Substrate
Flexibility and the Copernican Principle of Consciousness
Jeremy Pober
Centre of
Philosophy, University of Lisbon
(corresponding
author: Jeremy.pober@gmail.com)
Eric Schwitzgebel
Department of
Philosophy, University of California, Riverside
0.
Abstract
We present a novel argument for the substrate flexibility of
consciousness – that is, for the idea that conscious experiences can arise in a
variety of different types of physical media, not just in biological animals as
they currently exist on Earth. Some recent critiques of standard arguments for the
substrate flexibility of consciousness (e.g., Cao 2022; Block 2025; Seth
forthcoming) have emphasized that humanlike
consciousness might require our specific biological substrate. However, such
critiques are too narrowly focused to address the issue of consciousness in
entities whose experience may be very different from ours, for example alien
life forms or future AI systems designed along unfamiliar lines. Given that
it’s likely that functionally complex, behaviorally sophisticated entities have
arisen or will arise many times in the observable universe, in diverse
substrates, we argue that it would be a violation of a principle of Copernican
mediocrity to hold that among these diverse entities, only we, or only we and a
small proportion of others who share our substrate, are conscious.
1.
Introduction
Who is conscious? We—both authors and you,
the reader—are. Some non-human animals presumably are (Trestman, Birch, and Allen
1995/2026; Birch 2024). Likely, as we will argue, some extraterrestrial
entities (“aliens”) are. We don’t think there’s good evidence that current
technology has produced conscious artifacts (Artificial Intelligence, or “AI”),
though we don’t rule out that possibility in the future (Schwitzgebel and Pober 2026; Schwitzgebel forthcoming).
Conceptually prior to “who is conscious?”
is the question “what kinds of things can be conscious?” In this article, we
assume physicalism (or materialism), which holds that all conscious entities
are fundamentally composed of matter. We’re generally inclined to think that
only certain kinds of matter, organized in very specific and complex ways, can
give rise to consciousness, though we won’t directly argue for this claim here.
What we will argue for is that
consciousness can be realized in very different kinds of physical
arrangement—different substrates. Consciousness is substrate flexible.
After defining substrate flexibility in
Section 2, we offer the following three-premise argument in sections 3, 4, and
5, respectively:
P1: The universe contains at least a
thousand different behaviorally sophisticated species.
P2: These species have substantially
different substrates.
P3: Some of these species with
substantially different substrates are conscious.
Conclusion: Therefore, consciousness is
possible in substantially different substrates.
P1 is justified on cosmological grounds; P2
from astrobiological and biochemical considerations
concerning the possible substrates of life, and P3 from the Copernican
Principle of Consciousness (Schwitzgebel and Pober
2026). In S6 we discuss how our argument relates to the current debate on
substrate flexibility, in particular the best-known argument against it,
advanced by Peter Godfrey-Smith (2016, 2024), Rosa Cao (2022), and Ned Block
(2025). Godfrey-Smith, Cao, and Block aim to undercut the “neural replacement”
argument for substrate flexibility found in Tom Cuda (1985) and David Chalmers
(1996). As we will argue, although these authors convincingly show that the
precise duplication of humanlike cognitive function, and thus (presumably)
humanlike consciousness, is unlikely to be possible in a substrate different
from our own, their argument does not establish that non-humanlike consciousness is unlikely in a non-humanlike
substrate. Section 7 extends our reasoning to AI cases.
2.
Substrate
Flexibility
Glass is a different substrate than
ceramic. Since cups can be made of either glass or ceramic, cups can be
realized in different substrates. Computer hard drives can be solid state,
optical, or magnetic—different types of physical substrate for the long-term
storage of computer information. In this section, we articulate an account of
substrate flexibility, which also serves as an account of substrate dependence
and substrate independence.
In our way of speaking a target
phenomenon has, and is realized by, a physical substrate. Target
phenomena may be properties, events, states, or entities, as may their
substrates, though the categories need to correspond, i.e., properties are
realized in properties, states in states, and so on (Adams 1979). We will talk
of target and substrate properties as a general term except where differences
between properties and states or entities matter for our argument. We also
assume for ease of exposition that target and substrate properties are property
kinds and not property instances, though one can sensibly use
target/substrate/realizer talk for either.
A target property is substrate flexible to the extent it constrains its possible
substrates. In theory, if it completely constrains the space of possible
substrates, such that there is only one possible substrate property, it is
completely substrate dependent. If it places no constraints whatsoever, so that
any property may realize it, it is completely substrate independent. If the
property places some constraints on its substrates, such that it lies between
these two extremes, it is substrate flexible.
Substrate flexibility is a matter of
degree. Being a standard-isotope hydrogen atom is minimally substrate flexible
in the sense that it can be realized only in a very narrow range of physical
properties. Being a cup or a hard drive is much more substrate flexible.
However, they are not completely substrate independent, since a cup cannot be
made of air nor a hard drive of cheese (pace Putnam [1975]).
Crucially, the degree to which any property
is substrate flexible depends how coarsely or finely the target
property is specified. The more finely a target property is specified,
the less substrate flexible it is likely to be. The property of being a
corkscrew, specified coarsely, is at least moderately substrate flexible,
whereas the property of being a corkscrew made of molded plastic in
such-and-such a configuration, specified more finely, is much less substrate
flexible.
Properties can be specified more coarsely
or finely along multiple dimensions. One such dimension pertains to the
mereological parts, or mereological parts of parts (or parts of parts of parts,
etc.) of a bearer of the target property, and appealing to properties of those
parts; doing so is often described as giving a lower ‘level’ of description. Here
the difference between properties and other phenomena (states, entities, and
events) matters. For example, when discussing an entity, such as water,
more finely specified entities are parts of the entity, e.g., atoms in the H2O
molecule. But when discussing the property of being water, properties of
more finely specified parts are not parts of the property of being water (neither
properties of parts nor the property of being a part are parts of properties).[1]
Nonetheless, as our examples have already
shown, specifying properties more finely need not involve appealing to
mereological proper parts. We can sensibly speak of the substrate of a cup as
the whole lump of matter comprising it.
In our way of speaking, a target property
is realized in a substrate property. We borrow this term from the
literature on ‘multiple realization’, which focuses on the physical realizers
(what we call substrates) of functional mental states (Putnam 1967; Fodor 1974;
Kim 1992; Shapiro 2000).[2]
However, there are three differences between our use of the term and its
typical use in this literature. First, as we have already noted, the class of
phenomena that is or can be realized is far broader than functional properties
as standardly conceived. Second, while functional properties can be realized in
either physical or functional properties (Shoemaker 1981; Pober,
under review), substrates must be physical (including kinds of physical
properties). And third, there is an issue within literature about whether
properties with low substrate flexibility do or do not not
count as genuinely “multiply realizable” (Shapiro and Polger
2016; Piccinini 2020). Shapiro (2000), for example, has argued that although
single-handled (“waiters”) and double-handled (“flywheel”) corkscrews count as
multiple realizations of the functional kind corkscrew, otherwise identical
corkscrews made of differently colored molded plastics count not as “multiple
realized” but only as “variably realized”.
In our terminology, the property of being
a corkscrew made of molded plastic in such-and-such a configuration is
substrate flexible, albeit to a low degree.
Substrate flexibility as we have defined it
is cheap, and indeed, a property’s being completely
substrate dependent is, on our conception, an extreme case. We cast our
argument in terms of “substantial” substrate flexibility (P2 and P3 in S1). Life
is substantially substrate flexible, in our sense, if it can arise in
substrates very different from those found among terrestrial organisms. How
different is “very different”? We suggest that it would be enough for the
entities to have a chemical composition fundamentally different than any we
have so far found on Earth—for example, being composed of different “chemical
kinds” in the sense of Needham (2003, 2005). Entities are of different chemical
kinds if they are composed of different inorganic molecules or different
organic monomers, that is, different organic subcomponents on the scale of
amino or nucleic acids (Berg et al. 2018; this allows for various polymers of
the same subcomponents, such as polyesters, to count as the same chemical kind).
Despite all being carbon based, ethyl alcohol, polyester, and caffeine are
different types of substance, different chemical kinds (as can readily be
discerned by the effects of ingestion). If we were to discover conscious
entities whose basic chemical building blocks differ from ours, we could
reasonably say that they differ substantially from us in substrate.
3.
Behaviorally
Sophisticated Extraterrestrial Life
Each of the next three sections defends one
of our three premises. Here, we defend the first premise, that there are at
least a thousand of what we call ‘behaviorally sophisticated’ extraterrestrial
species across the span—and lifespan—of the universe. This premise follows from
astrobiological estimates of ‘technological
civilizations’ and our own account of behavioral sophistication (Schwitzgebel
and Pober, 2026).
Cosmologists and astrobiologists generally
estimate measurable characteristics, among which are “technological
civilization” and “technosignatures” (Frank and
Sullivan 2016; SETI 2021; Wright et al. 2022).
The idea is that an alien civilization with technology like our own is
potentially observable, for example if it broadcasts radio-frequency signals or
builds observable megastructures, such as a gigantic habitat or energy
collector around a star.
How common are technological civilizations?
While some astrobiologists think it’s possible that Earth contains the only one
in the approximately one trillion galaxies that currently form the observable
portion of the universe, estimates are typically several orders of magnitude
higher than that (e.g., Frank and Sullivan 2016). One recent survey found
median scientific estimates over one civilization per galaxy at some point in
that galaxy’s lifetime (Snyder-Beattie et al. 2021)—low enough to explain the
“Fermi Paradox” (the question of why we haven’t yet seen evidence of
technological civilizations) without making technological civilizations extremely rare. For purposes of the
present argument, the existence of at least a thousand technological
civilizations scattered in time and space across the observable universe—one
per billion galaxies, or 0.000000001% of the median scientific estimate—is more
than sufficient; we adopt this extremely conservative estimate.
Developing a technological civilization is
related to what we call behavioral sophistication. An entity is behaviorally
sophisticated, in our sense, if and only if it is capable of complex
goal-seeking, complex communication, and complex cooperation, where those terms
describe behavioral patterns without presupposing consciousness or any
particular internal architecture.
Suppose an alien species constructs
intricate devices that extract nutrition from its environment. These nutrition
sources are stored and consumed by species members as needed. Individuals
respond to environmental threats with highly specific evasive activities, often
well in advance and with narrow margins for error. The aliens interact in
highly complex patterns, engaging in activities that increase their chances of
survival or reproduction only if several others engage in specific activities
at specific times while physically remote and not in direct contact. This
interaction is enabled by complex signals with flexible contents and a
generative grammar. Such a species would be behaviorally sophisticated as we
intend the phrase.
Plausibly, all or almost all technological
civilizations would be behaviorally sophisticated, while some behaviorally
sophisticated species would not be technological (for example, if humans had
not advanced past first-century technology). It is thus reasonable to assume
that if at least a thousand technological civilizations have existed or will
exist, and if these civilizations arose independently (for example, in
different galaxies), then at least a thousand behaviorally sophisticated
species have evolved or will evolve.
Behavioral sophistication comes in degrees.
For simplicity, we assume approximately human-level behavioral sophistication.
However, as we will explain later, our argument from the Copernican Principle
to substrate flexibility can be generalized for any threshold of behavioral
sophistication above which almost all Earthly animals are conscious.
4.
Behaviorally
Sophisticated Life is Substrate Flexible
Our argument for the substrate flexibility
of behaviorally sophisticated life has two parts. First, we show that life
itself is substrate flexible with respect to biochemical properties and kinds. Second,
we contend that the evolutionary bottlenecks between life and behaviorally
sophisticated life do not add significant further restrictions on biochemical
substrates. Because we invoke (bio-)chemical property kinds, our argument will
suffice to show that life is substantially substrate flexible. (For ease of
exposition, we drop the explicit talk of properties when discussing chemical
substrates, that is, we talk of substrates as chemical kinds rather than
as having the property of being made of such-and-such chemical composition.)
4.1.
The
Substrates of Life
Philosopher David Lewis (1980) imagined
Martians whose cognition operates by the inflation of hydraulic sacs. Science
fiction writer Greg Egan (1997) imagined a life form consisting of a single,
planet-sized molecule. Scientists have hypothesized about the possibility of
life based in sulfur (Keller et al. 2017) or organoborates
(Petkowski, Bains, and Seager 2020; Grefensette et al. 2024), or employing alternative
solvents, such as ammonia (Raulin et al. 1995) or methane (McKay 2016; see Grefensette et al. 2024 for discussion of all examples).
There are three possible ways a biochemical
kind can vary: evidence suggests that the substrates of extraterrestrial life
will vary in at least two of them. The first type involves monomers made of the
same set of overall elements, but put together in a different combination than
the twenty amino or five nucleic acids we use. The second and third types
involve distinct kinds of differences in the chemical composition of the monomers.
Biochemical molecules require two types of atoms: a ‘scaffolding’ atom which
can form long, repeating chains, and ‘heteroatoms’ to give different parts of
the long chain molecule (different monomers of the polymer) functional
diversity. Terrestrial organisms use carbon as the scaffolding atom and
primarily oxygen, nitrogen, phosphorus, and sulfur as heteroatoms. The second
and third types of variation involve the use of other elements as heteroatoms
or as the scaffolding atom.
Restricting ourselves to elements used by
human life, the space of possible amino acids is far greater than the twenty we
use: using only the aforementioned heteroatoms plus carbon, there are 1054
possible amino acid ‘alphabets’ (Brown, Voracek, and Freeland 2023).
Terrestrial life did not choose the twenty we wound up with by chance: there
are functional properties that make this set nearly optimal (Ilardo et al.
2015). Specifically, the set covers a wide range of sizes and hydrophobic
states: hydrophobia is “the driver of protein folding” and volume “determines …
constraints that limit protein formation” (Brown, Voracek, and Freedland 2023,
537).[3]
In other words, the set provides among its members the properties needed for
diversity of polymer structure and function. Yet given the astounding number of
possible amino acid alphabets, ours is not alone in being optimal with respect
to certain functional properties: Brown and colleagues (ibid.) found thousands
of other ‘alphabets’ that differ in at least one amino acid in what they
consider a non-exhaustive search.
Yet life may also use other elements as
heteroatoms. Terrestrial life does not use silicon in natural circumstances,
yet Arnold and colleagues (Kan et al. 2016; Arnold 2018) have been able to induce
Rhodothermus marinus bacteria to use
their cytochrome c protein to synthesize silicon-carbon bonds that would be the
prerequisite for using silicon as a hetereoatom.
Further, other, non-terrestrial environments may be more amenable to the
use of silicon as a heteroatom (Petkowski, Bains, and
Seager 2020; Bains, Petkowski, and Seager 2024a).
To determine how effective a heteroatom an
element would be in a non-terrestrial environment, scientists examine how much
of its chemical space is available in that environment. An
element’s available chemical space is determined by two factors: the overall
chemical space of that element, or the set of all possible bonds it can form
across any environment, and the environmental factors which limit how much of
that chemical space is available. Thus, available chemical space measures the
range of stable chemical bonds an atom of it can form with other elements in
that environment (Lipinski and Hopkins 2004; Bains and Seager 2012).
Environments consist of multiple factors such as temperature and pressure, and
most chemical reactions which form bonds are only possible in a subset of
temperature/pressure settings. Another aspect of the environment is the solvent
or primary fluid in which the element in question is embedded, i.e., the
primordial solvent for life on earth was water (Bains 2004; Schulze-Makuch and
Irwin 2018; Bains, Petkowski, and Seager 2024b).[4]
The range of stable chemical compounds and
reactions can vary considerably among environments. For example, in
environments where the solvent is sulfuric acid rather than water—a situation
found in our solar system in the Venusian gas clouds and likely widely
throughout the universe (Ballesteros et al. 2019; Bains, Petkowski,
and Seager 2024a)—the available chemical space of most elements other than
silicon is vastly constricted (relative to their available chemical spaces in
water), while the available chemical space of silicon is greatly expanded (Petkowski, Bains, and Seager 2020). Sulfuric acid environments
might make the energy cost of bonding prohibitively high for familiar terrestrial
heteroatoms oxygen, nitrogen, phosphorus, and sulfur, while facilitating the
use of silicon as a heteroatom. The result would be life with a carbon
scaffold, but silicon as a widely used heteroatom and sulfuric acid as the
solvent. If somewhere in the vast universe there is life that employs
silicon-based heteroatoms in a sulfuric acid—with all the downstream structural
changes that would surely entail—life is substantially substrate flexible.
Unfortunately for silicon enthusiasts, Petkowski and colleagues argue that “silicon-based life”,
i.e., life using silicon as a scaffolding atom for its polymers is “almost
certainly impossible” (Ibid., 23). In short, they contend, the issue is that
the conditions required for silicon to form polymer chains, cryogenic
temperatures are mutually exclusive with the high-temperature conditions
required for silicon to form monomers with heteroatoms (Ibid). However, synthetic protein analogues made of
entirely different material, using metal oxides as the scaffolding and
heteroatoms have been hypothesized (Li, Zhao, and Liu 2024).
While silicon’s involvement in the
biochemical basis of life has received the most attention (e.g., Bains 2004;
Benner et al. 2004; Petkowski, Bains, and Seager
2020), if any chemically different systems can realistically evolve, then life
is highly substrate flexible—and maybe also consciousness, if such life could
develop consciousness.
.
4.2.
From
Life to Behaviorally Sophisticated Life
More has been said about the chemical basis
of life than about the physiological basis of complex life, and the little that
has been said about the latter discusses it from an evolutionary,
substrate-independent perspective (e.g., Kershenbaum 2021). Yet we can ground
our discussion by asking what evolutionary constraints behaviorally
sophisticated life places on the space of possible substrates in addition to
those posed by life itself. We suggest that there are two basic types of requirements
for the evolution of behaviorally sophisticated life which might plausibly
yield constraints on their substrates. We conclude that while these
requirements may pose constraints on cellular substrates (i.e., substrate
properties whose bearers are cells), neither offers reason to constrict the
space of possible biochemical substrates.
The first constraint comes from the need
for a behaviorally sophisticated organism to be large: on a scale of billions
of cells (or their equivalent). Not only does the organism need to be massively
multicellular, the cells need to exhibit specialization for various biological
functions. This requirement appears to impose no constraints on chemical
compounds involved in the cells, beyond the that the compounds have the
functional flexibility necessary for life (such as carbon chains with a
diversity of heteroatoms in a good solvent, as described in S4.1).
When we shift from specifying substrates in
terms of molecules and monomers to specifying substrates in terms of cells, some
alien life forms might be incapable of transitioning to massive
multicellularity. For example, some life may use nonlipid membranes or lack
membranes entirely, forming in moisture droplets or porous minerals (Grefensette et al. 2024). Such life would plausibly be
unable to scale up to massive multicellularity. However, this restriction of potential
substrate properties specified at a cellular grain does not imply any
restriction on substrates at a biochemical grain.
A second plausible constraint is that the
organism must develop a specialized type of cell for coordinating the various
specialized cells throughout its body, a ‘control system’ (Bechtel 2012). More
specifically, it must develop “some kind of central biological control system
that can facilitate and manage all of the internal bodily processes … needed to
keep them alive and well in their current environment, as well as successfully
respond to any threats to their survival that may result from changes in their
immediate external (or internal) environment” (Cranston 2015, 92). (Note that
this requirement may plausibly be understood as the physiological basis for
something like a mind or brain, though we do not insist on it). To achieve an
organ with such a function, it needs to develop a cell type optimized for
sending and receiving signals with other cells. Crucially, this signaling
ability needs to be adaptable at the cellular level: it needs to be able to
respond differently to the same stimulus over time in order for the organism to
adapt to—and eventually learn about—its environment.
Our neurons are the cells in our body
specialized for complex, coordinated, intercellular communication, and,
accordingly, our brains are our control systems. Neurons have evolved to send
and receive optimally in their environment inside (and as part of) our
body: all the major adaptations in the development of our neurons, from their
emergence in the first nervous systems (Munoz, Romanova, and Kohn 2021),
through animal evolution (Verkhatsky et al. 2022), to
the specific capacities of our cortical neurons (Galakhova
et al. 2022) have involved enhancements to their signaling abilities. Yet
signaling each other is hardly the purview of neurons alone. It is now widely
accepted that bacteria, fungi, and plants can signal among and within each
other without employing neurons (Koshland 1980; Lyon
et al. 2021; Ciaunica et al. 2023), and indeed the
cells neurons evolved from already had some signaling abilities (Munoz,
Romanova, and Kohn 2021). In our environment, neurons have great adaptive advantages
in (e.g.) speed and preservation of signal, but again we see no reason to suppose
that there couldn’t be fast, adaptive signaling in biochemical substrates
different from our own, especially if they can propagate electrical signals. We
thus see no reason to think that cognitive sophistication creates a bottleneck
requiring that the biological substrate of cognitively sophisticated alien life
must always be identical to our own.
Pulling together the ideas of S4.1 and
S4.2, we suggest that it’s unlikely that every behaviorally sophisticated
species in the universe happened to evolve the same substrate. Even if our
substrate is near-optimal given familiar environmental constraints (Ilardo et
al. 2015), the space of possible substrates is large, and other substrates may
be favored in other environments with different constraints (Bains 2004). In
this context, the constraints posed by conditions in the Venusian gas clouds
are an example of such environmental constraints, albeit an extreme one: the
Venusian environment is almost certainly more different from ours than
an environment would need to be to favor life developing in other chemical
kinds.
5.
The
Copernican Principle of Consciousness
We now turn to our third premise, that we
should believe at least some of the behaviorally sophisticated extraterrestrial
life is conscious. The Copernican Principle of Consciousness is a special case
of the Copernican Principle in cosmology. According to the general Copernican
Principle, “the Earth is not in a central, specially favored position” (Bondi
1968, p. 13) or, alternatively, “We do not occupy a privileged position in the
Universe” (Barrow and Tipler 1986, p. 1; cf. Scharf 2014). We have argued elsewhere
(Schwitzgebel and Pober 2026) that Copernican
Principles are best understood as default principles: they are defeasibly warranted and can be defeated by further
information to the contrary. If astronomical data suggest that we are in a specially privileged position,
that would nullify any warrant for the cosmological Copernican Principle.
Copernican Principles require a reference
class with respect to which they claim we are not special, and it is important
to specify this reference class properly. In some respects, we are in a special location. The surface
of the Earth is special in its ability to support life, compared to most other
locations within the Solar System and a radius of at least several dozen light
years (just how special depends on the conditions of exoplanets we’ve yet to
investigate sufficiently). For this reason, recent scientific applications tend
to emphasize the principle’s connection to the homogeneity and isotropy of the
universe at large scales (Caldwell and Stebbins 2008; Clarkson, Bassett, and Lu
2008; Camarena, Marra, Sakr, and Clarkson 2022). It is at the scale of
galaxies, or large parts of galaxies, that the cosmological Copernican
Principle applies because at that scale we have no data to suggest it is false.
We extend the idea of the Copernican
Principle to cosmology to consciousness, as follows:
The Copernican Principle of
Consciousness: Among
behaviorally sophisticated entities, we are not specially privileged with
respect to consciousness (Schwitzgebel and Pober
2026).
Just
as it’s reasonable to assume, pending counterevidence, that our large-scale
spatial position is not exceptional, it’s reasonable to assume, pending
counterevidence, that our mentality is not special relative to the mentality of
other behaviorally sophisticated entities. We do not occupy the center of the
trillion-galaxy-wide consciousness-is-here map. Although our mentality is
special relative to rocks, rocks are a reference class to which this Copernican
reasoning doesn’t apply, since we already know we’re more conscious than they
are. As with large-scale vs. small-scale spatial position, Copernican reasoning
only applies absent counterevidence.
If the Copernican Principle of
Consciousness is correct, we are not specially lucky to possess
consciousness-instilling Earthiform biology while
other behaviorally sophisticated entities’ architectures leave them entirely
nonconscious. Nor do we have especially more or better consciousness. Absent some
reason to think we are special, we Earthlings would then be suspiciously,
inexplicably lucky.
The Copernican Principle of Consciousness
thus implies that we should believe at least some of the other behaviorally
sophisticated species in the universe are conscious. It does not necessarily
imply that all, or even most of them, are conscious. And it does not give us an
exact threshold percentage, below which consciousness would count as rare
within this class, but we can make some rough stipulations: We would be
strikingly lucky to be conscious if only 2% of behaviorally sophisticated species
were conscious, less so if 20% were.
Suppose, then, on grounds of Copernican
mediocrity that at least 20%--at least 200—of the thousand-plus behaviorally
sophisticated species in the universe are conscious. It is possible that every
single one of the conscious species is composed of the same sets of low-level
chemical types as we are, and all of those who differ from us in substrate are
nonconscious. However, for two reasons, this seems unlikely.
First, a simplicity or symmetry principle suggests
that any division between conscious and nonconscious behaviorally sophisticated
entities should correlate with some important functional deficit in the
nonconscious ones (Schwitzgebel and Pober 2026). We
see no grounds to think that such a functional deficit would inevitably be
present in behaviorally sophisticated species of any substrate other than our
own.
Second, in the broader spirit of the
Copernican principles, we should probably also accept as a default assumption
that our particular substrate is not special. Our substrate is, of course, somewhat special: here on Earth, it
alone supports behaviorally sophisticated, conscious life. But among other
substrates in the universe that can support behaviorally sophisticated life and
may be favored in environments unlike ours with different resources and
constraints, we have no reason to believe that our substrate is uniquely
capable of supporting consciousness. Without such a positive reason, believing
our substrate special in this way would again violate a plausible principle of
mediocrity.
6.
Some
Possible Objections
Having now supported each of the three
premises we articulated in the introduction, in this section, we respond to two
possible objections about the relevance of our claims.
6.1.
Extant
Discussion of Substrate Flexibility
The recent literature on substrate
flexibility (e.g., Cao 2022; Block 2025; Seth forthcoming) focuses on the
plausibility of fine-grained functional equivalents to human beings existing in
a different substrate. We do not. As a result, one might worry that we are not
discussing the same topic. In some senses—e.g., Seth’s focus on the substrate flexibility
of computation—perhaps we are discussing a different topic. However, on
what we take to be the central issue—whether a conscious being can be realized
in different stuff—we are not. Rather, we simply believe we offer a better
strategy for conceptualizing the substrate flexibility of consciousness.
The literatures owes its current focus to
the neural replacement argument, from
Cuda (1985) and Chalmers (1996). Cuda and Chalmers imagine swapping one neuron
at a time from a biological human brain, replacing each with a functionally
identical silicon chip. If each silicon chip truly is functionally identical,
then the entity at the end of full neural replacement will have neural
structures made entirely of silicon chips, but functionally identical to an
ordinary conscious human brain. Such a hypothetical entity would inevitably report being conscious and never having
noticed a change. After all, if each silicon chip acts exactly like one neuron,
then it will signal forward to other neurons or chips in exactly the way the
original neuron would have, under exactly the same conditions. All motor
outputs, including speech outputs, will thereby be just the same as they would
be in the unmodified human. Plausibly, if we trust human introspective reports
of consciousness, we should also trust the reports of no change in experience
during the swapping procedure. Therefore, the argument concludes, we should
allow that an entity with a silicon-chip based neural structure could be
conscious. Substrate flexibility
follows. (Actually, we think it’s reasonable to doubt the mid-swap
introspective reports—see Udell and Schwitzgebel 2021; Schwitzgebel 2022; Block
2023—but grant introspective infallibility for the sake of argument.)
Godfrey-Smith (2016, 2024), Cao (2022), and
Block (2023) argue convincingly against a crucial premise of this argument: it
would not be possible (that is,
nomically possible, or consistent with the laws of nature) to swap a silicon
chip for a neuron while preserving all relevant function. The activity of neurons depends on intricate
biological details. Signal speed depends
on axon and dendrite lengths, and small differences of timing can have big
downstream consequences. Cell membranes
have tens of thousands of ion channels that are sensitive in different ways to
different chemicals. Nitric oxide serves
as a diffuse signal, passing freely through cell membranes to interact with
intracellular structures. Blood flow matters, both in total amount and in the
specific chemicals transported. Glial cells, which provide support structures,
also influence neuronal behavior. Many cell changes accumulate over time
without causing immediate spiking activity. And so on. The silicon chip would
need to replicate not just activity at the membrane but many consequences of
many changes in interior structure. To
replicate all of this so precisely that the functional input-output profile
matches that of a real neuron probably requires another biological neuron.
Thus, a presupposition of the neural replacement argument fails: we probably
cannot create silicon substitutes for biological neurons that preserve all of
the functionality relevant to behavior.
If the neural replacement argument succeeded,
it would establish the substrate flexibility of consciousness. A substantial
chunk of the existing literature evaluates the feasibility of this argument.
Despite our commitment to consciousness being substrate flexible, we think the
skeptics have it right. This is not our argument.
Humanlike consciousness might require
humanlike functional architecture. We don’t commit to this, but we grant that
it’s plausible. If Cao and others are right, the property of having humanlike consciousness might
have low substrate flexibility: It might depend on highly specific biological
functions in us that can only be implemented by neurons very much like our own,
down to a fine degree of detail.[5]
However, the question at hand is not
whether humanlike consciousness requires a humanlike substrate. It is whether consciousness of any sort depends on
having a specific type of substrate, or whether instead it has a
moderate-to-high degree of substrate flexibility. The failure of the neural
replacement argument simply does not speak to that question.
The underlying issue—the reason the
literature has focused on specifically human consciousness—is because if the
neural replacement argument succeeds, it delivers substantial substrate
flexibility all in one leap; and it has seemed to many that it succeeds. Yet the
resulting focus on fine-grained functional equivalents to humans necessarily
limits the scope of conscious beings examined, and it invites the mistaken
conclusion that consciousness requires something extremely similar to our
biological configuration Indeed, Block (2025) explicitly argues that similarity
to humans at the right grain of functional or biological description is the
only sufficient ground we have for justifying attribution of consciousness to
nonhuman entities. We disagree, and have developed our argument from the
Copernican Principle of Consciousness precisely to overcome this limitation.[6]
6.2.
Substantial
Substrate Flexibility
We have made our case for the substrate
flexibility of consciousness in terms of biochemical kinds, which are a subset
of chemical kinds. One might argue that since the extant discussion is based in
terms of carbon and silicon, which are elements rather than compound molecules,
we are talking about substrate flexibility at the wrong level or degree of
specification. We disagree in two ways. First, we disagree exegetically: in biochemistry,
‘carbon-based’ life just means life made of molecules using carbon as their
scaffolding atom (Petkowski et al. 2020): it is not
as if we are made of carbon in the way that diamonds are. The question has always
been about chemical kinds.
More importantly, however, we disagree in
substance (pun intended). The substrate flexibility of consciousness is clearly
a matter of degree. Chimpanzee consciousness takes place in a somewhat
different substrate; a chimpanzee brain works just fine for hosting
consciousness. But not anything will suffice: hunks of granite are (we assume)
not conscious and cannot be made so. The advantage of the phrase “substrate
flexibility” over the sometimes used “substrate dependence” and “substrate
independence” is that the degreed nature of the phenomenon is evident from the
terminology. Although some people might be interested specifically in question
of whether consciousness could ever be realized in silicon chips, this is a
very specific application of a more general question. We will address this
application in the next section. But proper understanding of the issue requires
a broader framing.
With this broader framing in view, we think
everyone should agree that consciousness is substrate flexible to some degree.
The question is only how flexible and in what respects. Astrobiology gives us a
scientific toehold on some of the possible variation. And by appealing to
Copernican mediocrity among behaviorally sophisticated life forms, we provide
grounds for endorsing substantial substrate flexibility while avoiding commitments
to computational functionalism and other such hotly contested issues.
Flexibility in chemical kinds – different
heteroatoms, different amino or nucleic acids – suggests flexibility at larger
scales. For the same types of reasons that Cao, Block, and Godfrey-Smith
emphasize, we should not expect that systems so different in their microproperties would be identical in their macroproperties and functional properties. This is
speculative, of course; we’re aware of no studies of how brains would be
different if silicon were a primary heteroatom or if the molecule of genetic
information employed different building blocks. But it’s a reasonable
conjecture that there would be important differences. Even here on Earth,
different animal phyla have somewhat different neurochemistries
and neural structures. Mollusks and vertebrates employ different (but
overlapping and related) ranges of neuropeptides, and basic morphology and
function can differ. For example, the differentiation between axon and dendrite
is often much less pronounced in mollusks, which often employ bidirectional “neurites”
serving both the input (axon) and output (dendrite) roles or other specialized
morphologies unfamiliar from vertebrate cases (Chase 2002; Nixon and Young
2003). Given this variation among (distantly) related animals even on Earth, it
would be amazing if the control system cells (or cell-equivalents) were
uniformly similar in all conscious life across the universe, in very different
planetary (or non-planetary?) environments, with very different temperature
ranges, solvents, pressure levels, gravitational levels, radiation types and
levels, energy sources, ratios of available chemicals, and contingent chances
of evolutionary history. Astrobiological evidence
suggesting the possibility of different chemical kinds, combined with these
more general plausibility considerations, combined with the likelihood of many
independent evolutionary sites throughout spacetime, suggests that even if one
accepts that all naturally evolved conscious life must be carbon based, it will
be heterogenous from the chemical level on up.
In any case, difference in chemical substance is sufficient for
‘substantial’ substrate flexibility, in our sense—pun (again) intended.
At this point, Copernican reasoning
applies. The diversity of life forms throughout the universe will likely employ
different chemical structures and different cellular architectures, arranged
into different gross morphologies in the brain, brain-analogue, or more
distributed signaling system (think again, even on Earth, of octopuses and
jellyfish). There will likely be different, complex and intermeshing functional
relationships from small-scale chemical bonding up to large-scale functional
differences in sensory, memory, and affective systems. To think that somehow,
among this diversity, only entities with our particular architecture and
functionality would be conscious, would be unmotivated terrocentrism.
7.
Implications
for Artificial Intelligence
Substrate flexibility is typically brought
up in the context of AI consciousness, in particular, the possibility of
entities made of the same basic material as our current best generative AI
technology. While we have argued that consciousness is substrate flexible, we
have not argued that it is capable of being realized in that substrate,
which does use silicon as (the closest thing a computer chip has to) a
scaffolding atom. Should we then assume that consciousness cannot exist in such
a substrate? The two authors differ on this question.
One answer (preferred by JMP) is: until we
have reason to believe otherwise, we should assume that our current computer
chips cannot realize consciousness. Not being able to realize consciousness is
the default property we attribute to substrates until we have a reason to think
otherwise, and our argument does not give us a reason to think otherwise for
the substrates of current AI’s. Note that this position does not imply a
‘biological naturalism’ as suggested by Seth, wherein “consciousness depends on
intrinsic properties of its material biological basis” (Seth forthcoming). It
does not assume the necessity of any particular biological process, such as
autopoiesis (Maturana and Varela 1972/1980; Seth forthcoming) or low-level
biochemical metabolism (Godfrey-Smith 2016). It is instead a simpler default to
assuming that specialness needs to be established.
The other answer (preferred by ES) is: we
should be open to the possibility of AI consciousness. Once we acknowledge that
consciousness does not require our particular substrate, it seems unmotivated
to draw the line in any one specific place, as long as the substrate in
question shows the capacity to support sufficient behavioral sophistication. To
reject AI consciousness specifically on grounds of substrate would require some
argument either (1) that silicon chips cannot support the degree of behavioral
sophistication that would justify attributing consciousness to sophisticated
but differently constructed aliens, or (2) that, even if it can support the
requisite sophistication, silicon is for some other reason disprivileged
relative to carbon. The failure of the neural replacement argument gives at
best very weak evidence for such a general incapacity or disprivilege.
8.
Conclusion:
Relaxing Behavioral Sophistication
We have argued for the substrate
flexibility of consciousness by supporting the three premises we set out in the
introduction. In doing so, we used specifically human-level behavioral
sophistication as our criterion, erring on the side of conservatism. Relaxing
the required level of behavioral sophistication would presumably increase the
likelihood of conscious species existing in different substrates.
Suppose your best guess estimate is that,
on Earth, consciousness is present in all vertebrates, plus cephalopods and
some insects (Birch 2024). And suppose
that your best guess estimate is that on average each galaxy contains a million
planets where species of approximately that level of behavioral sophistication
eventually evolve (even if technological civilizations rarely arise). The
observable universe would then host, over its lifetime, a quintillion (1018)
qualifying planets. With that many draws from the lottery, some of these life
forms will be strange indeed. We do not think you will then also want to
suppose that consciousness will be limited to only those life forms lucky
enough to be made of the same substances that compose us here on Earth.
References
Adams, Fred (1979). “Properties, functionalism, and the
identity theory.” Eidos 1:153-179.
Antony, Louise,
and Joseph Levine (1997). “Reduction with Autonomy.” Philosophical
Perspectives 11:83-105.
Aronld, Frances H. (2017). “Directed Evolution: Bringing new
chemistry to life.” Angewante Chemie
International Edition 57:4143-4148.
Bains, William (2004). “Many Chemistries Could Be Used to
Build Living Systems.” Astrobiology 4:137-167.
Bains, William, Janusz J. Petkowski, and Sara Seager (2024a).
“Venus' atmospheric chemistry and cloud characteristics are compatible with venusian life.” Astrobiology 24:371-385.
Bains, William, Janusz J. Petkowski, and Sara Seager (2024b).
“Alternative Solvents for Life: Framework for evaluation, current status, and
future research.” Astrobiology 24:1231-1256.
Bains, William, and Sara Seager (2012). “A combinatorial
approach to biochemical space: description and application to the redox
distribution of metabolism.” Astrobiology 12: 271-281.
Ballesteros, Fernando Jesus, Alberto Fernandez Soto, and
Vincente J. Martinez (2019). “Diving into exoplanets: Are water seas the most
common?” Astrobiology 19:642-654.
Barrow, John D., and Frank J. Tipler (1986). The Anthropic Cosmological Principle.
Oxford University Press.
Bechtel, William G. (2012). “Representing Time of Day in
Circadian Clocks.” In (Eds.) Albert Newen, Andreas Bartels, and Eva-Maria
Young, Knowledge and representation (pp. 129-162). CSLI
Publications.
Benner, Steven A., Alonso Ricardo, and Matthew A. Carrigan (2004).
“Is there a common chemical model for life in the universe?" Current
Opinion in Chemical Biology 8:672-689.
Berg, Jeremy M., John L. Tymoczko,
Gregory J. Gatto, Jr., and Lubert Stryer. (2018). Stryer’s
Biochemistry, 8th Edition. W.H. Freeman & Co.
Birch, Jonathan (2024). The Edge of Sentience. Oxford
University Press.
Block, Ned (2023). The
Border Between Seeing and Thinking. Oxford University Press.
Block, Ned (2025). “Can only meat machines be conscious?” Trends
in Cognitive Sciences 30:298-308.
Brown, Sean M., Václav Voráček, and Stephen Freeland (2023). "What
would an alien amino acid alphabet look like and why?." Astrobiology 23:536-549.Bondi 1968
Caldwell, R. R., and A. Stebbins (2008). “A test of the
Copernican Principle.” Physical Review
Letters 100:191302.
Camarena, David, Valerio Marra, Ziad Sakr, and Chris Clarkson
(2022). “The Copernican principle in light of the latest cosmological data.” Monthly Notices of the Royal Astronomical
Society 509:1291-1302.
Cao, Rosa (2022). “Multiple realizability and the spirit of
functionalism.” Synthese 200:506.Chalmers,
David J. (1996). The Conscious Mind. Oxford University Press.
Chase, Ronald A. (2002). Behavior
and Its Neural Control in Gastropod Molluscs.
Oxford University Press.
Ciaunica, Anna, Evgeniya V. Shmeleva, and Michael
Levin (2023) "The brain is not mental! coupling neuronal and immune
cellular processing in human organisms." Frontiers in integrative
neuroscience 17:1057622.Clarkson, Bassett, and Lu 2008
Cranford, Jerry L. (2015). Astrobiological
Neurosystems. Springer.
Cuda, Tom (1985). “Against neural chauvinism.” Philosophical Studies, 48:111-127.Egan,
Greg (1997). Diaspora. Millenium.
Figdor, Carrie (2026). A
defense of substrate dependence in the mind sciences. Unpublished manuscript.
Fodor, Jerry A.(1974). “Special sciences (or: The disunity of
science as a working hypothesis).” Synthese
28:97-115.
Frank, A., and W. T. Sullivan, III (2016). “A new empirical
constraint on the prevalence of technological species in the universe.” Astrobiology 16:359-362.
Galakhova, A.A., S. Hunt, R. Wilbers, D.B. Heyer, et al.
(2022). “Evolution of cortical neurons supporting human cognition.” Trends
in Cognitive Sciences 26:909-922.
Gillet, Carl (2003). “The metaphysics of realization, multiple
realizability, and the special sciences.” The Journal of Philosophy 100:591-603.
Godfrey-Smith, Peter (2016). “Mind, matter, and metabolism.” Journal of Philosophy 113:481-506.Godfrey-Smith,
Peter (2024). “Simulation scenarios and philosophy.” Philosophy & Phenomenological Research 109:1036-1041.
Grefensette, Natalie, Luoth
Chou, Stephanie Colon-Santos, et al. (2024). “Chapter 9: Life as We Don’t Know
It.” Astrobiology 24, Supplement 1:S186-S201.
Ilardo, Melissa, Markus Meringer, Stephen Freeland, et al. (2015).
“Extraordinarily Adaptive Properties of the Genetically Encoded Amino Acids.” Scientific
Reports 5:9414.
Ionnidis, Stavros, Gal Vishne,
Meir Hemmo, and Orly Shenker, Eds. (2022). Levels of Reality in Science and
Philosophy. Springer.
Kan, S.B. Jennifer, Russell D. Lewis, Kai Chen, and Frances H.
Arnold (2016). “Directed Evolution of cytochrome c for carbon-silicon bond
formation: Bringing silicon to life.” Science 354:1048-1051.
Keller, Markus A., Domen Kampjut,
Stuart A. Harrison, and Markus Raiser (2017). "Sulfate radicals enable a
non-enzymatic Krebs cycle precursor." Nature ecology &
evolution 1: 0083.
Kershenbaum, Arik (2021). The Zoologist’s Guide to the
Galaxy. Penguin Press.
Kim, Jaegwon (1992). “Multiple Realization and the Metaphysics
of Reduction.” Philosophy and Phenomenological Research 52:1-26.
Koshland, Daniel E. (1980). Bacterial chemotaxis
as a model behavioral system. Raven Press.
Lewis 1980 Mad Pain and Martian Pain
Li, Jingyi, Xi Zhao, and Jing Liu (2024). “Biosimilar liquid
metal living matter.” Matter 7:2033-2065.
Lipinski, Christopher, and Andrew Hopkins (2004). “Navigating
Chemical Space for Biology and Medicine.” Nature 432:855-861.
Lyon, Pamela, Fred Keijzer, Detlev Arendt, and Michael Levin (2021).
“Reframing cognition: getting down to biological basics.” Philosophical
Transactions of the Royal Society of London B 376:20190750.
Maturana, Humberto R., and Francisco J. Varela (1972/1980). Autopoiesis and cognition. D. Reidel.Morris, Kevin(2018). Physicalism Deconstructed. Cambridge
University Press.
Munoz, Leonid L., Daria Y. Romanova, and Andrea B. Kohn.
(2021). “Neural versus alternative integrative systems: molecular insights into
origins of neurotransmitters.” Philosophical Transactions of the Royal
Society of London B, 376:20190762.
Needham, Paul (2003). "Chemical substances and intensive
properties." Annals of the New York Academy of Sciences 988.1:
99-113.
Needham, Paul (2005). “Mixtures and Modality.” Foundations
of Chemistry 7:103-118.
Nixon, Marion, and John Z. Young (2003). The Brains and Lives of Cephalopods. Oxford University Press.
Petkowski, Janusz J., William Bains, and Sara Seager (2020).
“On the potential of silicon as a building block for life.” Life 10:84.
Piccinini, Gualtiero (2020). Neurocognitive Mechanisms. Oxford
University Press.
Pober, Jeremy M. (under review). “Hybrid Functionalism: Between
Superficial and Deep”
Putnam, Hilary (1967). “The Nature of Mental States.”
Reprinted in H. Putnam (1975) Mind, Language, and Reality,
Cambridge: Cambridge University Press 429-440
Putnam, Hilary (1975). “Philosophy and our mental life.” In
H. Putnam (1975) Mind, Language, and Reality: Philosophical Papers Volume 2.
Cambridge, UK: Cambridge University Press.
Raulin, F., P. Bruston, P. Paillous, and R. Sternberg (1995). “The low temperature
organic chemistry of Titan’s geofluid.” Advances in Space Research 15:321-333.
Scharf, Caleb (2014). The Copernicus Complex. Farrar, Straus and Giroux.
Schwitzgebel, Eric (2022). “An objection to Chalmers’s fading
qualia argument”. Blog post at The
Splintered Mind (Dec 13). URL: https://schwitzsplinters.blogspot.com/2022/12/an-objection-to-chalmerss-fading-qualia.html
Schwitzgebel, Eric (forthcoming). AI and Consciousness. Cambridge University Press.
Schwitzgebel, Eric and Jeremy M. Pober
(2026). “The Copernican Argument for Alien Consciousness; The Mimicry
Argument Against Robot Consciousness.” Philosophers’ Imprint,
forthcoming; DOI https://doi.org/10.3998/phimp.7823
SETI (2021). “The Drake Equation”. URL: https://www.seti.org/research/seti-101/drake-equation [accessed May 22, 2026].
Seth, Anil K. (2025). “Conscious artificial intelligence and
biological naturalism.” Behavioral and Brain Sciences. Published
online 2025:1-42. doi:10.1017/S0140525X25000032
Shapiro, Lawrence A. (2000). “Multiple Realizations.” The
Journal of Philosophy 97:635-654.
Shapiro, Lawrence A. and Thomas Polger (2016). The multiple
realization book. Oxford University Press.
Shoemaker, Sydney (1981). “Some Varieties of Functionalism.” Philosophical
Topics 12:93-119.
Snyder-Beattie, Andrew E., Anders Sandberg, K. Eric Drexler,
and Michael B. Bonsall (2021). “The timing of evolutionary transitions suggests
intelligent life is rare.” Astrobiology
21:265-278.Trestman, Michael, Jonathan Birch, and Colin Allen (1995/2026). “Animal
consciousness”, Stanford Encyclopedia of
Philosophy (spring 2026 edition).
Udell, David B., and Eric Schwitzgebel (2021). “Susan
Schneider’s proposed Tests for AI Consciousness: Promising but Flawed”. Journal of Consciousness Studies 28(5-6):121-144.
Verkhatsky, Alexei, Amaia M. Arranz, Katarzyna Ciuba, and
Aleksandra Pekowska (2022). “Evolution of neuroglia.”
Annals of the New York Academy of Sciences 1518:120-130.
Wright, Jason T., Jacob Haqq-Misra, Adam Frank, Ravi
Kopparapu, Manasvi Lingam, and Sofia Z. Sheikh (2022). “The case for technosignatures: Why they may be abundant, long-lived, highly
detectable, and unambiguous”. Astrophysical
Journal Letters 927:L30.
[1] Discussion of
mereological parts of different sizes, or after a different number of
decompositions from the target entity, are often discussed in terms of
different ‘levels of reality’ (Ioannidis et al. 2022). However, we eschew such
talk here, which we take to be unnecessary for—and problematic for reasons
unrelated to—the current discussion.
[2] With respect to other
issues in the metaphysics of realization, we aim to be as ecumenical as
possible. As discussed with the example of cups, we allow for the realization
relation to hold between phenomena and substrates along “flat” and
“dimensioned” lines (Gillet 2003), that is, respectively, wherein a property
and its substrate property obtain of the same entity or state, or wherein the
substrate property obtains of a proper part, entity, or state. We are
indifferent as to whether the realization relation is identity, constitution,
or some other sui generis relation (Morris 2018) and indeed accept that
there may be multiple classes of relations which count as ‘realization’, e.g.,
physical phenomena may be identical to their substrates whereas functional
states may stand in a ‘role-filling’ (Antony and Levine 1997) relation to their
physical substrates.
[3] The same is likely true
for the nucleic acids, which only use five combinations (four each in DNA and
RNA), though we do not know of any studies which have investigated the issue.
[4] The solvent must be a
fluid: as Bains explains, “chemical life is dependent on its molecular
components being suspended in a fluid of comparable density of the
macromolecules” (Bains 2004, 139).
[5] Godfrey-Smith (2016) and
Seth (forthcoming) also suggest that consciousness might require, respectively,
metabolism and autopoiesis. However, neither gives clear positive grounds for
thinking of such biological properties as a strict requirement on consciousness,
and in any case, such abstractly specified categories are plausibly highly
substrate flexible (Schwitzgebel forthcoming).
[6] Figdor (MS) argues
against substrate flexibility on the grounds that if mental states are
substrate flexible, then it would falsely imply that phylogeny is irrelevant to
mental state individuation. We don’t disagree with Figdor any
more that we do with Cao; we just take a broader target. The
individuation of specific human or animal mental states may depend on phylogeny
in the way Figdor suggests, but not consciousness or its absence per se.