (such as differentiation
and migration) in anatomical morphospace. Like the brain, they are subject to reprogramming by stimuli or experiences
and they carry out numerous goal-directed activities that serve the collective. Current research has ported many of the
same tools as those used for neural decoding to understand the morphogenetic code implemented in the somatic bioelectric
medium. Functional tools cannot tell the difference between neurons and non-neural cells, because they are fundamentally
similar in their ability to execute homeostatic loops and scale up to larger agents via bioelectrical coupling. Images in panels
A,B courtesy of Jeremy Guay of Peregrine Creative, used with permission from Levin, M. Life, death and self: Fundamental
questions of primitive cognition viewed through the lens of body plasticity and synthetic organisms. Biochem Biophys Res
Commun 2021, 564, 114-133.
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

ENGINEERING NOVEL BODIES AND NOVEL MINDS 15
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
from) a multi-scale architecture. All the engineering
accomplishments listed in Table 1 benefitted from
the fact that scientists did not have to micromanage
(create from scratch) every part and its activity;
instead, they recombined biological systems at chosen
levels and depended on reliable components to exert
their competencies in new circumstances.
The standard picture of evolution is that of an
undirected search of morphospace. The problem with
the much more efficient ‘Lamarckian’ algorithm is not
the contingent and porous Weismann barrier between
germline and soma (Jablonka etal., 1998; Anava
etal., 2014); instead, it is the much deeper issue of
the inverse problem (Lobo etal., 2014): because the
relationship between genome and anatomy is highly
indirect, it is, in general, very hard to compute which
changes in DNA should be made to canalize any desired
bodily change (such as a neck stretching after taller
branches). The same credit assignment problem is at
the heart of efforts in machine learning: how to modify
the subunits (and which ones) to reduce error, given
feedback from the environment? Both these issues
are strongly impacted by a multi-scale competency
architecture. One instructive example is the fact that
when eye primordia are grafted onto tadpole tails and
the primary eyes are removed, the resulting animals
can still see (perform visual learning tasks and
behaviours). The primordial cells still make a proper
eye even in a novel environment (in the midst of muscle
tissue); the eyes put out an optic nerve that connects
to the spinal cord (a novel target instead of the brain),
and the brain properly interprets data arriving by this
novel route and folds it into the behavioural repertoire
of the animals (Blackiston & Levin, 2013). Likewise,
experimental introduction of additional bones in
vertebrate limbs results in adjustments of ligaments,
muscles, motor neurons, etc. to enable a functional limb
that balances mechanical load correctly (Hallgrímsson
& Hall, 2011; Sultan etal., 2022).
These examples readily illustrate the implications
for evolution, because macro-scale changes are readily
accommodated: it is much easier to explore a fitness
landscape when the goal-directed competency of the
parts can be relied upon not to wreck the adaptive
character of the body when things change. When the
parts themselves are goal-directed agents, evolution is
greatly accelerated. Indeed, teleonomy is at the core
of reliability (which enables a searchable, continuous
fitness space). Complex systems can persist and be
improved and built upon continuously because the goal
directedness of their parts enables other parts (and the
collective) to trust that they will accomplish their task
(making it practical to invest energy in activity and
architectures that rely upon them working properly).
This begins to blur distinctions, and emphasizes
commonalities, between the processes implemented by
evolution and rational engineers (Kauffman, 1971).
Owing to these dynamics, multi-scale competency
increases the apparent intelligence quotient (IQ) of the
evolutionary process (Watson etal., 2014, 2016; Watson
& Szathmary, 2016; Szilagyi etal., 2020; Czégel etal.,
2022). It is no longer as short-sighted in fitness space
(although it is undirected in genotype space), because
the competency of the parts enables the specific moves
it makes to be much better than they otherwise would
be. Teleonomic robustness in various spaces enables
competency in fitness space without directed mutation
(which the inverse problem makes extremely difficult,
in any case). This not only accelerates the search,
but also changes the very nature of what is evolved.
Evolution does not produce specific solutions to specific
environmental problems; instead, it produces problem-
solving machines that can handle novelty. The nature
of evolution as a learning process (Watson etal., 2014,
2016; Power etal., 2015; Watson & Szathmary, 2016;
Czégel etal., 2019; Szilagyi etal., 2020) is transformed
by the teleonomy. The competency of the parts
enables better generalization during evolutionary-
scale learning; it does not only learn one way of
being a successful organism, it effectively learns a
class of feasible organisms, because the competition
and cooperation of goal-directed components (Gawne
etal., 2020) map many different starting points to a
functional anatomy.
Over time, a lineage learns not only how to exploit
a specific niche successfully, but also how to manage
successfully in novel circumstances (changes of
the environment and of its own parts, including
mutations and other perturbations, as in the many
examples above). Moreover, by working through an
indirect, complex intermediate layer (development
and physiology), evolution is forced to generalize
(which would not happen with direct encodings).
This enables evolution to give rise to problem-solving
machines with generalization to novel circumstances.
This is becoming strikingly apparent in the advent of
biorobotics with emergent properties.
EVOLUTION HAS ALREADY LEARNED
TO GENERALIZE BEYOND DEFAULT
MORPHOLOGIES
The robustness of life under perturbation of the
external environment and of internal components
reveals the essence of teleonomic systems: competency
in pursuing goals. Where do the specific goal states
come from? What determines the set points towards
which a given system will expend energy and effort?
When we observe that cells of a given species work to
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

16 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
implement a specific structural and functional outcome
reliably, we generally say that these are explained by
evolution. The goals towards which cellular collectives
navigate competently in various spaces traditionally
have had two explanations: direct selection ‘forces’
and side-effects (spandrels) of other features that have
been selected (Gould & Lewontin, 1979; Kull, 2014).
This might, perhaps, be reasonable for the standard
class of biological systems evolved in the phylogenetic
tree on Earth. However, the limits of this paradigm
are revealed when we expand traditional epigenetics
to include genuinely novel configurations that have
never existed before onEarth.
Questions about selves, autonomy, plasticity and
the origin of biological novelty led us to ask what
would happen if skin cells were removed from a frog
embryo, dissociated and given a chance to reboot
their multicellularity in vitro (Blackiston etal., 2021).
Many outcomes are possible a priori: they could have
spread out or died or formed a monolayer, etc. Instead
(Fig. 7), what they did was to reassemble and form a
novel proto-organism known as a Xenobot (Kriegman
etal., 2020). These spherical constructs move through
water by the coherent action of cilia, exhibiting a
variety of self-actuated types of motility. They have a
developmental sequence of novel forms that are unlike
the typical Xenopus stages; they repair after damage,
interact with their environment and show spontaneous
changes in behaviour. These novel morphologies and
behaviours do not require transgenes or genomic
editing; Xenobots repurpose their native hardware (e.g.
cilia, which are normally used to redistribute mucus)
to new functionality. Amazingly, deprived of their
normal way of reproducing, the emergent processes of
Xenobots discover kinematic self-replication (a novel
mode of reproduction not used by any other organism
on Earth, to our knowledge), which they implement by
herding loose collections of cells in their environment
together to form the next generation of Xenobots
(Kriegman etal., 2021). Nothing has been added
to their completely wild-type frog genome; instead,
developmental constraints have been removed.
Without the normal instructions from the rest of the
body telling these skin cells to form a passive, two-
dimensional boundary layer to keep out the bacteria (a
system of low agency), the true capacities of this cellular
collective are revealed; it forms a three-dimensional
individual with a more exciting life of self-initiated
motile behaviour. The collective intelligence of these
cells is revealed as, despite a novel environment and
novel internal configuration that never existed in the
frog evolutionary lineage, they discover novel ways to
be a coherent organism. Their option space is normally
distorted by the larger collective, but their default
geodesic through option space (and their baseline
preferences in morphospace) are revealed when these
influences are removed. All this self-assembly and
emergent organization takes place in 48h and does
not require aeons of evolutionary forces to become a
good Xenobot.
If the answer to ‘Where do a frog’s shape and
behaviour come from?’ is ‘Long periods of selection
and interaction with the environment that sculpt
the genome to be a great frog’, then where do the
anatomical and behavioural goals of Xenobots
originate? Their anatomical and behavioural goals are
emergent (Veloz, 2021), rather than directly selected
for over aeons of sculpting by selection. Anumber of
researchers have emphasized information arising from
generic laws of form (Beloussov & Grabovsky, 2007;
Beloussov, 2008; Newman, 2014, 2019; Zhang etal.,
2021), from mathematics (Brigandt, 2013; Lange, 2013;
Green & Batterman, 2017; Reutlinger, 2017) and from
environmentally initiated novelties (West-Eberhard,
1998, 2005a, b; Shapiro, 2022). These Xenobots are
only the beginning of a large class of beings that
challenge us to develop a better understanding of how
goal states arise in novel contexts and how evolution
exploits the laws of physics and computation in the
context of teleonomic processes.
A key area of research concerns how goal dir-
ectedness is scaled during evolution, enabling beings
with increasingly large cognitive horizons with
respect to the goals they can pursue (Levin, 2019).
Bioengineering is a powerful means of studying this
aspect of collective intelligence, because it allows us to
manipulate which components (with what goals) are
connected together and precisely how their different
modules can interact in the swarm. One set of models
concerns the coupling of subunits with respect to the
three components of a homeostatic loop: sensing state,
storing a memory of the set point, and taking actions
to reduce error. This is beginning to be explored in
robotics and machine learning; in one example, robots
can adapt to new situations because their components
are behaving homeostatically (Di Paolo, 2000).
Amulti-scale competency architecture, as exploited
by chimaeric techniques, is not only an interesting
path forward (biological inspiration) for artificial
intelligence, but is also helping to dissolve artificial
barriers across fields.
CHIMAERISM HELPS TO DISSOLVE
OUTDATED CONCEPTS
Chimaerism is a type of conceptual universal acid,
dissolving existing terminology that is not based on
deep concepts but is instead a relic of parochial contin-
gencies of our familiar forms (Bongard & Levin, 2021).
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

ENGINEERING NOVEL BODIES AND NOVEL MINDS 17
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
Figure 7. Xenobots as a tractable system for understanding novel teleonomic properties. A, epidermal cells from a frog
embryo can be removed and placed in a novel environment, where they are freed from the instructive influences of the
collective, which normally force them into a quiescent, two-dimensional existence. B–G, they become Xenobots (B), swimming
by rowing cilia against the medium and exhibiting a variety of behaviours, such as those tracked in C and the circling,
maze-traversal and tube-traversal behaviours shown in D–G. H, remarkably, they even find a way to replicate, using their
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

18 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
The ability to mix any two biological systems, in any
arbitrary proportion, reveals the continuous nature of
terms that are often thought to be binary (Boldt,2018).
This strongly extends the gradual view, already
required when taking evolution and developmental
biology seriously, which lends no support to any sharp
line between ‘true cognition’ in complex brains and
‘just physics’ that is sometimes said to occur in their
phylogenetic and ontogenic precursors.
As we combine living material with electronics
and computer software (Table 1), sharp distinctions
between ‘machine’ and ‘organism’ become untenable.
There is no principled way to draw a clean line, and any
combination (from a being with 95% human brain + 5%
smart implant to one that is 5% human brain tissue in
a 95% robotic body, and every proportion in between)
is a viable being. Rather than providing a rigorous way
to distinguish life from ‘mere machines’ (as many have
argued), teleonomy emerges as a much more valuable
scientific tool: a parameter for defining a continuum
of beings on which can be built a deep framework for
understanding agency in diverse implementations. The
same is true for terms such as evolved vs. designed; by
themselves, these terms provide no powerful insight
into the nature, capabilities or moral worth of any
agent. Evolutionary algorithms are increasingly being
used by engineers to design new constructs, while
we ourselves are the products of evolution (and our
engineering is thus a secondary effect of operations of
evolved systems).
There is an important lesson in the remarkable fact
that cells and engineered materials can work together
so seamlessly: life itself does not respect the distinction
between the two, which persists only because of our
historical limitations (which are now increasingly
being lifted by progress in bioengineering). The life
vs. machine distinction is relevant only to the lowest
class of machines available in prior decades and is not
fundamental. The dichotomy is seen to be fallacious
in several ways. What is essential about machines
is not what they are made of (metal vs. cytoplasm),
but that they are systems that operate according to
predictable rules and thus can be manipulated. If this
were not true, evolution would not work. Evolution
does not build upon a blank canvas; instead, it alters
signals given to biological components that already
have a background behaviour (e.g. new signalling
factors to change the way in which cells move or
differentiate during development). This process is
greatly potentiated by the fact that each component is
a machine in the important functional sense of being
controllable and reliable for specific outcomes.
The modules of living things (from organs to
molecular networks) constitute a sophisticated class
of machines, which have homeostatic and allostatic
loops. This means that both evolution and engineering
can be potentiated by the same property: competency
to get a job done even when the environment and
composition change (within limits, of course). Life
and engineering are interoperable precisely because
of teleonomy. Each subsystem has evolved to exploit
the physics of internal and external components
without knowing in advance what the situation will
be. The multi-scale competency architecture is highly
opportunistic, because each goal-driven subunit
has no access to the reality of its environment, only
to inputs it receives, and must construct a model of
what to do, on the fly (Friston, 2013; Constant etal.,
2018). This means that, as with the new class of robots
with no prior fixed map of their structure (Bongard
etal., 2006), they will exert adaptive function even
in novel circumstances. To living creatures arising
from a single cell, all scenarios (whether ‘natural’ or
‘artificial’) are new circumstances, in which they learn
to survive.
Already today, human bodies are augmented with
prosthetics that allow them to control assistive
devices, use novel and additional effectors, such as
third limbs (Penaloza & Nishio, 2018), and make use
of new inputs in sensory substitution (Danilov & Tyler,
2005; Nagel etal., 2005; Ptito etal., 2005). These drive
changes in cognition (Kieliba etal., 2021), illustrating
the deep plasticity of life. Teleonomic functions enable
the immediate, efficient use of these evolutionarily
novel configurations because life does not manage
outcomes bottom up: the implementation details are
black boxed and do not break control systems when
each level can focus on providing top-down guidance,
letting the underlying modules achieve the goals with
existing mechanisms.
motions to collect nearby loose cells into piles that assemble into the next generation of Xenobots, and the cycle repeats.
I–I″, the same genome produces a machine that, in normal circumstances, becomes frog embryo with stereotypical stages
(I, I′), but can also produce a very different machine which makes a coherent proto-organism that re-uses its genomically
specified hardware for a novel developmental sequence (I″) and behaviours that arise spontaneously in 48h, rather than
over aeons of shaping by selection. Images by Douglas Blackiston, Levin lab (except for C, produced by Simon Garnier
and used with his permission). D-G, I’ taken with permission from Blackiston etal. (2021). H taken with permission from
Kriegman etal. (2021). Frog embryo stages sourced from Niewkoop and Faber (1994), hosted by Xenbase (www.xenbase.org
RRID:SCR_003280).
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

ENGINEERING NOVEL BODIES AND NOVEL MINDS 19
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
Of course, the plasticity and interoperability of living
things has been exploited by evolutionary arms races,
enabling some biological systems to hijack others,
such as zombie ants produced by a fungalcontroller
(Hughes etal., 2011, 2016; deBekkeretal., 2015;
Steinkraus et al., 2017; Elya etal., 2018) and
commensal bacteria being able to dictate the number
of heads in regenerating flatworms (Williams
etal., 2020). Importantly, the plasticity revealed by
chimaerism (and older work on sensory augmentation;
Bach-y-Rita etal., 1969) also reminds us that modern
humans do not represent the upper limit of goal-driven
cognition. The charge of ‘anthropomorphism’ is thus a
pre-scientific world view, in which human beings had
some sort of magic that was unique and did not exist,
even in weaker forms, anywhere else. The progressive
augmentation, diversification, biomedical engineering
and instrumentization of human brains will strongly
emphasize the limitations of the notion of absolute,
discrete natural kinds (e.g. species) and the need to
look for what is essential or deep about such categories.
It is clear now that terminology and distinctions
based on contingencies of material and origin story
will not survive the coming decades of chimaeric
technologies that will erase boundaries and extend
capabilities. We propose that the best conceptual
framework to take their place is one based on
understanding what is essential about agents: their
degree of competency in pursuing goals, and the scale
of goals they are capable of pursuing. This cybernetic
view, together with bioengineering tools, provides
researchers and philosophers with an enormous
option space of possible beings, in which the whole
panoply of evolved natural forms is only a small subset
(Fig. 2). All of Darwin’s ‘endless forms most beautiful’
exist in a small region within the space of viable
configurations. The field of evo-devo can expand to
bio-robo, and synthetic bioengineering is moving from
synthetic biology working in chemical and metabolic
spaces (Ollé-Vila etal., 2016) to synthetic morphology
working in anatomical and behavioural spaces, in
order to explore the variety of possible bodies and
minds (Sloman, 1984; Yampolskiy, 2015).
WHAT IS NEXT: EXPANDING THE OPTION
SPACE OF BEINGS DRIVES ARESEARCH
PROGRAMME
Deriving the rules of emergent morphogenesis and
behaviour using only the examples of natural species
on Earth is like testing a hypothesis on the same set
of data that generated it; chimaeric and bioengineered
beings give us the opportunity truly to evaluate the
quality of future models of robustness and plasticity
with respect to anatomical goal states. Moreover,
the interoperability of life at all levels (i.e. the fact
that chimaeras and hybrids are viable) reveals that
there are no natural kinds with respect to species
(Devitt, 2010; Austin, 2017). Natural life forms do not
represent privileged, perfect outcomes. The general
distaste for novel beings seen in science fiction (e.g.
Wells’ ‘The Island of Dr. Moreau’) is a hold-over
from a pre-Darwinian essentialism that neglects
the fact that natural species simply exemplify a
subset of possibilities for staying represented in the
biosphere, possibilities that have been found thus far
by a meandering, randomly driven search process
(evolution) that optimizes for nothing more intelligent
than biomass. It is very likely that rational design
(bioengineering) can do better than this, once we
understand the collective intelligence of molecular
pathways, cells and tissues and learn to guide their
teleonomic activity.
The option space of beings enables us truly to see
life for the first time, in the absence of standard
phylogenetic relationships that enlighten some
aspects but obscure many others. Akey aspect that
is revealed by this way of viewing life is the multi-
scale competency; the basal intelligence of many levels
of organization in a given body, all of which exhibit
teleonomic behaviour in their own problem space.
Recognizing this behaviour, and the specific goals
being pursued, is an IQ test that scientists (observers)
take when evaluating the agency of unconventional
agents; our ability to detect, understand and manage
teleonomy says as much about our intelligence as it
does about the system being studied.
The emerging field at the intersection of synthetic
developmental biology, computer science and cognitive
science implies numerous opportunities for next steps
and further progress. From the perspective of theory/
conceptual advances and specific research directions,
the following questions need to be developed:
• What is an effective eigenspace for modelling
agency? What would be the minimal axes for the
space of all possible teleonomic agents? And how
do we recognize, quantify and compare teleonomic
agents in radially diverse embodiments? Even
gene-regulatory networks, a paradigmatic case of
deterministic genetic hardware, appear to have
learning capacity (Watson et al., 2010; Szabó
etal., 2012; Gabalda-Sagarra etal., 2018; Herrera-
Delgado et al., 2018; Biswas et al., 2021), and
teleonomy has been explored in signal transduction
processes in plants (Gilroy & Trewavas, 2022). It
is imperative that we abandon the tendency for
armchair pronouncements of what can and cannot
be seen as cognitive and instead develop toolkits
for generating and testing teleonomic models of
arbitrary systems.
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

20 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
• If evolution is blind and always prefers immediate
fitness payoffs, how is it that it not only gives rise to
creatures highly adapted for specific environments,
but also evolves hardware that can problem solve
in numerous novel configurations never seen
before? How does evolution capitalize on the laws
of physics and computation to generalize so well
from specific examples to highly diverse possible
instantiations? Specifically, how does teleonomy
potentiate the evolutionary search for yet larger
teleonomic systems, and how can engineers do the
same?
• Can the notion of external environment (Umwelt;
von Uexküll, 2010) be extended to a multi-scale
concept, in which adjacent cells, tissues, etc. are
each other’s environment? Can molecular pathways
and biophysical dynamics be thought of as
affordances for systems to compete and cooperate
within and across levels in the organism (Queller
& Strassmann, 2009; Gawne etal., 2020)? Links to
existing thought on the extended organism (Turner,
2000) and extended mind (Clark & Chalmers, 1998)
are within reach.
• What is the relationship or overlap between the
sets demarcated by ‘life’ and ‘cognition’? If all
(most?) components in living things are teleonomic
agents and are thus somewhere on the continuum
of cognition (Fig. 1), are all living things cognitive?
What is a useful definition of ‘life’, given that
teleonomic agents can be produced by engineering
with organic or inorganic parts? Although modern
life is necessarily teleonomic (in order to survive
in the biosphere), could there have been very early
life forms that were not teleonomic? Could current
efforts at truly minimal synthetic life (Hanczyc etal.,
2011; Cejkova et al., 2017) clarify the relationship
between teleonomy and physics?
• How can we develop a semiotics of synthetic agents
(Tsuda etal., 2009) and their Umwelten (Manicka
& Harvey, 2008; Beer, 2014) to gain a better
understanding of the ways in which teleonomy
provides the ratchet that drives the great transitions
of cognitive capacity along the continuum? How can
we develop ways of manipulating biological systems
(such as morphogenesis) via communication and
motivation, rather than micromanagement?
• Can we develop artificial neural networks whose
output is not specific solutions to specific problems,
but rather seeds that determine machines that
solve problems (Moore et al., 2018)? Can we
imitate the learning and generalization capacity
of evolution in our machine learning architectures,
such that each level of the output specifies a
flexible design for an agent that functions in the
problem space?
• Can we make new autonomous robotics via a multi-
scale competency architecture (a generalization of
the subsumption architecture; Brooks, 1986), where
each layer is an agent that constructs models of
itself and its environment? Robots made of parts
having agency (teleonomy all the way down) will, for
the first time, become vulnerable to the occasional
defections of cancer, but will gain the flexibility and
robustness of life owing to the massive adaptability
to novelty that results from dynamic cooperation
and competition of goal-driven sub-agents.
TOWARDS A NEW ETHICS BASED ON DEEP
AGENCY IN THE OPTION SPACE OF LIFE
There is an aspect of teleonomy that has a positive
impact on the ethics of chimaeric and bioengineered
technologies. Making changes at the lower levels
(e.g. genomic editing) tends to result in system-
level outcomes that are hard to predict (Lobo etal.,
2014). When one is forced to manage all the outcomes
directly (as when manipulating a system bottom up),
a likely outcome is ‘unhappy monsters’ (i.e. ethically
unacceptable creatures containing mismatched
components that do not work harmoniously together).
Asimple example occurs in planarians: inducing
secondary heads with biochemical inducers makes
heads that are not properly scaled to the rest of the
organism, whereas nudging the top-level ‘head vs.
tail’ decision in the native bioelectric circuit leads
to all of the downstream properties being handled
by the system itself and results in perfectly scaled,
functional heads (Durant etal., 2019). In general,
changes introduced at the micro-level tend to wreck
complex systems more than changes that are input at
the higher level; optimizing the level of intervention
for specific biomedical and synthetic purposes will
require the incorporation of teleonomic models into
the biophysics frameworks used almost exclusively
today (despite the work on modularity in evolution;
Payne etal., 2014; Watson etal., 2014). Controlling
biology top down, via experiences and stimuli that
rely as much as possible on the native collective
intelligence of its parts, is likely to produce much
more coherent organisms and will provide us with a
better understanding oflife.
The impact of rapidly advancing technologies will
go far beyond science, encompassing many issues that
have been dealt with in science fiction but have not
yet been worked through by thinkers in the fields of
philosophy of mind, ethics and policy. Our culture is
in for an upheaval that will far eclipse the controversy
that fomented around Darwin’s ‘Origin of Species’.
Our ethics structures are barely sufficient to optimize
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

ENGINEERING NOVEL BODIES AND NOVEL MINDS 21
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
Figure 8. Goal directedness is an invariant for a continuous spectrum of cognition. A, biological systems are not only
structurally hierarchical but also functionally hierarchical: each layer solves unique problems in a relevant problem space,
exhibiting teleonomy. B, the degree of competency and complexity that can be handled by a system in its pursuit of goal
states defines major transitions along a continuum of cognition ranging from passive matter to advanced self-reflective
minds, which enables comparison of highly diverse intelligences (Rosenblueth etal., 1943). C, given that agency claims
are, in effect, engineering protocol claims, the search for efficient prediction and control strategies defines an ‘axis of
persuadability’, ranging from brute force micromanagement to persuasion by rational argument, C1–C4 show only a few
representative waypoints. C1, the simplest physical systems (e.g. mechanical clocks) cannot be persuaded, argued with or
rewarded/punished; only physical hardware-level ‘rewiring’ changes their behaviour. C4, on the far right are human beings
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

22 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
intraspecies interactions; they will need a total
overhaul when we are surrounded by novel agents of
every possible configuration. How does one prosecute
a 50% cyborg for its misdeeds? Legal systems based
on one dimension of human IQ (with a cut-off for
‘competent’) are in no way ready for a multidimensional
space of beings with diverse components contributing
to their mind and behaviour.
Many current legal and moral structures are built
around a crisp category of ‘human’ within that axis.
Although we have some provision for ‘diminished
capacity’ (in legal proceedings), we do not yet have
any for increased capacity. Neural real-estate is
readily taken over by functional modules (e.g. in
the cases of the blind whose visual processing areas
become used for hearing and other modalities; Van
Ackeren etal., 2018). Thus, it is very likely that the
addition of neural tissue to standard brains will
result in increased processing capacity. The system
uses a highly adaptive design that can make use of
resources that do not have to be specified in advance,
because the various subunits (cortical layers, neurons,
etc.) are themselves teleonomic agents that relate
to their neighbours with the same plasticity with
which they face external environments. Members of
the transhumanist movement have already begun to
discuss what happens when humanity includes beings
with a wide range of IQ and moral capacities (e.g. a
larger cognitive horizon that allows them functionally
to care about far more than today’s human is capable
of handling). What is a human being? Agood answer,
in future decades, cannot be based on a genetic or
anatomical description; perhaps it will be based on the
minimal amount of active compassion or functional
caring (similar to today’s standard human) that a
being can muster within its cognitive boundary (i.e. a
definition based on capacity for teleonomic action).
Moreover, although Darwin’s revolution created
a continuous spectrum along which to compare
intelligence (and place moral categories for
relationships to primates, whales, dogs, etc., in addition
to complex cases, such as octopuses), a far wider
reality is beginning to emerge. There is not only a
large axis representing different amounts of cognitive
sophistication, but also an immense space of multiple
dimensions of different types of cognitive capacity. It is
entirely unclear how creatures within that space will
relate to each other, making essential the search for an
invariant on which to base norms (such as complexity
of the creature’s goal space).
CONCLUSION
Bioengineering expands our subject of inquiry to a
now-realizable immense option space of novel living
agents: true biology in the sense of life as it can be
(Langton, 1995), rather than zoology/botany. The study
of these novel beings does two important things. First,
it helps us to move beyond the historical contingencies
and familiar (limiting) categories of where agency can
be found in the specific products of one evolutionary
stream on Earth. We must become comfortable
thinking about new scales of size and duration, new
material substrates for life and mind, and new spaces
in which goals can be sought. Second, it helps us to
get to the root of a key question: where do goals come
from in the first place? Synthetic creatures such
as chimaeras, hybrots, cyborgs and biobots provide
empirically tractable model systems in which to study
emergence of the morphological and behavioural goals
of collective intelligences (any being made of parts)
that were not shaped by aeons of evolution towards
specific environments.
(and others to be discovered; Bostrom, 2003; Kurzweil, 2005), whose behaviour can be changed radically by a communication
encoding a rational argument that changes the motivation, planning, values and commitment of the agent receiving this.
C2, C3, between these extremes lies a rich range of intermediate agents, such as simple homeostatic circuits (C2), which
have set points encoding goal states, and more complex systems, such as animals that can be controlled by training using
stimuli that communicate to the system how it can achieve its goal of receiving a reward (C3). This continuum is not a linear
scala naturae; evolution is free to move in any direction in this option space of cognitive capacity. The goal of the scientist is
to find the optimal position for a given system. Too far to the right results in complex models that do not improve prediction
and control. Too far to the left and one loses the benefits of top-down control in favour of intractable micromanagement.
This is also a continuum with respect to how much knowledge one has to have about the details of the system in order to
manipulate its function. For systems in classC1, one has to know a lot about their workings to modify them. For classC2,
one has to know how to read–write the set point information but does not need to know anything about how the system
will implement those goals. For classC3, one does not have to know how the system modifies its goal encodings in light
of experience, because the system does all of this on its own; one only has to provide suitable rewards and punishments.
Ascertaining the optimal level of teleonomy in the objects around us is a key task for scientists interested in understanding
and managing novel complex systems; this capacity is also a built-in cognitive module for animals navigating complex
environments, conspecifics, prey, etc. B created after Rosenblueth etal. (1943). A, C images courtesy of Jeremy Guay of
Peregrine Creative, taken with permission from Levin (2022).
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

ENGINEERING NOVEL BODIES AND NOVEL MINDS 23
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
These new model systems dissolve outdated, binary
categories, such as ‘machine’ and ‘organism’, that were
based on the temporary technological limitations of the
past and will surely not survive the coming advances
of the next few decades (Bongard & Levin, 2021).
Fixed Linnean relationships and categories based on
genomes have served biologists well (although perhaps
not so well for microbiologists), but these need to be
replaced in light of the increasingly obvious plasticity
and interoperability of life (Fig. 8). We will soon be
surrounded by truly ‘endless forms most beautiful’
(Darwin, 1859), filling an option space that Darwin
could not even have dreamed of (Ollé-Vila etal., 2016).
They challenge us to define new categories based on
deep (empirically useful and more philosophically
sophisticated) criteria and to move beyond an
assumption that goals are always set by selection.
Future impacts of these ideas will be driven by
a progressively improved understanding of the
relationship between the crafter (whether evolution
or an engineer) and its material. For example, the
current intellectual property system was developed to
address work with classical, passive materials, where
everything is in what the craftsman did; patenting the
craftsman’s recipe is key. It is not yet suitable for work
with agential materials, for which the inventor is a
collaborator with the components. As with the Xenobots,
where the outcome is as much (or more) dependent on
the competency of cells to carry out specific goals as it is
on the human providing cues, the outcome of this new
type of creative work is partly the method, but it is also
partly what has been discovered about the competency
of the agential material. These challenges to our ideas
of intellectual ownership dovetail with similar concerns
over future inventions by AI agents that serve as tools
to augment human creativity.
A mature science of teleonomy is no more about
understanding Xenobots and their kin than computer
science is the study of our current computers. The goal is
far deeper: understanding the relationship between the
genome, the software that produces bodies and behaviour,
and the ability to reach adaptive ends despite diverse
means on very rapid time scales. Critically, the issue
of forming and detecting goals requires a specification
of a complex system to whom the goals belong. What
defines a self? Although immunological (Pradeu, 2019)
and evolutionary-scale (Strassmann & Queller, 2010)
theories have been proposed, these are intimately tied to
(and thus limited by) the types of organisms we observe
naturally in our biosphere. Amore general framework,
able to encompass and directly compare truly diverse
agents, needs to be based on teleonomy at its core: a self
is a goal-directed system, and its level of sophistication
(ranging from modest inorganic systems to trans-human
beings) is set by the spatiotemporal scale of the goals it
can pursue (Levin, 2019).
Chimaeric and synthetic bioengineering enables us
to leave the Garden of Eden of a finite set of natural
species and to continue Adam’s task of naming novel
creatures; more specifically, discovering their true
nature beyond the facts of their composition and
origin. The sciences of cybernetics and the deep lessons
of neuroscience, which extend well beyond neurons
(Friston etal., 2014; Ramstead etal., 2019; Fields etal.,
2020; Fields & Levin, 2020a, b), will be key components
of this future. At stake are transformative advances in
regenerative medicine (to get beyond the low-hanging
fruit reachable by conventional stem cell biology and
genomic editing approaches), robotics and generalAI.
Crucially, this new field suggests not only novel
capabilities and advances in knowledge, but also the
need for a new ethics. The frequently voiced statements
that ‘living things are not machines’ reflect an outdated
essentialism and a type of magical thinking that trusts
in clear, binary lines separating evolved living beings
from designed machines to define our moral duty to
various agents comfortably. These lines do not exist,
which will be made painfully clear in the next decades
as we become surrounded by collections of agents that
make the iconic Cantina scene in ‘Star Wars’ look tame
in comparison. Significant effort will need to be made as
science and society mature to include designed beings in
addition to natural beings, in order to avoid the types of
ethical lapses to which humans are prone: mistreatment
of those who do not resemble a familiar in-group in
composition or origin. The nature of teleonomy as a
guiding principle cutting across contingencies of origin
story and composition, and the inevitable expansion of
life throughout the option space of hybrid forms (Fig. 2),
provide important conceptual tools for a path forwards
to a future where we cannot simply guess the capacity
of an agent to think and suffer based on what it looks
like or how closely it resembles a familiar touchstone
species in the Earth’s phylogenetic lineage. It is not
clear what a new ethics of life as it can be will look like,
but some sort of golden rule about compassion towards
systems proportional to their teleonomic capacity might
be a place from which tostart.
ACKNOWLEDGEMENTS
We thank Julia Poirier for assistance with the
manuscript. We are grateful to Francis Heylighen,
Peter A.Corning, Richard Vane-Wright and two
anonymous reviewers for helpful comments on the
paper. We also thank Joshua Bongard, Avery Caulfield,
Anna Ciaunica, Pranab Das, Daniel Dennett, Thomas
Doctor, Bill Duane, Christopher Fields, Jacob Foster,
Karl Friston, James F.Glazebrook, Erik Hoel, Eva
Jablonka, Santosh Manicka, Noam Mizrahi, Aniruddh
Patel, Giovanni Pezzulo, Andrew Reynolds, Matthew
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

24 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
Simms, Elizaveta Solomonova, Richard Watson,
Daniel Weiskopf, Olaf Witkowski and numerous others
from the Levin Lab and the Diverse Intelligences
community for helpful conversations and discussions.
M.L. gratefully acknowledges support by the
Templeton World Charity Foundation (TWCF0606),
the John Templeton Foundation (62212) and The
Elisabeth Giauque Trust. We have no conflicts of
interest to declare.
This article is a contribution to a special issue on
Teleonomy in Living Systems, guest edited by Richard
I.Vane-Wright and Peter A.Corning, based on a
Linnean Society meeting held on 28 and 29 June 2021.
DATA AVAILABILITY
This manuscript is based on information in the
accessible published literature.
REFERENCES
AaserP, KnudsenM, RamstadOH, vandeWijdevenR,
NicheleS, SandvigI, TufteG, BauerUS, HalaasØ,
HendsethS, SandvigA, ValderhaugV. 2017. Towards
making a cyborg: a closed-loop reservoir-neuro system. In:
ECAL 2017: the 14th European conference on artificial life.
Lyon: MIT Press, 430–437.
AdamatzkyA. 2018. Towards fungal computer. Interface
Focus 8: 20180029.
AdewoleDO, Serruya MD, HarrisJP, BurrellJC,
PetrovD, ChenHI, WolfJA, CullenDK. 2016. The
evolution of neuroprosthetic interfaces. Critical Reviews in
Biomedical Engineering 44: 123–152.
AllenM, FristonKJ. 2018. From cognitivism to autopoiesis:
towards a computational framework for the embodied mind.
Synthese 195: 2459–2482.
AnavaS, PosnerR, RechaviO. 2014. The soft genome.
Worm 3: e989798.
AndersenSB, GerritsmaS, YusahKM, MayntzD, Hywel-
JonesNL, BillenJ, BoomsmaJJ, HughesDP. 2009. The
life of a dead ant: the expression of an adaptive extended
phenotype. The American Naturalist 174: 424–433.
AndersonMJ, Sullivan JG, HoriuchiT, FullerSB,
DanielTL. 2020. A bio-hybrid odor-guided autonomous
palm-sized air vehicle. Bioinspiration and Biomimetics 16:
026002.
AndoN, KanzakiR. 2020. Insect-machine hybrid robot.
Current Opinion in Insect Science 42: 61–69.
AshbyWR. 1952. Design for a brain. London: Chapman &
Hall.
AustinCJ. 2017. Aristotelian essentialism: essence in the age
of evolution. Synthese 194: 2539–2556.
AydinO, ZhangX, NuethongS, Pagan-DiazGJ, BashirR,
GazzolaM, SaifMTA. 2019. Neuromuscular actuation of
biohybrid motile bots. Proceedings of the National Academy
of Sciences of the United States of America 116: 19841–19847.
Bach-y-RitaP. 1967. Sensory plasticity. Applications to a
vision substitution system. Acta Neurologica Scandinavica
43: 417–426.
Bach-y-RitaP, CollinsCC, Saunders FA, WhiteB,
ScaddenL. 1969. Vision substitution by tactile image
projection. Nature 221: 963–964.
BadcockPB, FristonKJ, RamsteadMJD, PloegerA,
HohwyJ. 2019. The hierarchically mechanistic mind: an
evolutionary systems theory of the human brain, cognition,
and behavior. Cognitive, Affective & Behavioral Neuroscience
19: 1319–1351.
Bakkum DJ, Chao ZC, GamblenP, Ben-AryG,
Shkolnik AG, DeMarseTB, PotterSM. 2007a.
Embodying cultured networks with a robotic drawing
arm. Annual International Conference of the IEEE
Engineering in Medicine and Biology Society (EMBC)
2007: 2996–2999.
Bakkum DJ, Gamblen PM, Ben-AryG, Chao ZC,
PotterSM. 2007b. MEART: the semi-living artist. Frontiers
in Neurorobotics 1: 5.
BalazsiG, vanOudenaardenA, CollinsJJ. 2011. Cellular
decision making and biological noise: from microbes to
mammals. Cell 144: 910–925.
BaluškaF, LevinM. 2016. On having no head: cognition
throughout biological systems. Frontiers in Psychology 7:
902.
BaluškaF, MillerWB Jr, ReberAS. 2021. Biomolecular
basis of cellular consciousness via subcellular nanobrains.
International Journal of Molecular Sciences 22: 2545.
BeerRD. 2014. The cognitive domain of a glider in the game
of life. Artificial Life 20: 183–206.
deBekkerC, OhmRA, LoretoRG, SebastianA, AlbertI,
MerrowM, Brachmann A, HughesDP. 2015. Gene
expression during zombie ant biting behavior reflects
the complexity underlying fungal parasitic behavioral
manipulation. BMC Genomics 16: 620.
BeloussovLV. 2008. Mechanically based generative laws of
morphogenesis. Physical Biology 5: 015009.
BeloussovLV, GrabovskyVI. 2007. Information about a
form (on the dynamic laws of morphogenesis). BioSystems
87: 204–214.
BertalanffyL. 1951. Towards a physical theory of organic
teleology: feedback and dynamics. Human Biology 23:
346–361.
BiswasS, ManickaS, HoelE, LevinM. 2021. Gene regulatory
networks exhibit several kinds of memory: quantification of
memory in biological and random transcriptional networks.
iScience 24: 102131.
BlackistonD, Lederer E, KriegmanS, Garnier S,
BongardJ, LevinM. 2021. A cellular platform for the
development of synthetic living machines. Science Robotics
6: eabf1571.
BlackistonDJ, LevinM. 2013. Ectopic eyes outside the
head in Xenopus tadpoles provide sensory data for light-
mediated learning. The Journal of Experimental Biology 216:
1031–1040.
BoldtJ. 2018. Machine metaphors and ethics in synthetic
biology. Life Sciences, Society and Policy 14: 12.
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

ENGINEERING NOVEL BODIES AND NOVEL MINDS 25
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
BongardJ, LevinM. 2021. Living things are not (20th
century) machines: updating mechanism metaphors in light
of the modern science of machine behavior. Frontiers in
Ecology and Evolution 9: 650726.
BongardJ, ZykovV, LipsonH. 2006. Resilient machines
through continuous self-modeling. Science 314: 1118–1121.
Bonifazi P, Difato F, Massobrio P, Breschi GL,
PasqualeV, LeviT, GoldinM, BornatY, TedescoM,
BisioM, KannerS, GalronR, TessadoriJ, TavernaS,
ChiappaloneM. 2013. In vitro large-scale experimental
and theoretical studies for the realization of bi-directional
brain-prostheses. Frontiers in Neural Circuits 7: 40.
Bostrom N. 2003. Human genetic enhancements: a
transhumanist perspective. Journal of Value Inquiry 37:
493–506.
vandenBrand R, MignardotJB, von ZitzewitzJ,
LeGoff C, FumeauxN, WagnerF, CapogrossoM,
MartinMoraudE, MiceraS, SchurchB, CurtA, CardaS,
BlochJ, CourtineG. 2015. Neuroprosthetic technologies to
augment the impact of neurorehabilitation after spinal cord
injury. Annals of Physical and Rehabilitation Medicine 58:
232–237.
BraunE. 2015. The unforeseen challenge: from genotype-to-
phenotype in cell populations. Reports on Progress in Physics
78: 036602.
BrigandtI. 2013. Systems biology and the integration of
mechanistic explanation and mathematical explanation.
Studies in History and Philosophy of Biological and
Biomedical Sciences 44: 477–492.
BrooksRA. 1986. A robust layer control system for a mobile
robot. IEEE Journal of Robotics and Automation RA-2:
14–23.
BuccelliS, BornatY, ColombiI, AmbroiseM, MartinesL,
PasqualeV, BisioM, TessadoriJ, NowakP, GrassiaF,
AvernaA, TedescoM, BonifaziP, DifatoF, MassobrioP,
LeviT, ChiappaloneM. 2019. A neuromorphic prosthesis
to restore communication in neuronal networks. iScience 19:
402–414.
CejkovaJ, BannoT, HanczycMM, StepanekF. 2017.
Droplets as liquid robots. Artificial Life 23: 528–549.
CerveraJ, LevinM, MafeS. 2020. Bioelectrical coupling of
single-cell states in multicellular systems. The Journal of
Physical Chemistry Letters 11: 3234–3241.
CerveraJ, PaiVP, Levin M, MafeS. 2019. From non-
excitable single-cell to multicellular bioelectrical states
supported by ion channels and gap junction proteins:
Electrical potentials as distributed controllers. Progress in
Biophysics & Molecular Biology 149: 39–53.
CerveraJ, PietakA, LevinM, MafeS. 2018. Bioelectrical
coupling in multicellular domains regulated by gap junctions:
a conceptual approach. Bioelectrochemistry 123: 45–61.
ClarkA, ChalmersD. 1998. The extended mind (active
externalism). Analysis 58: 7–19.
Cohen-KarniT, LangerR, KohaneDS. 2012. The smartest
materials: the future of nanoelectronics in medicine. ACS
Nano 6: 6541–6545.
ConstantA, RamsteadMJD, VeissiereSPL, CampbellJO,
FristonKJ. 2018. A variational approach to niche
construction. Journal of the Royal Society Interface 15:
20170685.
CzégelD, GiaffarH, TenenbaumJB, SzathmáryE. 2022.
Bayes and Darwin: How replicator populations implement
Bayesian computations. Bioessays 44: e2100255.
CzégelD, ZacharI, SzathmáryE. 2019. Multilevel selection
as Bayesian inference, major transitions in individuality as
structure learning. Royal Society Open Science 6: 190202.
DanilovY, TylerM. 2005. BrainPort: an alternative input to
the brain. Journal of Integrative Neuroscience 4: 537–550.
DarwinC. 1859. On the origin of species by means of natural
selection, or, the preservation of favoured races in the struggle
for life. London: J. Murray.
DegenhartAD, BishopWE, ObyER, Tyler-KabaraEC,
ChaseSM, BatistaAP, YuBM. 2020. Stabilization of
a brain–computer interface via the alignment of low-
dimensional spaces of neural activity. Nature Biomedical
Engineering 4: 672–685.
DeMarseTB, WagenaarDA, BlauAW, PotterSM. 2001.
The Neurally Controlled Animat: biological brains acting
with simulated bodies. Autonomous Robots 11: 305–310.
DevittM. 2010. Species have (partly) intrinsic essences.
Philosophy of Science 77: 648–661.
DiPaoloEA. 2000. Homeostatic adaptation to inversion
of the visual field and other sensorimotor disruptions. In:
MeyerJ-A, BerthozA, FloreanoD, RoitblatHL, WilsonSW,
eds. SAB2000 sixth international conference on simulation of
adaptive behavior: from animals to animats, 440-449. MIT
Press: Paris. Available from: http://cognet.mit.edu/pdfviewer/
book/9780262316071/chap45.
DingS, O’BanionCP, WelfareJG, LawrenceDS. 2018.
Cellular cyborgs: on the precipice of a drug delivery
revolution. Cell Chemical Biology 25: 648–658.
DongX, KheiriS, LuY, XuZ, ZhenM, LiuX. 2021. Toward a
living soft microrobot through optogenetic locomotion control
of Caenorhabditis elegans. Science Robotics 6: eabe3950.
Doursat R, SanchezC. 2014. Growing fine-grained
multicellular robots. Soft Robotics 1: 110–121.
DoursatR, Sayama H, MichelO. 2013. A review of
morphogenetic engineering. Natural Computing 12:
517–535.
DurantF, BischofJ, FieldsC, MorokumaJ, LaPalmeJ,
HoiA, LevinM. 2019. The role of early bioelectric signals
in the regeneration of planarian anterior/posterior polarity.
Biophysical Journal 116: 948–961.
DurantF, MorokumaJ, FieldsC, WilliamsK, AdamsDS,
LevinM. 2017. Long-term, stochastic editing of regenerative
anatomy via targeting endogenous bioelectric gradients.
Biophysical Journal 112: 2231–2243.
EbrahimkhaniMR, LevinM. 2021. Synthetic living
machines: a new window on life. iScience 24: 102505.
ElgartM, SnirO, SoenY. 2015. Stress-mediated tuning of
developmental robustness and plasticity in flies. Biochimica
et Biophysica Acta 1849: 462–466.
Elya C, Lok TC, Spencer QE, McCausland H,
MartinezCC, EisenM. 2018. Robust manipulation of the
behavior of Drosophila melanogaster by a fungal pathogen in
the laboratory. eLife 7: e34414.
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

26 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
Emmons-BellM, DurantF, TungA, PietakA, MillerK,
KaneA, Martyniuk CJ, DavidianD, MorokumaJ,
LevinM. 2019. Regenerative adaptation to electrochemical
perturbation in Planaria: a molecular analysis of
physiological plasticity. iScience 22: 147–165.
FankhauserG. 1945a. The effects of changes in chromosome
number on amphibian development. The Quarterly Review of
Biology 20: 20–78.
FankhauserG. 1945b. 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 100: 445–455.
FieldsC, BischofJ, LevinM. 2020. Morphological coordination:
a common ancestral function unifying neural and non-neural
signaling. Physiology (Bethesda, Md.) 35: 16–30.
FieldsC, LevinM. 2018. Are Planaria individuals? What
regenerative biology is telling us about the nature of
multicellularity. Evolutionary Biology 45: 237–247.
FieldsC, LevinM. 2020a. How do living systems create
meaning? Philosophies 5: 36.
FieldsC, LevinM. 2020b. Scale-free biology: integrating
evolutionary and developmental thinking. BioEssays 42:
e1900228.
FredericksenMA, Zhang Y, Hazen ML, LoretoRG,
MangoldCA, ChenDZ, Hughes DP. 2017. Three-
dimensional visualization and a deep-learning model
reveal complex fungal parasite networks in behaviorally
manipulated ants. Proceedings of the National Academy of
Sciences of the United States of America 114: 12590–12595.
FristonK. 2013. Life as we know it. Journal of the Royal
Society Interface 10: 20130475.
FristonK, LevinM, SenguptaB, PezzuloG. 2015. Knowing
one’s place: a free-energy approach to pattern regulation.
Journal of the Royal Society Interface 12: 20141383.
FristonK, SenguptaB, AulettaG. 2014. Cognitive dynamics:
from attractors to active inference. Proceedings of the IEEE
102: 427–445.
Gabalda-SagarraM, CareyLB, Garcia-OjalvoJ. 2018.
Recurrence-based information processing in gene regulatory
networks. Chaos 28: 106313.
GawneR, McKennaKZ, LevinM. 2020. Competitive and
coordinative interactions between body parts produce
adaptive developmental outcomes. BioEssays 42: e1900245.
GershlakJ, Hernandez S, FontanaG, Perreault L,
HansenK, Larson S, BinderB, DolivoD, YangT,
DominkoT. 2017. Crossing kingdoms: Using decellularized
plants as perfusable tissue engineering scaffolds.
Biomaterials 125: 13–22.
GilroyS, TrewavasT. 2022. Agency, teleonomy and signal
transduction in plant systems. Biological Journal of the
Linnean Society.
GiselbrechtS, RappBE, NiemeyerCM. 2013. The chemistry
of cyborgs—interfacing technical devices with organisms.
Angewandte Chemie (International ed. in English) 52:
13942–13957.
GouldSJ, LewontinRC. 1979. The spandrels of San Marco
and the Panglossian paradigm: a critique of the adaptationist
programme. Proceedings of the Royal Society B: Biological
Sciences 205: 581–598.
GreenS, BattermanR. 2017. Biology meets physics:
reductionism and multi-scale modeling of morphogenesis.
Studies in History and Philosophy of Biological and
Biomedical Sciences 61: 20–34.
GrossbergS. 1978. Communication, memory, and development.
In: RosenR, SnellF, eds. Progress in theoretical biology,
183-232.
HallgrímssonB, HallBK. 2011. Epigenetics: linking genotype
and phenotype in development and evolution. Berkeley:
University of California Press.
HanczycMM, CascheraF, RasmussenS. 2011. Models of
minimal physical intelligence. Procedia Computer Science 7:
275–277.
HeamsT. 2012. Selection within organisms in the nineteenth
century: Wilhelm Roux’s complex legacy. Progress in
Biophysics and Molecular Biology 110: 24–33.
Herrera-DelgadoE, Perez-Carrasco R, Briscoe J,
SollichP. 2018. Memory functions reveal structural
properties of gene regulatory networks. PLoS Computational
Biology 14: e1006003.
HughesDP, AndersenSB, Hywel-JonesNL, HimamanW,
BillenJ, BoomsmaJJ. 2011. Behavioral mechanisms and
morphological symptoms of zombie ants dying from fungal
infection. BMC Ecology 11: 13.
Hughes DP, Araujo JP, Loreto RG, QuevillonL,
deBekkerC, EvansHC. 2016. From so simple a beginning:
the evolution of behavioral manipulation by fungi. Advanced
Genetics 94: 437–469.
HughesDP, LibersatF. 2019. Parasite manipulation of host
behavior. Current Biology 29: R45–R47.
ImirzianN, HughesDP. 2021. An agent-based model shows
zombie ants exhibit search behavior. Journal of Theoretical
Biology 526: 110789.
JablonkaE, Lamb MJ, AvitalE. 1998. ‘Lamarckian’
mechanisms in Darwinian evolution. Trends in Ecology &
Evolution 13: 206–210.
JamesW. 1890. The principles of psychology, Vol. 1. Henry Holt
and Co. Available from: https://doi.org/10.1037/10538-000.
JenningsHS. 1906. Behavior of the lower organisms. New
York: The Columbia University Press, The Macmillan
Company, agents; etc.
JonasE, KordingKP. 2017. Could a neuroscientist
understand a microprocessor? PLoS Computational Biology
13: e1005268.
KammRD, BashirR. 2014. Creating living cellular machines.
Annals of Biomedical Engineering 42: 445–459.
KauffmanSA. 1971. Articulation of Parts Explanation in
Biology and the Rational Search for them. In: BuckRC,
CohenRS, eds. PSA 1970. Boston Studies in the Philosophy
of Science, vol 8. Springer, Dordrecht. Available from: https://
doi.org/10.1007/978-94-010-3142-4_18.
KeijzerF, vanDuijnM, LyonP. 2013. What nervous systems
do: early evolution, input–output, and the skin brain thesis.
Adaptive Behavior 21: 67–85.
KielibaP, ClodeD, Maimon-MorRO, MakinTR. 2021.
Robotic hand augmentation drives changes in neural body
representation. Science Robotics 6: abd7935.
KriegmanS, BlackistonD, LevinM, BongardJ. 2020. A
scalable pipeline for designing reconfigurable organisms.
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

ENGINEERING NOVEL BODIES AND NOVEL MINDS 27
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
Proceedings of the National Academy of Sciences of the
United States of America 117: 1853–1859.
KriegmanS, BlackistonD, LevinM, BongardJ. 2021.
Kinematic self-replication in reconfigurable organisms.
Proceedings of the National Academy of Sciences of the
United States of America 118: e2112672118.
KullK. 2014. Adaptive evolution without natural selection.
Biological Journal of the Linnean Society 112: 287–294.
KurzweilR. 2005. The singularity is near: when humans
transcend biology. New York: Viking.
LanderAD. 2004. A calculus of purpose. PLoS Biology 2: e164.
LangeM. 2013. What makes a scientific explanation
distinctively mathematical? British Journal for the
Philosophy of Science 64: 485–511.
LangtonCG. 1995. Artificial life: an overview. Cambridge:
MIT Press.
LebedevMA, NicolelisMA. 2011. Toward a whole-body
neuroprosthetic. Progress in Brain Research 194: 47–60.
LevinM. 2011. The wisdom of the body: future techniques and
approaches to morphogenetic fields in regenerative medicine,
developmental biology and cancer. Regenerative Medicine 6:
667–673.
LevinM. 2019. The computational boundary of a ‘self’:
developmental bioelectricity drives multicellularity and
scale-free cognition. Frontiers in Psychology 10: 2688.
LevinM. 2022. Technological approach to mind everywhere:
an experimentally-grounded framework for understanding
diverse bodies and minds. Frontiers in Systems Neuroscience
16: 768201.
LevinM, PietakAM, BischofJ. 2019. Planarian regeneration
as a model of anatomical homeostasis: Recent progress in
biophysical and computational approaches. Seminars in Cell
& Developmental Biology 87: 125–144.
LiW, MatsuhisaN, LiuZ, WangM, LuoY, CaiP, ChenG,
ZhangF, LiC, LiuZ, LvZ, ZhangW, ChenX. 2021. An
on-demand plant-based actuator created using conformable
electrodes. Nature Electronics 4: 134–142.
LoboD, SolanoM, BubenikGA, LevinM. 2014. A linear-
encoding model explains the variability of the target
morphology in regeneration. Journal of the Royal Society
Interface 11: 20130918.
LyonP. 2006. The biogenic approach to cognition. Cognitive
Processing 7: 11–29.
LyonP. 2015. The cognitive cell: bacterial behavior
reconsidered. Frontiers in Microbiology 6: 264.
MadhavanR, Chao ZC, Wagenaar DA, BakkumDJ,
PotterSM. 2006. Multi-site stimulation quiets network-
wide spontaneous bursts and enhances functional plasticity
in cultured cortical networks. Conference Proceedings IEEE
Engineering in Medicine and Biology Society 2006: 1593–
1596. doi:10.1109/IEMBS.2006.260571. PMID: 17946052.
ManickaS, HarveyI. 2008. Psychoanalysis of a minimal
agent. In: Proceedings of the Eleventh International
Conference on the Simulation and Synthesis of Living
Systems. Winchester, UK: MIT Press. Available from: http://
sro.sussex.ac.uk/id/eprint/23984/.
ManickaS, LevinM. 2019. The cognitive lens: a primer
on conceptual tools for analysing information processing
in developmental and regenerative morphogenesis.
Philosophical Transactions of the Royal Society B: Biological
Sciences 374: 20180369.
MathewsJ, LevinM. 2018. The body electric 2.0: recent
advances in developmental bioelectricity for regenerative and
synthetic bioengineering. Current Opinion in Biotechnology
52: 134–144.
MaturanaHR, VarelaFJ. 1980. Autopoiesis and cognition
the realization of the living. In: Boston studies in the
philosophy of science. Dordrecht: Springer Science and
Buisness Media, 191. Available from: https://books.google.
com/books?id=nVmcN9Ja68kC.
McCuskerC, GardinerDM. 2011. The axolotl model for
regeneration and aging research: a mini-review. Gerontology
57: 565–571.
McSheaDW. 2016. Freedom and purpose in biology. Studies
in History and Philosophy of Biological and Biomedical
Sciences 58: 64–72.
MehraliM, BagherifardS, AkbariM, ThakurA, MiraniB,
MehraliM, Hasany M, OriveG, DasP, EmneusJ,
AndresenTL, Dolatshahi-Pirouz A. 2018. Blending
electronics with the human body: a pathway toward a
cybernetic future. Advanced Science 5: 1700931.
MerrittT, HamidiF, AlistarM, DeMenezesM. 2020. Living
media interfaces: a multi-perspective analysis of biological
materials for interaction. Digital Creativity 31: 1–21.
MooreDG, WalkerSI, LevinM. 2018. Pattern regeneration
in Coupled Networks. Proceedings of the ALIFE 2018: The
2018 Conference on Artificial Life. ALIFE 2018: The 2018
Conference on Artificial Life, 204-205. Tokyo, Japan: ASME.
Available from: https://doi.org/10.1162/isal_a_00043.
NagelSK, CarlC, KringeT, MärtinR, König P. 2005.
Beyond sensory substitution—learning the sixth sense.
Journal of Neural Engineering 2: R13–R26.
Nanos V, Levin M. 2022. Multi-scale chimerism: an
experimental window on the algorithms of anatomical
control. Cells & Development 169: 203764.
Newman SA. 2014. Chapter 6.Physico-genetics of
morphogenesis: the hybrid nature of developmental
mechanisms. In: Alessandro M, Thomas P, eds. Towards
a Theory of Development, 95–113. doi:10.1093/acprof:
oso/9780199671427.001.0001.
NewmanSA. 2019. Inherency and homomorphy in the
evolution of development. Current Opinion in Genetics &
Development 57: 1–8.
Ollé-Vila A, Duran-NebredaS, Conde-Pueyo N,
MontañezR, SoléR. 2016. A morphospace for synthetic
organs and organoids: the possible and the actual. Integrative
Biology 8: 485–503.
OriveG, TaebniaN, Dolatshahi-PirouzA. 2020. A
new era for cyborg science is emerging: the promise of
cyborganic beings. Advanced Healthcare Materials 9:
e1901023.
OviedoNJ, MorokumaJ, WalentekP, KemaIP, GuMB,
AhnJM, HwangJS, GojoboriT, LevinM. 2010. Long-
range neural and gap junction protein-mediated cues control
polarity during planarian regeneration. Developmental
Biology 339: 188–199.
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

28 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
Pagan-DiazGJ, Ramos-CruzKP, SamR, KandelME,
AydinO, SaifMTA, PopescuG, Bashir R. 2019.
Engineering geometrical 3-dimensional untethered in vitro
neural tissue mimic. Proceedings of the National Academy of
Sciences of the United States of America 116: 25932–25940.
ParkSJ, GazzolaM, ParkKS, ParkS, DiSantoV,
Blevins EL, Lind JU, CampbellPH, Dauth S,
CapulliAK, PasqualiniFS, AhnS, ChoA, Yuan H,
MaozBM, Vijaykumar R, ChoiJW, DeisserothK,
LauderGV, MahadevanL, ParkerKK. 2016. Phototactic
guidance of a tissue-engineered soft-robotic ray. Science 353:
158–162.
PayneJL, Moore JH, WagnerA. 2014. Robustness,
evolvability, and the logic of genetic regulation. Artificial Life
20: 111–126.
PenalozaCI, NishioS. 2018. BMI control of a third arm for
multitasking. Science Robotics 3: eaat1228.
PezzuloG, Levin M. 2015. Re-membering the body:
applications of computational neuroscience to the top-down
control of regeneration of limbs and other complex organs.
Integrative Biology 7: 1487–1517.
PezzuloG, LevinM. 2016. Top-down models in biology:
explanation and control of complex living systems above the
molecular level. Journal of the Royal Society Interface 13:
20160555.
PfeiferR, BongardJ, GrandS. 2007. How the body shapes the
way we think: a new view of intelligence. Cambridge: MIT Press.
PietakA, Levin M. 2017. Bioelectric gene and reaction
networks: computational modelling of genetic, biochemical
and bioelectrical dynamics in pattern regulation. Journal of
the Royal Society Interface 14: 20170425.
PinetK, DeolankarM, Leung B, McLaughlinKA.
2019. Adaptive correction of craniofacial defects in pre-
metamorphic Xenopus laevis tadpoles involves thyroid
hormone-independent tissue remodeling. Development 146:
dev175893.
Pio-LopezL. 2021. The rise of the biocyborg: synthetic biology,
artificial chimerism and human enhancement. New Genetics
and Society 40: 599–619.
PotterSM, WagenaarDA, MadhavanR, DeMarse TB.
2003. 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 25:
3690–3693. Available from: https://ieeexplore.ieee.org/
document/1280959.
PowerDA, WatsonRA, SzathmaryE, MillsR, PowersST,
DoncasterCP, CzappB. 2015. What can ecosystems
learn? Expanding evolutionary ecology with learning theory.
Biology Direct 10: 69.
PradeuT. 2019. Immunology and individuality. eLife 8:
e47384.
PrindleA, LiuJ, AsallyM, LyS, Garcia-OjalvoJ, SuelGM.
2015. Ion channels enable electrical communication in
bacterial communities. Nature 527: 59–63.
PtitoM, MoesgaardSM, GjeddeA, KupersR. 2005. Cross-
modal plasticity revealed by electrotactile stimulation of the
tongue in the congenitally blind. Brain 128: 606–614.
QuellerDC, StrassmannJE. 2009. Beyond society: the
evolution of organismality. Philosophical Transactions of the
Royal Society B: Biological Sciences 364: 3143–3155.
RamsteadMJD, ConstantA, BadcockPB, FristonKJ.
2019. Variational ecology and the physics of sentient
systems. Physics of Life Reviews 31: 188–205.
ReutlingerA. 2017. Explanation beyond causation? New
directions in the philosophy of scientific explanation.
Philosophy Compass 12: e12395.
RicottiL, TrimmerB, FeinbergAW, RamanR, ParkerKK,
BashirR, SittiM, MartelS, DarioP, MenciassiA. 2017.
Biohybrid actuators for robotics: a review of devices actuated
by living cells. Science Robotics 2: eaaq0495.
RosenbluethA, WienerN, Bigelow J. 1943. Behavior,
purpose, and teleology. Philosophy of Science 10: 18–24.
RothschildRM. 2010. Neuroengineering tools/applications
for bidirectional interfaces, brain–computer interfaces, and
neuroprosthetic implants – a review of recent progress.
Frontiers in Neuroengineering 3: 112.
SahaD, MehtaD, AltanE, ChandakR, TranerM, LoR,
GuptaP, SingamaneniS, ChakrabarttyS, RamanB.
2020. Explosive sensing with insect-based biorobots.
Biosensors and Bioelectronics: X 6: 100050.
SalóE, AbrilJF, AdellT, CebriàF, EckeltK, Fernandez-
TaboadaE, Handberg-Thorsager M, IglesiasM,
MolinaMD, Rodriguez-EstebanG. 2009. Planarian
regeneration: achievements and future directions after
20 years of research. The International Journal of
Developmental Biology 53: 1317–1327.
SchreierHI, SoenY, BrennerN. 2017. Exploratory adaptation
in large random networks. Nature Communications 8: 14826.
ShapiroJA. 2022. Engines of innovation: biological origins of
genome evolution. Biological Journal of the Linnean Society
blac041.
ShullPB, DamianDD. 2015. Haptic wearables as sensory
replacement, sensory augmentation and trainer – a review.
Journal of Neuroengineering and Rehabilitation 12: 59.
SimsM, PezzuloG. 2021. Modelling ourselves: what the
free energy principle reveals about our implicit notions of
representation. Synthese 199: 7801–7833.
SimunovicM, BrivanlouAH. 2017. Embryoids, organoids
and gastruloids: new approaches to understanding
embryogenesis. Development 144: 976–985.
SlomanA. 1984. The structure of the space of possible minds.
In: TorranceS, HorwoodE, eds. The mind and the machine:
philosophical aspects of artificial intelligence. MIT Press,
35–42. Available from: https://www.cs.bham.ac.uk/research/
projects/cogaff/81-95.html#49a.
SpemannH. 1938. Embryonic development and induction
(Ser. Yale university. mrs. hepsa ely silliman memorial
lectures, [1933]). Yale University press; London, H. Milford,
Oxford University Press.
SteinkrausDC, HajekAE, Liebherr JK. 2017. Zombie
soldier beetles: Epizootics in the goldenrod soldier
beetle, Chauliognathus pensylvanicus (Coleoptera:
Cantharidae) caused by Eryniopsis lampyridarum
(Entomophthoromycotina: Entomophthoraceae). Journal of
Invertebrate Pathology 148: 51–59.
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

ENGINEERING NOVEL BODIES AND NOVEL MINDS 29
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
StrassmannJE, QuellerDC. 2010. The social organism:
congresses, parties, and committees. Evolution; international
journal of organic evolution 64: 605–616.
SultanSE, MoczekAP, WalshD. 2022. Bridging the
explanatory gaps: what can we learn from a biological agency
perspective? BioEssays 44: e2100185.
SzabóA, VattayG, KondorD. 2012. A cell signaling model
as a trainable neural nanonetwork. Nano Communication
Networks 3: 57–64.
SzilagyiA, SzaboP, SantosM, Szathmary E. 2020.
Phenotypes to remember: evolutionary developmental
memory capacity and robustness. PLoS Computational
Biology 16: e1008425.
TsudaS, ArtmannS, ZaunerK-P. 2009. The phi-bot: a robot
controlled by a slime mould. In: AdamatzkyA, KomosinskiM,
eds. Artificial life models in hardware. London: Springer,
213–232.
TuringAM. 1952. The chemical basis of morphogenesis.
Philosophical Transactions of the Royal Society of London,
Series B: Biological Sciences 237: 37–72.
TurnerDA. 2016. Functional Outcome from Traumatic
Brain Injury with Brain -Machine Interface
Neuromodulation: Neuroprosthetic Scaling in Relation
to Injury Severity. In: LaskowitzD, GrantG, eds.
Translational research in traumatic brain injury. Boca
Raton (FL): CRC Press/Taylor and Francis Group. Chapter
18. Available from: https://www.ncbi.nlm.nih.gov/books/
NBK326719/.
TurnerJS. 2000. The extended organism: the physiology of
animal-built structures. Cambridge: Harvard University
Press.
TurnerJS. 2017. Purpose & desire: what makes something
“alive” and why modern Darwinism has failed to explain it.
New York: HarperOne.
vonUexküllJ. 2010. A foray into the worlds of animals and
humans with a theory of meaning Posthumanities 12, 1st
edn. Minneapolis: University of Minnesota Press. 1 online
resource.
VanAckerenMJ, BarberoFM, MattioniS, BottiniR,
CollignonO. 2018. Neuronal populations in the occipital
cortex of the blind synchronize to the temporal dynamics of
speech. eLife 7: e31640.
VandenbergLN, AdamsDS, LevinM. 2012. Normalized
shape and location of perturbed craniofacial structures in
the Xenopus tadpole reveal an innate ability to achieve
correct morphology. Developmental Dynamics 241:
863–878.
Vane-WrightRI. 2014. What is life? And what might be said
of the role of behaviour in its evolution? Biological Journal of
the Linnean Society 112: 219–241.
VarelaF, Maturana H. 1972. Mechanism and biological
explanation. Philosophy of Science 39: 378–382.
VarelaFG, MaturanaHR, UribeR. 1974. Autopoiesis: the
organization of living systems, its characterization and a
model. BioSystems 5: 187–196.
VarelaFJ, ThompsonE, RoschE. 1991. The embodied
mind: cognitive science and human experience. Cambridge:
MIT Press.
VelozT. 2021. Goals as emergent autopoietic processes.
Frontiers in Bioengineering and Biotechnology 9: 720652.
VogtN. 2021. Assembloids. Nature Methods 18: 27.
WagnerA, RosenW. 2014. Spaces of the possible: universal
Darwinism and the wall between technological and
biological innovation. Journal of the Royal Society Interface
11: 20131190.
WarwickK. 2014. The cyborg revolution. NanoEthics 8:
263–273.
WarwickK, NasutoSJ, BecerraVM, WhalleyBJ. 1998.
Experiments with an in-vitro robot brain. In: CaiY, ed.
Computing with instinct. Berlin, Heidelberg: Springer.
WatsonRA, Buckley CL, MillsR, Davies A. 2010.
Associative memory in gene regulation networks. In: Harold
F, Mark D, Martin H, Lone LL, Sarah M, Daniel M, Pierre-
Alain M, Kasper S, Steen R, eds. Artificial life conference XII.
Odense, Denmark, 194–201.
WatsonRA, MillsR, BuckleyCL, KouvarisK, JacksonA,
PowersST, CoxC, TudgeS, DaviesA, KouniosL,
PowerD. 2016. Evolutionary connectionism: algorithmic
principles underlying the evolution of biological organisation
in evo-devo, evo-eco and evolutionary transitions.
Evolutionary Biology 43: 553–581.
WatsonRA, SzathmaryE. 2016. How can evolution learn?
Trends in Ecology & Evolution 31: 147–157.
WatsonRA, WagnerGP, PavlicevM, WeinreichDM,
MillsR. 2014. The evolution of phenotypic correlations and
‘developmental memory’. Evolution; international journal of
organic evolution 68: 1124–1138.
West-EberhardMJ. 1998. Evolution in the light of
developmental and cell biology, and vice versa. Proceedings
of the National Academy of Sciences of the United States of
America 95: 8417–8419.
West-EberhardMJ. 2005a. Developmental plasticity and
the origin of species differences. Proceedings of the National
Academy of Sciences of the United States of America 102:
6543–6549.
West-EberhardMJ. 2005b. Phenotypic accommodation:
adaptive innovation due to developmental plasticity.
Journal of Experimental Zoology. Part B.Molecular and
Developmental Evolution 304: 610–618.
WhitingJG, JonesJ, BullL, LevinM, AdamatzkyA.
2016. Towards a Physarum learning chip. Scientific Reports
6: 19948.
WienerN. 1961. Cybernetics: or control and communication
in the animal and the machine. Cambridge: The M.I.T.
Press.
WillettFR, AvansinoDT, HochbergLR, HendersonJM,
Shenoy KV. 2021. High-performance brain-to-text
communication via handwriting. Nature 593: 249–254.
WilliamsKB, BischofJ, LeeFJ, MillerKA, LaPalmeJV,
WolfeBE, LevinM. 2020. Regulation of axial and
head patterning during planarian regeneration by a
commensal bacterium. Mechanisms of Development 163:
103614.
WrightJ, MacefieldVG, vanSchaikA, TapsonJC. 2016.
A review of control strategies in closed-loop neuroprosthetic
systems. Frontiers in Neuroscience 10: 312.
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023

30 W. P.CLAWSON and M.LEVIN
© 2022 The Linnean Society of London, Biological Journal of the Linnean Society, 2022, XX, 1–30
YampolskiyRV. 2015. The space of possible mind designs.
In: BiegerJ, GoertzelB, PotapovA, eds. 8th International
Conference, Artificial General Intelligence 2015. Berlin:
Springer International, 218–227.
YangCY, Bialecka-FornalM, WeatherwaxC, LarkinJW,
Prindle A, LiuJ, Garcia-Ojalvo J, Suel GM.
2020. Encoding membrane-potential-based memory
within a microbial community. Cell Systems 10: 417–
423 e413.
ZhangH, LiZ, GaoC, FanX, PangY, LiT, WuZ, XieH,
HeQ. 2021. Dual-responsive biohybrid neutrobots for active
target delivery. Science Robotics 6: eaaz9519.
Downloaded from https://academic.oup.com/biolinnean/advance-article/doi/10.1093/biolinnean/blac073/6643577 by guest on 27 March 2023