The Copernican Argument for
Alien Consciousness;
The Mimicry Argument Against
Robot Consciousness
Eric Schwitzgebel
University of California, Riverside
Riverside, CA
92521
USA
Jeremy Pober
University of Lisbon / University of Antwerp
1649-004 Lisboa, PT / Antwerpen 2000, BE
November 11, 2024
The Copernican Argument for
Alien Consciousness;
The Mimicry Argument Against
Robot Consciousness
Abstract: On broadly Copernican grounds, we are entitled to
default assume that apparently behaviorally sophisticated extraterrestrial
entities (“aliens”) would be conscious.
Otherwise, we humans would be inexplicably, implausibly lucky to have
consciousness, while similarly behaviorally sophisticated entities elsewhere
would be mere shells, devoid of consciousness. However, this Copernican default assumption is
canceled in the case of behaviorally sophisticated entities designed to mimic superficial
features associated with consciousness in humans (“consciousness mimics”), and
in particular a broad class of current, near-future, and hypothetical robots. These considerations, which we formulate,
respectively, as the Copernican and Mimicry Arguments, jointly defeat an
otherwise potentially attractive parity principle, according to which we should
apply the same types of behavioral or cognitive tests to aliens and robots,
attributing or denying consciousness similarly to the extent they perform
similarly. Instead of grounding
speculations about alien and robot consciousness in metaphysical or scientific
theories about the physical or functional bases of consciousness, our approach appeals
directly to the epistemic principles of Copernican mediocrity and inference to
the best explanation. This permits us to
justify certain default assumptions about consciousness while remaining to a
substantial extent neutral about specific metaphysical and scientific theories.
Word Count: ~11,000 words
Keywords: alien life, Artificial
Intelligence, consciousness, Copernican Principle, Large Language Models
The Copernican Argument for
Alien Consciousness;
The Mimicry Argument Against
Robot Consciousness
1. Our Argument in Sketch.
Engineers might soon build
robots so sophisticated that people are tempted to regard them as meaningfully
conscious or sentient – robots capable, or seemingly capable, of genuinely
feeling pleasure and genuinely suffering; robots that possess, or seem to
possess, real conscious awareness of themselves and their environments.[1] Less likely, but not impossible, we might
soon discover complex, seemingly intelligent, seemingly conscious space aliens
– maybe under the surface of Europa or via signals from a nearby star. It would be lovely if we had a scientifically
well-justified consensus theory of consciousness that we could
straightforwardly apply to determine whether such robots or aliens are
genuinely conscious or only superficially seem so. Unfortunately, there is no consensus theory,
and consensus is unlikely anytime soon.[2]
Still, as we will argue, we
are not wholly in the dark regarding robot and alien consciousness. When apparently confronted with what we will
call behaviorally sophisticated
robots or aliens, a particular pair of default attitudes is justified.
Toward behaviorally
sophisticated extraterrestrial entities (“aliens”), the appropriate default
attitude is to attribute consciousness.
We will defend this view on broadly Copernican grounds. According to the cosmological principle of
Copernican mediocrity, we do not occupy any special position in the
universe. Copernican mediocrity, we
argue, would be violated if, among all the behaviorally sophisticated entities
that have presumably arisen in the universe, we humans are among the rare few
who also happen to be conscious, while the rest are, so to speak, empty shells,
devoid of consciousness. That would make
us weirdly, inexplicably lucky – much as if we just happened, inexplicably, to
occupy the exact center of the universe.
Toward a broad range of
behaviorally sophisticated current, near-future, and hypothetical robots – entities
that we will call “consciousness mimics” – we will argue that the appropriate
default attitude is skepticism or uncertainty about their consciousness. If a system is designed to behave in ways
that mimic the behavior of conscious entities in order to evoke a certain type
of reaction in an audience, we normally cannot justifiably infer the existence
of consciousness from features that would otherwise suggest it. While the effective mimicry of consciousness
might in some cases require corresponding consciousness beneath, there’s no
compelling reason to think this is generally so. There’s a vast gulf between a robot designed
or selected to emit the superficial appearance of a friendly greeting – a robot
perhaps modeled on us and engineered specifically to pass various tests – and a
robot who actually experiences friendship.
Thus, the alien and robot
cases are asymmetric. If an alien exits
a spaceship and greets you, it’s reasonable to assume that you have encountered
a conscious being; if a social robot designed to mimic human behavior rolls off
a factory floor and greets you, it’s reasonable to doubt its
consciousness. This asymmetry holds even
if the alien and robot both exhibit an overall similar degree of behavioral
sophistication.
One might have thought a
certain type of parity principle would be plausible, according to which we
should approach robots and aliens similarly, applying the same types of
behavioral or cognitive tests and attributing or denying consciousness
similarly to the extent they perform similarly.
If you’d apply a certain type of Turing Test to one, for example, apply
that same Turing Test to the other, and draw similar conclusions given similar
performance.[3] If you’d look for recurrent processing loops
of Type A or cognitive performance of Type B in one, look for recurrent
processing loops of Type A or cognitive performance of Type B in the other. The Copernican Argument and the Mimicry
Argument jointly provide reason to reject robot-alien parity. This is because what we know about the
history of an entity’s creation makes an epistemic difference to our assessment
of its consciousness.
Our argumentative approach
is unusual in the following respect: Instead of attempting to ground guesses
about robot or alien consciousness in specific metaphysical or scientific
theories about the physical or functional bases of consciousness, we appeal
directly to epistemic principles – a Copernican principle and a version of
inference to the best explanation. Even
absent knowledge of what particular types of processes give rise to
consciousness and of the inner workings of the entities being evaluated, these
principles ground default assumptions about consciousness, based in our
knowledge of the entity’s history. These
broad principles can give us an epistemic toehold on alien and robot
consciousness with minimal commitment to the tendentious specifics of
particular theories of consciousness.
2. Behavioral Sophistication.
An entity is behaviorally sophisticated in our
intended sense if it is capable of complex goal-seeking, complex communication,
and complex cooperation, where “goal-seeking”, “communication”, and
“cooperation” are all understood in terms of superficial behavioral patterns without
commitment to the existence of consciousness or any particular internal
cognitive architecture.[4]
Suppose an alien species
creates complex devices that extract nutrition from its environment. The sources of nutrition are stored and then
consumed by members of the species when they would otherwise be nutritionally
deprived. Members of the species respond
to changes in their environment that threaten their survival by engaging in
highly specific activities that evade the threat, often well in advance and
with relatively narrow margins of error.
The aliens interact in highly complex and seemingly technologically
advanced ways, engaging in activities that substantially increase their chances
of survival or reproduction only if several other group members engage in
specific activities at specific times when they are physically remote and not
in direct contact. This interaction is
enabled by complex signals with flexible contents. For good measure, suppose that the aliens fly
spacecraft at least as good as our own.
Such a species would be behaviorally sophisticated as we intend
the phrase. When used by biologists
discussing the goals, communication, and cooperation of bacteria, birch trees,
and mycorrhizal fungi, such terms need not presuppose consciousness or the
types of cognitive mechanisms we are familiar with from animal cases. Similarly for our use of “goals”,
“communication”, and “cooperation” here.
(Employ scare quotes if you like.)
Maybe complex behavior of the sort described nomologically
or metaphysically requires consciousness or a particular cognitive architecture? In that case, consciousness would be built in
for free, so to speak, given the sophistication described above. But that would depend on matters of theory
about which we aim to stay neutral. The
point we insist on is that behavioral sophistication bears an epistemic relationship
to consciousness. Specifically, when it
is observed in a terrestrial being, it provides prima facie epistemic
warrant for attributing consciousness to that being.
How sophisticated is
“behaviorally sophisticated”? Our
Copernican Argument works in its simplest form if approximately human-level
sophistication is the required minimum.
However, as we will see in Section 12, a generalization of the
Copernican Argument might also work with a much lower standard of
sophistication. To forestall a possible
confusion: Our argument will be that behavioral sophistication is a sufficient condition for the default
attribution of consciousness, not a necessary condition.
3. The Copernican Principle
of Consciousness.
We assume that the emergence
of behavioral sophistication is not extremely
rare. Though it’s impossible to know
exactly how rarely behavioral sophistication arises, on a conservative
estimate, behavioral sophistication might independently evolve at least once per
every billion galaxies at some point in those galaxies’ history.[5] This would yield at least a thousand
independently arising behaviorally sophisticated clades or groups in the
galaxies currently occupying the observable portion of the universe.
Among these entities, we can
imagine various possibilities for the distribution of consciousness. On one extreme, all behaviorally
sophisticated extraterrestrial (“alien”) entities might be conscious. On the other extreme, maybe only Earthlings
are. And of course a vast range lies
between the extremes. If among all of
these entity-types, only we Earthlings were conscious, or only we and a few
others, we would be strikingly special.
According to an epistemic principle we’ll call the Copernican Principle
of Consciousness, we can default assume that we wouldn’t be in that way
special.
The Copernican Principle of
Consciousness is a special case of the more general Copernican Principle in
cosmology, according to which “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; various other
formulations similar in spirit are possible, e.g., Scharf 2014). Recent scientific applications tend to
emphasize the principle’s connection to the homogeneity and isotropy of the
universe at large scales.[6] If everything is pretty much the same
everywhere we look, no position (including our own) is particularly special.
Extending this thinking to
the case of consciousness, we propose the following Copernican Principle of Consciousness:
Among behaviorally
sophisticated entities, we are not specially privileged with respect to
consciousness.
If the Copernican Principle of Consciousness is
correct, we are not specially lucky to be blessed with a consciousness-instilling
Earthiform biology while other behaviorally
sophisticated entities’ architectures leave them entirely non-conscious. Nor are we in the specially privileged
position of being especially more
conscious than other behaviorally sophisticated entities. Nor are we blessed with a special kind of
awesomely good consciousness. Otherwise, absent some reason to think we are
special, we Earthlings would be suspiciously, inexplicably lucky. Among all the spaceship-flying,
linguistically communicating, long-term-goal planning, socially cooperative
entity-types in the universe, only we (and maybe a few others) are conscious?! Although (at least on some metaphysical and
scientific theories of consciousness) that could be so, we probably
ought to default to the assumption that we’re not uniquely bestowed with magic
light.
Scientific cosmologists generally interpret
the Copernican Principle spatially or spatiotemporally, in terms of
“position”. The Copernican Principle of
Consciousness can also be understood spatiotemporally, if desired. If the observable universe has at least a
thousand different spatiotemporal regions where behaviorally sophisticated entities
have arisen, but only a small percentage of those regions host conscious entities, then Earth is in a
special location: a bright spot of consciousness in a mostly-consciousnessless cosmos – the, or a, center of the
“consciousness is here” map.
We see Copernican principles
as default epistemic principles. Compelling evidence could potentially arise
that we are at the center of the
universe. Maybe our best theory of
consciousness will someday imply we’re lucky to have Earthiform
neurons instead of Andromedean hydraulics. The assumption of mediocrity is just a
reasonable starting place.
4. Four Concerns about the
Copernican Principle of Consciousness.
We hope that most readers
will find the Copernican Principle of Consciousness plausible on its face. However, to better articulate its boundaries
and assess its plausibility, we consider four objections.
Concern A.
“Specialness” is vacuous. Copernican principles depend
on the somewhat murky notion of “special privilege”.[7] Specified narrowly enough, any position could
be identified as unique or extremely unusual on some metric of evaluation. Maybe Earth is the only planet in the
observable universe with a mass of X, average orbital distance of Y, and moon
with magnesium-to-iron ratio of Z, with X, Y, and Z specified to 100
digits. Similarly, if I deal out five
cards from a shuffled deck and receive J♣, 4♦, 7♣, Q♠,
7♠ in that order, in some sense that is a “special” or low-probability
outcome, since the chance of receiving those exact cards in that order is under
one in three hundred million. But if
everything is special, then claims of specialness are nearly vacuous and,
contra Copernican principles, we should be unsurprised to find ourselves in a
“special” location.
Responding to this worry
requires clarifying “special”. In poker,
A♠, K♠, Q♠, J♠, 10♠ is a special or privileged
hand, while J♣, 4♦, 7♣, Q♠, 7♠ is not, because no
other hand can defeat it. For poker
purposes, a royal flush has distinctive importance. But even if we’re not playing poker, A♠,
K♠, Q♠, J♠, 10♠ is striking in a way J♣,
4♦, 7♣, Q♠, 7♠ is not (unless something of importance
hinges on drawing those exact cards).
Similarly, occupying the precise center of the universe is intuitively
striking in a way that being 3.24 billion light years from the center of an
approximately Hubble-sized universe is not.
A license plate of OOO 000 is intuitively striking in a way that XNE 488
is not. If your neighbor arrives home
with OOO 000 on their new car, you can reasonably expect an explanation other
than chance. Alternatively, specialness
might be captured by information compressibility: We can describe the royal
flush as “the top five spades in descending order”, while to describe the other
hand we probably can’t do better than just list the specific cards. We can describe the center of the universe as
just “the center”, while most other positions would require three spatial
coordinates. Thus, there are at least
three different ways of characterizing “specialness” or “privilege”: relative
to goals or rewards, in terms of something like intuitive strikingness, and in
terms of information compressibility. Underlying
all three is the idea that there is an aspect of “specialness” beyond
statistical rarity or uniqueness. If any
of these three approaches works, then Copernican principles are not in general
vacuous. Alternatively, we are open to
other ways of cashing out this additional aspect. Our list is not meant to be exhaustive.
Concern B.
Consciousness is not special. The Copernican
Principle of Consciousness requires specifically that consciousness is special – that being the one conscious entity in a
pool of behaviorally sophisticated aliens would be more like drawing A♠,
K♠, Q♠, J♠, 10♠ than like drawing J♣, 4♦,
7♣, Q♠, 7♠. But maybe
consciousness isn’t special?
On the face of it,
consciousness is very special. Some
philosophers have argued that consciousness is the most intrinsically valuable
thing in the universe, the grounds of moral status, and the primary source of
meaning in life.[8] Even if we don’t go that far, that
consciousness is even a plausible candidate for such high status suggests that
it has some importance. If you were to
learn you would never again be conscious, you would probably be dismayed in a
way you would not be dismayed to learn you will never again weigh exactly X
grams. It is also an intuitively
striking property: If it were somehow discoverable that we human beings are
conscious while our nearest behaviorally sophisticated alien neighbors are non-conscious,
that would be quite a notable difference between us. Arguably, too, when we describe an entity as
“conscious” or not, we compress substantial information into a brief statement,
in the sense that the attribution of consciousness captures some general
pattern or capacity in the target entity – though exactly which pattern or
capacity will depend on what account of consciousness is correct (or
alternatively, on what patterns or capacities the attributer regards as typical
of conscious entities).[9]
Concern C. The
Anthropic Principle makes consciousness unsurprising even if rare. Brandon Carter remarks on an important
qualification to Copernican principles: Our position is “necessarily privileged
to the extent of being compatible with our existence as observers” (1974, p.
293). Even if most of the universe
(e.g., large stretches of near vacuum) is incompatible with the existence of
intelligent observers, we should be unsurprised to find ourselves inhabiting
one of the rare hospitable spots. The
resulting Anthropic Principle can be
formulated in various ways (Barrow and Tipler 1986; Bostrom 2002). Earth is pretty special in its friendliness
to complex life – so in one sense we do occupy a specially privileged
spot. The Copernican Principle requires,
then, a qualification: We are not in a specially privileged location among locations capable of hosting intelligent
observers.[10]
So, what is an “intelligent observer”? One possibility is: any behaviorally
sophisticated entity in the sense of Section 2.
This characterization has some plausibility: If we were to interact with
them, we would naturally interpret them as making “observations”. We might then come to rely on these
“observations” in testing our theories.
Maybe members of an alien species send a signal naturally interpreted as
conveying the observation of a supernova from their home planet on
such-and-such a date. Employing this
definition of an “observer”, our originally stated version of the Copernican
Principle already has the required Anthropic qualification built in: “Among behaviorally
sophisticated entities [i.e., “observers”], we are not specially privileged
with respect to consciousness.”
Still, we have some sympathy
with the reader who wants to insist that only conscious entities rightly count
as “observers”. An Anthropic “observer” qualification
would then generate a considerably weaker Copernican Principle of
Consciousness: “Among conscious behaviorally
sophisticated entities, we are not specially privileged with respect to
consciousness.” This is a plausible
principle, and it isn’t vacuous, for it still suggests that we wouldn’t have
especially more or better consciousness.
However, this principle is insufficient for our purposes. It would no longer capture the Copernican
idea that it would be suspiciously lucky if we happened to have the right stuff
for consciousness while almost all of our alien peers – similar in behavioral
sophistication when viewed from outside – lack it.
Imagine switching observer
and observed: Some alien entity (perhaps conscious) observes human beings and various
other behaviorally sophisticated entities in other locations. Absent some particular grounds for regarding
us as special, this observing alien presumably ought to see us as just one of
the bunch. This neutral, rather than
Earth-centered, perspective is the correct, Copernican perspective. We shouldn’t regard ourselves as special,
unless there’s some evidence that we are special that could be recognized from
a neutral, outside perspective.
Otherwise we err like the self-absorbed teen who regards their
adolescent anguish and yearnings as somehow invisibly unique. Our original, unmodified Copernican Principle
of Consciousness reflects this neutral perspective, while implicitly capturing
an Anthropic qualification specified in terms of observers as behaviorally
sophisticated entities.
Concern D. What
justifies treating behaviorally sophisticated entities as the reference group,
rather than, say, entities containing carbon or having two eyes? One might imagine a variant of the Copernican
Principle of Consciousness according to which among entities containing carbon,
we are not specially privileged with respect to consciousness. Such a principle would seem to prove too much
since, plausibly, we are specially privileged with respect to consciousness,
among that class of entities.[11]
We have two avenues of response.
One approach is to apply the Anthropic qualification: Among entities containing
carbon who are also observers, we are not specially privileged
with respect to consciousness. Arguably, this plausible principle is the
appropriately Anthropically qualified carbon version of the Copernican
principle.
Second, behavioral sophistication, unlike being
made of carbon, directly targets the heart of the issue. If mainstream
scientific theories are correct, we already know that we are special, with
respect to our consciousness, among all entities containing carbon. Thus
Copernican reasoning doesn’t apply. Consider the parallel case concerning
position: If we have good evidence that we are at the center of the
universe, we can’t use the Copernican Principle to defeat that evidence.
Copernican Principles only justify default assumptions absent
countervailing evidence. Our focus on behavioral sophistication reflects the
target question it is our project to address, and on which we lack direct
evidence: not whether we are special among carbon-containing entities but
rather whether very differently constructed, but behaviorally sophisticated,
aliens would or would not be conscious.
We hope the reader finds the
fundamental idea plausible. There would
be something odd, something strikingly un-Copernican, in supposing that among
all the thousands or millions or trillions of behaviorally sophisticated
entities that have arisen in the universe, only we, or only we and a small
proportion of others, are fortunate enough to be conscious, while the rest merely
seem to feel love, to experience
ecstasies of poetry, to share excited news of their discoveries, and to thrill
with reverent awe before towers to their gods, but are in fact as
experientially empty as toasters.
5. From the Copernican
Principle to Default Liberalism about Consciousness.
The Copernican Principle of
Consciousness supports the conclusion that we are not specially privileged with respect to consciousness. It doesn’t follow that we aren’t slightly privileged. A royal flush in a single draw rightly draws
suspicions of funny business, but two pairs could easily enough be chance. Maybe only 10% of behaviorally sophisticated
entities are conscious? Then we wouldn’t
have to be too lucky to be among
them.
Let’s define Default Liberalism about consciousness
as the view that, absent specific grounds for doubt, when confronted with an
apparently behaviorally sophisticated entity, we would be warranted in
attributing consciousness. We propose
that symmetry and simplicity justify moving from the Copernican Principle of
Consciousness to Default Liberalism. If
some behaviorally sophisticated entities are in fact non-conscious, that
shouldn’t just be an arbitrary fact about them. We should expect there to be some striking
difference between conscious and non-conscious aliens that constitutes specific
grounds for doubt.
Suppose we discover an
otherwise apparently behaviorally sophisticated alien species that lacks
behavior suggestive of complex self-representations. Or suppose we discover a species that appears
to lack flexible learning. It’s not
unreasonable to worry that such striking general deficits might reflect
corresponding deficits in consciousness.[12] Our approach can handle these cases in one of
two ways. We could define “behavioral
sophistication” so as to exclude such entities: Behavior doesn’t count as
sophisticated unless it’s suggestive of complex self-representation and
flexible learning. Alternatively, we
could liberally describe such entities as behaviorally sophisticated if their
behavioral patterns are advanced enough in other ways. Doubt is justifiable either way we run the
case: in the former, because they are not behaviorally sophisticated, in the
latter, because the difference is plausibly important enough to cancel the
default in default liberalism.
Here we can’t avoid some
broad commitments about what features plausibly do and do not justify canceling
the default. Among the differences that
plausibly matter are differences that suggest major deficits in cognitive
function (relative to the class of Earthly entities assumed to be conscious). Without taking a stand on exactly what
types of cognitive function track the existence versus absence of
consciousness, virtually all mainstream scientific theories of consciousness
treat conscious states or their neural correlates as functionally significant
in the production of intelligent or goal-directed behavior. This relationship needn’t be exceptionless: A
person with locked-in syndrome might be conscious with little or no behavior,
and conversely a “consciousness mimic” (see discussion below) might achieve
impressive behavioral sophistication without consciousness. But fundamentally, the reason we think rocks
aren’t conscious is that their behavior doesn’t suggest a cognitively complex
relationship with their environment. If
we discover an alien world teeming with life, we will look for consciousness
first among the entities whose behavior warrants the attribution of complex
cognitive mechanisms and last among entities that passively sit. As we emphasized in the introduction, we aim
to avoid commitment to particular scientific and metaphysical theories of
consciousness, but committing to some
sort of association between consciousness and cognitive function is plausible
on a very wide range of mainstream theories.
Accordingly, we can distinguish
ungrounded from grounded doubt about an entity’s consciousness. If we discover that some arbitrarily
encountered alien entity, as behaviorally sophisticated as we are, operates by
hydraulics instead of electrical neurons, we might entertain some ungrounded
doubt about their consciousness. Might
hydraulics not actually support consciousness, as opposed to our wonderful
neurons? The sophisticated behavior suggests
consciousness, given the connection between complex behavior and consciousness
in folk psychology and in most mainstream theories of consciousness; but we
might still wonder if we are epistemically justified in making the leap of
attributing consciousness to an alien entity built so differently from us. The Copernican Principle of Consciousness,
combined with considerations of symmetry or simplicity, provide a defeasible,
default reason to make that leap.
6. Mimicry Arguments Against
Robot Consciousness.
We will now shift focus to
what we will call Mimicry Arguments against Robot Consciousness. In the context of Copernican Default
Liberalism about consciousness, Mimicry Arguments constitute specific grounds
for doubt about the consciousness of a particular class of entities, which we
will call consciousness mimics. We can be Copernican liberals about the
consciousness of apparently behaviorally sophisticated entities in general
while reserving skepticism about a subclass of mimics.
Skeptics about AI
consciousness commonly warn against mistaking imitations of consciousness for
the real thing. Emily Bender and
colleagues (2021), for example, characterize Large Language Models, like
ChatGPT, as “stochastic parrots” that copy the form of language with no
understanding of its meaning. Johannes
Jaeger (2023/2024) recommends replacing the phrase “artificial intelligence”
with “algorithmic mimicry” to reflect the reality (as he sees it) that AI
systems merely mimic intelligence without actually possessing it. John Searle argues that computers only
“simulate” mental processes, and “no one supposes that a computer simulation of
a storm will leave us all wet” (1984, p. 37-38). Kristin Andrews and Jonathan Birch (2023)
warn us against too readily attributing sentience to AI systems that have been designed
specifically to mimic human responses.
Daniel Dennett (2023) warns against language models as “counterfeit
people”.
Some of the best-known
philosophical thought experiments meant to cast doubt on AI consciousness
implicitly or explicitly suggest a mimicry structure. In Searle’s (1980) “Chinese room”, a person
who does not know Chinese but who has an extremely large rulebook is
hypothesized to interact with the outside world in a manner indistinguishable
from a real Chinese speaker by accepting Chinese characters as inputs and
issuing other Chinese characters as outputs in accord with the advice of the
rulebook. Block’s (1978/2007) “Chinese
nation” thought experiment (relatedly, the “Blockheads” of Block 1981) features
a large group of people who control a body by consulting a bulletin board in
the sky and following simple if-then rules so as to duplicate the functional
structure of a real human being for an hour.
Without entering into the details of these thought experiments or
assessing how convincing they are, we note a common feature: The systems in
question are specifically designed to imitate human behavior.
Mimicry arguments can at
least sometimes be excellent grounds for doubt about the consciousness of an AI
system, for the same reason that we might reasonably doubt that an actor is
genuinely sad if we learn that they are imitating the superficial signs of
sadness. However, advocates of mimicry
arguments have not generally contextualized their arguments in a theory of
mimicry that explains why this is so.
The remainder of this essay addresses this challenge, then explores how
to reconcile mimicry-based concerns about AI consciousness with our Copernican
Principle.
7. Mimicry in General.
Suppose you encounter
something that looks like a rattlesnake.
One possible explanation is that it is a rattlesnake. Another is that it mimics a rattlesnake. Such
mimicry can arise through evolution (other snake species that evolve the
superficial appearance of a rattlesnake to discourage predators) or through
human design (a rubber rattlesnake).
Normally, it’s reasonable to suppose that things are what they appear to
be. But this default assumption can be
defeated – for example, if there’s reason to suspect sufficiently frequent
mimics.
In biology, deceptive mimicry (for example Batesian
mimicry) occurs when one species (the mimic) resembles another species or
entity (the model) in order to mislead another species such as a predator (the
receiver). For example, viceroy
butterflies evolved to visually resemble monarch butterflies in order to
mislead predator species that avoid monarchs due to their toxicity. Similarly, gopher snakes evolved to shake
their tails in dry brush in a way that resembles the look and sound of
rattlesnakes. Chameleons and cephalopods
mimic their physical environment to mislead predators into treating them as
unremarkable continuations of that environment.
Social mimicry occurs when one
animal behaves similarly to another animal for social advantage. For example, African grey parrots imitate
each other to facilitate bonding and to signal in-group membership, and their
imitation of human speech arguably functions to increase the affection of human
caretakers. In deceptive mimicry, the superficially
imitated feature normally does not correspond with the model’s relevant
trait. The viceroy is not toxic, and the
gopher snake has no poisonous bite. In
social mimicry, even if there is no deceptive purpose, the signal might or
might not correspond with the trait suggested by the signal: the bird might or
might not belong to the group it is imitating, and Polly the parrot might or
might not really “want a cracker”.[13]
All mimicry thus involves
three traits: the readily observable trait (S2) of the mimic, the corresponding
observable trait (S1) of the model, and a further feature (F) of the model that
is normally indicated by the presence of S1 in the model. (In the Polly-want-a-cracker case, F might be
the intention to utter a sentence with that propositional content.) All mimicry also involves three protagonists:
a model, a mimic, and a receiver whose pattern of reaction to S2 explains the
resemblance of S2 to S1. Crucially, in
mimicry, the resemblance of S2 to S1 is explained by a potential receiver’s
reaction to that resemblance, given that the receiver treats S1 as normally
indicating F, rather than being explained directly by the presence of F is the
mimic. See FIGURE 1.
FIGURE 1: The mimic’s possession of a readily observable feature S2 is
explained by its resemblance to observable feature S1 in the model, because of how
a receiver, who treats S1 as indicating F, responds to the resemblance of S1
and S2. S1 reliably indicates F in the
model, but S2 need not reliably indicate F in the mimic.
Six clarifications:
(1.) The protagonists need
not be distinct. For example, in
parrots’ social mimicry the model might also be the receiver. Alternatively, if you mimic someone just to
delight yourself with your mimicry skills, then you are both mimic and
receiver.
(2.) The receiver need not
be fooled. The parrot owner doesn’t
mistake Polly’s speech for human speech.
In the wild, the parrot’s mate does not mistake its partner’s singing
for its own.
(3.) Although in the
simplest cases of deceptive mimicry, the receiver will react similarly to S1
and S2, the receiver might in principle react very differently to S1 and S2 if the
receiver is capable of distinguishing them.
A parrot might react very differently to its own song than to its mate’s
imitation of its song. A public clown
might mimic a depressed person’s gait, generating appreciation and laughter in
the audience. Mimicry requires only that
the resemblance of S2 to S1 be explained by the reactions of potential
receivers, given that the receivers treat S1 as an indicator of F.
(4.) The mimic need not lack
feature F. The viceroy might happen to
be toxic to some predator species. As
long as S2 in the mimic is better explained by its resemblance to S1 than by
the presence of F in the mimic, it’s still mimicry.
(5.) Emulation or imitation
alone is insufficient for mimicry. One
animal might imitate another for any of a variety of reasons, for example, to
duplicate their success in a challenging task. Mimicry distinctively requires a specific
relationship to a receiver.[14]
(6.) Mimicry involves a gap
between S1 and F. If I imitate your
holding up three fingers by also holding up three fingers, I have copied or
imitated you, but I have not necessarily (in our intended sense) mimicked
you. If the three fingers are intended
as a signal of membership in the Cool Kids Club and I raise them in imitation
of you so as to trick the doorkeeper, then I have mimicked you.[15]
Mimicry in the intended
sense is thus a specific, causally complex relationship in which S2 in the
mimic is better explained by a
complex relationship involving model, mimic, receiver, and the receiver’s
tendency to treat S1 as an indicator of F in the model than by a direct
relationship between S2 and F in the mimic.
Mimicry arguments against the consciousness of AI systems rely,
implicitly or explicitly, on the presence of this causal structure.
8. Mimicry in Artificial
Intelligence.
Consider a child’s toy that
says “hello” when powered on. We can
treat this as a simple case of mimicry. English-speaking
humans are the models, the toy is the mimic, S1 and S2 are the sound “hello”, F
is the intention to greet, and the receiver is a child who reacts by hearing the
sound as an English “hello”. In human
models, “hello” normally (though not perfectly) indicates an intention to greet. But of course the child’s toy has no
intention to greet. (Maybe the toy’s
designer, years ago, intended to craft a toy that would “greet” its user when
powered on, but that isn’t the toy’s
intention.) S2 is better explained by
having been modeled on S1, because of an anticipated receiver who hears the
toy’s sound as similar to a human “hello”, than as an indicator of F.
Large Language Models like
GPT, PaLM, LLaMA, and
Claude are more complex but are – at least in their simpler instantiations –
structurally mimics, as has been argued by, for example, Bender (Bender and
Koller 2020; Bender et al. 2021), Millière (2022), and
Jaeger (2024). Our account of mimicry renders
this claim more precise. If you ask
ChatGPT “What is the capital of California?”, it might respond “The capital of
California is Sacramento.” Treating this
text string as S2, the appropriate explanation of why ChatGPT exhibits S2 is
that its outputs are modeled on human-produced text that also correctly
identifies the capital of California as Sacramento, designed with a receiver in
mind who will interpret that output linguistically. While human-produced text with that content
(S1) reliably indicates the producer’s knowledge that Sacramento is the capital
of California (F), we cannot infer the presence of F from S2 – not, at least,
without substantial (and in this case dubious) further argument.
When a simple Large Language
Model outputs a novel sentence not present in the training corpus, S2 and S1
will need to be described more abstractly (e.g., “a summary of Hamlet” or even
just “text interpretable as a sensible answer to an absurd question”), but the
underlying considerations are the same.
The LLM’s output is modeled on patterns in human-generated text and can
be explained as mimicry of those patterns, leaving open the question of whether
the LLM has the further features we would attribute to a human being who gave a
similar answer to the same prompt. To
the extent an LLM is further shaped by reinforcement that rewards answers to
the extent they seem humanlike, the mimicry structure is reinforced.
We can call something a consciousness mimic if it exhibits readily
observable features suggestive of consciousness whose presence is best
explained by having been modeled on similar observable features of another
system, for the sake of a receiver who responds to those features in a way that
reflects the fact that they treat the features, when they occur in the model,
as indicating a conscious state or capacity. Note that this account of mimicry can apply to
both evolved and intentionally designed or artificial beings: all that is
required is that the etiology of the feature appeals to the right kind of
modelling.
ChatGPT and the “hello”-toy are consciousness
mimics in this sense. Their observable
language-like features (their S2s) are modeled on us, for the sake of a certain
sort of receiver with a certain set of interests and background knowledge. In us, the modeled features (the S1s)
reliably indicate the further feature (F) of consciousness (humans who say
“hello” or answer questions about the capital of California are normally
conscious), but there needn’t be any such reliable connection in a
consciousness mimic. Most “social AI”
programs, like Replika, are similarly consciousness
mimics, combining the structure of Large Language Models with superficial signals
of emotionality through an avatar with an animated, expressive face. This adds detail to the mimicry but does not
change the fundamental mimicry relation.
In our current technological
climate, any AI system that outputs a diverse range of interpretable text or
speech will almost inevitably rely on mimicry.
The best explanation of the shape of the text strings or the sound
structure will involve modeling the superficial features (such as word
co-occurrence relationships) of another entity (us) for the sake of a receiver
(also us) who interprets them in accord with their understanding of linguistic
capacities in model cases. Their S2s are
modeled on human S1s for the sake of interpretability by human receivers, with
no structure that ensures the presence of F in the mimic.
Contrast this with children’s
language learning. Children learn
language through imitation, and sometimes mimicry, but the primary structure is
not mimicry. If you say “blicket” in the
presence of a blicket and the child repeats “blicket” back to you, their
behavior arises in part from their (presumably conscious) appreciation of the
fact that the novel word refers to the novel object (Sobel, Yoachim,
Gopnik, Meltzoff, and Blumenthal 2007).
Their S2 resembles your S1, for the sake of a receiver (you), but S2
indicates their own underlying F, and you rightly interpret it as such. Children’s crying, demands for food, and
insistent “no” are not S2s explained by S1s linked to Fs in the model. Rather, they are direct signals (normally) of
genuine unhappiness, desire, and reluctance.
Similarly, Alex the parrot’s utterances of “four corner paper” in the
presence of square pieces of paper – though drawing on the same capacities that
underlie parrot mimicry – function as true signals in the ordinary, direct
sense (Pepperberg 1999). Childhood
linguistic mimicry occurs, but it is different from ordinary language
leaning. One example would be in shared
pretense, when a child says, “I’ve got to hurry or I’ll be late to work!”,
theatrically grabbing a pretend briefcase.
The situation becomes more
complicated as the models become more complicated. To the extent models are trained not to
behave like humans but rather to give the “right” answers or to act
successfully toward goals, and to the extent they incorporate elements like
maps and calculators that have representational rather than mimicry functions, they
diverge from being pure mimics. The
Mimicry Argument might still apply to the extent their fundamental structure
relies on superficial mimicry of S1s without reproduction of the architectures
that produce the S1s in the human case – but its application becomes more
complicated. They are no longer 100%
“parrot” (or underground octopus[16]).
However, they are still beings whose
properties are designed to mimic, perhaps not consciousness directly, but other
properties of human beings whose possession presupposes consciousness.
9. The Structure of the
Mimicry Argument.
The default in Default Liberalism
about the consciousness of behaviorally sophisticated entities can thus be
canceled in at least two ways: (a.) if the behavior suggests a sufficiently
important cognitive deficit, or (b.) if the entity is a consciousness mimic.
In particular, the Mimicry
Argument can be expressed as follows:
(1.) In cases of mimicry, the
appearance of S2 in the mimic is not normally sufficient grounds to infer the
presence of F in the mimic.
(2.) An important range of AI
systems are consciousness mimics.
(3.) Therefore, in this range of
AI systems, we cannot infer consciousness from the (S2) features that, were their
analogs (S1) to occur in humans, would normally indicate conscious states (F).
As discussed, S2 mimicry traits aren’t incompatible with the existence, in the
mimic, of the F traits normally indicated by S1s in the model. A viceroy might be toxic; a parrot mimicking
a group signal might belong (in fact, usually does belong) to the targeted group. But establishing that requires further
evidence or argument. In the case of
current and near-future AI systems, no further evidence or argument is likely
to merit scientific consensus on their consciousness.
The Mimicry Argument doesn’t
establish that consciousness mimics are definitely non-conscious. It only undercuts the inference from
superficial appearance (S2) to underlying reality (F). Perhaps a compelling argument can be made for
AI consciousness in some range of current or near-future cases; but given the
degree of uncertainty in consciousness science, we are disinclined to think any
such argument would warrant scientific and philosophical consensus.[17]
10. Are AI Systems
“Behaviorally Sophisticated” in the Sense Relevant to the Copernican Principle?
In Section 2, we described
an entity as “behaviorally sophisticated” if it is capable of complex
goal-seeking, complex communication, and complex cooperation, where
“goal-seeking”, “communication”, and “cooperation” are all understood in purely
behavioral terms, without commitment to the existence of consciousness or any
particular underlying cognitive architecture.
Are current or near-future AI systems behaviorally sophisticated in this
sense, and if so, would the Copernican Principle of Consciousness then apply to
them?
For alien cases, we favor a
somewhat relaxed sense of “behavioral sophistication”. As contemplated in Section 5, differently
structured entities might be capable of approximately human-level goal-seeking,
communication, and cooperation while having cognitive deficits that we would
find shocking (e.g., major deficits in learning capacities or
self-representational capacities). We
human beings also have remarkable cognitive weaknesses. Consider the Wason Selection Task. In this task, you are presented with four
cards. Each card, you are told, has a
letter on one side and a number on the other side. Two cards are letter-face up: “A” and
“C”. The other two are number-face up:
“4” and “7”. What card or cards must you
flip over to determine whether the rule “If a card has a consonant on one side,
it has an even number on the other side” is true? The large majority of ordinary people, even
if highly educated, fail to recognize the correct answer as “C” and “7”.[18] We wouldn’t want an alien species kicking us
out of the club of behaviorally sophisticated entities because of our –
possibly to them shocking – weakness in simple formal logical reasoning. Entities with different cognitive structures
will very likely have very different patterns of cognitive strength and
weakness. A proper Copernican
perspective won’t insist that alien intelligence look too much like our own.
Applying this same
liberality to AI cases, we don’t want to insist that present and realistic
near-future AI systems as they exist on Earth lack approximately human-level
behavioral sophistication. On some tasks
– such as walking across uneven surfaces and identifying objects in blurry,
partly occluded photographs – human beings (so far) substantially exceed AI
systems. On other tasks – such as
playing chess and multiplying large numbers – AI systems substantially exceed
even expert humans (interpreting “playing chess” and “multiplying large
numbers” purely behaviorally). It might
be tempting for opponents of AI consciousness to rest their case on insisting
that AI systems are not as sophisticated as human beings in some crucial range
of tasks. This is the “cognitive
deficit” grounds for canceling Default Liberalism. However, this move relies on correctly
specifying the relevant tasks, and correctly specifying the relevant tasks is a
challenge. AI enthusiasts with some
justification object to a history of moving goalposts. Back when success in chess seemed like the
pinnacle of human expertise, skeptics about AI intelligence sometimes held
chess up as a special sign of human intelligence. The Turing-inspired emphasis on the ability
to hold humanlike conversation also increasingly looks like it might be a poor
indicator, or at least one that will have to be specified in a highly precise
manner that is potentially difficult to justify. The Mimicry Argument allows AI skeptics to
express their skepticism about AI consciousness while (a.) conceding that AI
might be overall as behaviorally sophisticated as humans and (b.) without
committing to any particular type of especially revelatory behavioral benchmark
that AI cannot pass or any specific theory about the cognitive structures associated
with consciousness. These are, we
submit, attractive features of the Mimicry Argument.
Could the pressures on a
consciousness mimic lead it to develop genuine consciousness? The case for yes depends on the
interests and sophistication of the receiver.
If the pressures of natural selection are severe and specific enough to
generate complex goal-seeking, communication, and cooperation of the sort
described in Section 2, those same pressures will almost certainly produce
complex cognitive structures of the sort practically required to sustain
goal-seeking, communication, and cooperation – for example, long-term memory,
self-representation, and linguistic understanding. Fantastical creatures, such as “Blockheads”
(Block 1981) might be behaviorally identical to a naturally selected
intelligent organism, but such things could never evolve: The most efficient
way to reliably act, over evolutionary time, as if one has a long-term memory
is to actually have a long-term memory. In
contrast, the behavioral sophistication of a consciousness mimic need only be
good enough to achieve the ends of mimicry – for example, to fool or delight a
receiver. We are often delighted by obviously-less-than-perfect
mimics (the “hello”-toy) and, as the recent history of language models
suggests, more sophisticated mimics can sometimes easily fool us.
Different types and degrees
of similarity between S2 and S1 are required for different mimicry
relationships. Deceptive mimicry can
become an arms race between a mimic motivated to appear similar to the model in
the eyes of the intended dupe and an intended dupe motivated to acquire the
capacities to distinguish mimic from model.
In the limit, the mimic could only “fool” a god-like receiver by
actually acquiring feature F. Fortunately
for the mimics, we are far from god-like. We thus see no reason to suppose that
consciousness would be required for successful mimicry when receivers are imperfectly
discerning or when the motive isn’t deception.
As current large language models show, some impressive feats of mimicry
can be achieved without long-term memory or sophisticated self-representation. These are major cognitive differences that,
even independently of mimicry considerations, might trigger reasonable
cancelation of Default Liberalism as described in Section 5.
There is a reason most
features of most evolved organisms aren’t explained by mimicry. Mimicry is a complex relationship that
requires a special set of evolutionary pressures to evolve and sustain, often
parasitic on the simpler true signals or cues of more prevalent non-mimic
species. If viceroys outnumber monarchs
(as they now do in California) the evolutionary advantage of mimicking the
monarch’s wing coloration decreases. And
technologically, we speculate, mimicry structures might not prevail in the long
term. In general, human design goals
might be met more fully and efficiently by creating systems that really have F
than by creating systems that imitate the superficial signs of F. If we create an artificial entity that really
has underlying linguistic representations of real-world events and a conscious
understanding of itself and its interlocutors (whatever architecture that
requires), that will presumably be a more reliable and satisfying interaction
partner, for a broad range of purposes, than today’s Large Language Models. Similarly, if our space alien neighbors send
behaviorally sophisticated robotic probes across the galaxy, it’s a reasonable
guess that those probes will be designed to explore, measure, experiment, communicate,
self-repair, respond to emergencies, and perhaps devise theories on the fly
that guide their exploratory choices, rather than that they will mainly be
designed as mimics to fool or delight some receiver by virtue of their
resemblance to some model entity.
Since mimicry traits evolve
less commonly than non-mimicry traits, and since the evolution or design of
mimicry traits requires a highly specific set of purposes or conditions, we suggest
that it is reasonable to default assume that a newly encountered behaviorally
sophisticated extraterrestrial entity will not be a consciousness mimic, even
if it has a technological origin – though we might hold this default assumption
only lightly. The crucial epistemic
distinction between mimicry and non-mimicry explanations of behavioral sophistication
crosscuts distinctions between design vs. evolution and artifice vs. biology.
11. Rejecting the Parity
Principle.
We sympathize with those who
reject a simple sort of organic chauvinism to argue that AI isn’t conscious. The most obvious alternative to organic
chauvinism is one that treats alien and robot cases symmetrically – holding,
for example, that the same external, behavioral criteria should be applied to
both aliens and robots in assessing them for consciousness. If Behavioral Test X would be sufficient to warrant
attribution of alien consciousness, then – on this way of thinking –
attribution of consciousness would also be warranted to robots who similarly
pass Behavioral Test X. Similarly, on
this way of thinking, theorists who emphasize interior architecture should attribute
consciousness to aliens and robots based on the same internal, architectural
criteria. If Architecture Y would
warrant attribution of consciousness to aliens, then attribution of
consciousness would also be warranted to any robot possessing Architecture
Y. Mixed behavioral/architectural views
might also apply identical criteria to aliens and robots. According to what we’ll call the Parity Principle, in evaluating an alien
or robot for consciousness, we should apply the same behavioral and/or
architectural criteria.
We acknowledge the potential
appeal of the Parity Principle. But to
accurately assess this principle, we need to distinguish it from a closely
related metaphysical thesis.
Metaphysically, if an alien and a robot have identical structures with
respect to what matters for consciousness according to the correct theory of
consciousness (whatever that may be), then it cannot be the case that one but
not the other has consciousness. This
metaphysical parity thesis is highly plausible.
The Parity Principle we
reject differs from the plausible metaphysical thesis in two respects. First, the Parity Principle is epistemic
rather than metaphysical. It concerns
not what must be true if robots and aliens share whatever behavioral or
architectural features matter to consciousness; rather it concerns what we are warranted in believing about alien and
robot consciousness given our ignorance
about the correct theory of consciousness.
Second, it applies independent of stipulations about the internal
workings of the entities involved.
These two differences
matter. Even if we know virtually
nothing about the internal workings of a particular (individual or type of)
behaviorally intelligent alien, we can apply the Copernican Principle, at least
by default (that is, assuming we have no evidence it is a consciousness mimic
or in some other respect likely to be a member of a disprivileged
class with respect to consciousness).
Even if we know virtually nothing about the internal workings of a
particular (individual or type of) robot or AI system, if we know that the
superficial features suggestive of consciousness arise from a mimicry
relationship of the sort described in Section 7, we are licensed to withhold
inference to consciousness, barring some positive evidence that the superficial
features are in fact coupled with consciousness.
The Copernican and Mimicry
Arguments yield different results for alien and robot systems not because of
some intrinsic inferiority of engineered to naturally evolved life but rather
because aliens and a wide range of current and near-future robots – normally or
as a default presumption – have very different histories. Consequently,
very different explanations of their structure and behavior can be
warranted. Aliens, normally or as a
default presumption, are naturally evolved entities not modeled on other
entities (or if designed, not specifically designed as mimics). In contrast, in present and near-future
cases, the most convincing robot conversationalists will be consciousness
mimics. Contra the Parity Thesis, this
difference generates an epistemic asymmetry in attribution of consciousness to
aliens and robots, even if the aliens and robots have similar outward behavior
or similar interior architectures (at some level of descriptive specificity).
This is not to endorse the teleofunctionism of, for example, Millikan (1984), Dretske (1995), and Neander (2017). On a teleofunctional
view of mental properties, etiology also matters: Donald Davidson has a mind
and his molecule-for-molecule identical chance-generated Swampman
duplicate lacks a mind because of their different origins. But such teleofunctionalism
concerns the metaphysical relationship of mental properties to their physical
realizers at a given time. Our approach is
neutral about the metaphysical importance of an entity’s history while
remaining committed to the history’s epistemic importance.
12. Generalizing the
Copernican Argument to Less Sophisticated Aliens.
In Section 2, we advanced a
high bar for “behavioral sophistication” because we were seeking a sufficiency
criterion for default attributions of alien consciousness which would be
appealing to a broad range of consciousness theorists, from the very liberal
(who hold that consciousness is widespread on Earth) to the very conservative
(who hold that consciousness is limited to just one or a few species). But this is a movable parameter in our
theory. Our Copernican argument can be
adapted to animal cases involving lower standards of behavioral sophistication
as follows.
Barring some reason to think
we are exceptional, Earth would be a strangely lucky place if only Earth has
conscious non-linguistic life forms and in almost every other region of the
observable universe, similarly complex or sophisticated life forms were
non-conscious. Thus, a generalization of
the Copernican Principle is possible.
With respect to whatever types of animal life you are justified in
regarding as conscious on Earth (and different readers might reasonably favor
different views), it is reasonable to default assume that similarly complex or
sophisticated alien life forms would also be conscious.[19]
References:
Anderson, Bruce, and Marinus
L. de Jager (2020). Natural selection in
mimicry. Biological Reivews, 95, 291-304.
Andrews, Kristin, and
Jonathan Birch (2023). What has
feelings? Aeon (Feb 23) https://aeon.co/essays/to-understand-ai-sentience-first-understand-it-in-animals.
Barrow, John D., and Frank
J. Tipler (1986). The anthropic cosmological principle. Oxford University Press.
Bender, Emily M., and
Alexander Koller. 2020. Climbing towards
NLU: On Meaning, Form, and Understanding in the Age of Data. Proceedings
of the 58th Annual Meeting of the Association for Computational Linguistics,
5185-5198.
Bender, Emily M., Timnit
Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell (2021).
On the dangers of stochastic parrots: Can language models be too
big? FAccT '21: Proceedings of the 2021 ACM Conference on Fairness,
Accountability, and Transparency, 610-623.
Birch, Jonathan (2022). The search for invertebrate consciousness. Noûs,
56, 133-153.
Block, Ned (1978/2007). Troubles with functionalism. In N. Block, Consciousness, function, and representation. MIT Press.
Block, Ned (1981). Psychologism and
behaviorism. Philosophical Review, 90, 5-43.
Bondi, H. (1968). Cosmology. Cambridge University Press.
Bostrom, Nick (2002). Anthropic
bias. Routledge.
Builes, David, and Caspar
Hare (2023). Why aren’t I part of a whale? Analysis,
83, 227-234
Butlin, Patrick, Robert Long, Eric
Elmozino, et al. (2023). Consciousness in Artificial Intelligence:
Insights from the science of consciousness.
ArXiv:2308.08708.
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 (1), 1291-1302.
Carruthers, Peter, and Rocco
Gennaro (2001/2023). Higher-order theories
of consciousness. Stanford Encyclopedia of Philosophy (fall 2023 edition).
Carter, Brandon (1974). Large number coincidences and the anthropic
principle in cosmology. In M.S. Longair, ed., Confrontation
of cosmological theories with observational data. D Reidel.
Chalmers, David J.
(2023). Could a Large Language Model be
conscious? ArXiv:2303.07103.
Clarkson, Chris, Bruce
Bassett, and Teresa Hui-Ching Lu (2008).
A general test of the Copernican Principle. Physical
Review Letters, 101, 011301.
Dehaene, Stanislas, Hakwan Lau, and Sid Kouider
(2021). What is consciousness, and could
machines have it? Science, 358 (6362), 486-492.
Davidson, Donald
(1984). Inquiries into truth and interpretation. Oxford University Press.
Dennett, Daniel C.
(1987). The intentional stance. MIT
Press.
Dennett, Daniel C.
(2023). The problem with counterfeit
people. The Atlantic (May 16). https://www.theatlantic.com/technology/archive/2023/05/problem-counterfeit-people/674075.
Dretske, Fred (1995). Naturalizing
the mind. MIT Press.
Elga, Adam (2000).
Self-locating belief and the Sleeping Beauty problem. Analysis,
60, 143-147.
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.
Frankish, Keith (2016). Illusionism as a theory of
consciousness. Journal of Consciousness Studies, 23 (11-12),11-39.
Ginsburg, Simona, and Eva
Jablonka (2019). The evolution of the sensitive soul. MIT Press.
Godfrey-Smith, Peter
(2016). Mind, matter, and metabolism. Journal
of Philosophy, 113, 481-506.
Gott, J. Richard, III. (1993). Implications of the Copernican principle for our future prospects. Nature, 363, 315–19.
Harnad, Stevan (2003). Can a machine be conscious? How? Journal of Consciousness Studies, 10
(4-5), 67-75.
Jaeger, Johannes
(2023/2024). Artificial intelligence is
algorithmic mimicry: Why artificial “agents” are not (and won't be) proper
agents. ArXiv:2307.07515.
Jamie, Gabriel A.
(2017). Signals, cues and the nature of
mimicry. Proceedings of the Royal Society B, 284 (1849): 20162080. doi:
10.1098/rspb.2016.2080.
Kammerer, François
(2019). The illusion of conscious
experience. Synthese, 198, 845-866.
Kriegel, Uriah (2019). The value of consciousness. Analysis,
79, 503-520.
Lacki, Brian C. (2023). The Noonday argument: fine-graining, indexicals, and the nature of Copernican reasoning. International
Journal of Astrobiology, 22, 354–398.
Lee, Geoffrey (2019). Alien subjectivity and the importance of
consciousness. In A. Pautz and D. Stoljar, eds., Blockheads! MIT Press.
Levy, Neil (2024). Consciousness ain’t
all that. Neuroethics, 17 (21), https://doi.org/10.1007/s12152-024-09559-0.
Millière, Raphaël (2022). Moving beyond mimicry in artificial
intelligence. Nautilus (Jul 1). https://nautil.us/moving-beyond-mimicry-in-artificial-intelligence-238504.
Millikan, Ruth (1984). Language,
thought, and other biological categories.
MIT Press.
Neander, Karen (2017). A mark
of the mental. MIT Press.
Papineau, David (2024). Consciousness
is not the key to moral standing.
Unpublished manuscript.
Pepperberg, Irene M.
(1999). The Alex studies. Harvard
University Press.
Scharf, Caleb (2014). The
Copernicus complex. Farrar, Straus
and Giroux.
Schneider, Susan
(2019). Artificial you. Princeton
University Press.
Schwitzgebel, Eric
(2016). Phenomenal consciousness, defined
and defended as innocently as I can manage. Journal
of Consciousness Studies, 23 (11-12), 224-235.
Schwitzgebel, Eric
(2024). The weirdness of the world.
Princeton University Press.
Searle, John R. (1980). Minds, brains, and programs. Behavioral
and Brain Sciences, 3, 417-457.
Searle, John R. (1984). Minds,
brains, and science. Harvard
University Press.
Sebo, Jeff, and Robert Long
(2023). Moral consideration for AI
systems by 2030. AI and Ethics. https://doi.org/10.1007/s43681-023-00379-1.
Seth, Anil K. (2021). Being
you. Dutton.
Seth, Anil K., and Tim Bayne
(2022). Theories of consciousness. Nature
Reviews Neuroscience, 23, 439-452.
Shepherd, Joshua
(2024). Sentience, Vulcans, and zombies:
The value of phenomenal consciousness. AI & Society, https://doi.org/10.1007/s00146-023-01835-6.
Sherratt, Thomas N. (2008). The evolution of Müllerian mimicry. Naturwissenschaften 95, 681-695.
Shevlin, Henry (2020). General intelligence: An ecumenical heuristic
for artificial consciousness research? Journal of Artificial Intelligence and
Consciousness, 7, 245-256.
Silver, David, Thomas
Hubert, Julian Schrittwieser, et al. (2018). A general reinforcement learning algorithm
that masters chess, shogi, and Go through self-play. Science,
362, 1140-1144.
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.
Sobel, David M., Caroline M.
Yoachim, Alison Gopnik, Andrew N. Meltzoff, A. N.,
and Emily J. Blumenthal (2007). The
blicket within: Preschoolers’ inferences about insides and causes. Journal
of Cognition and Development, 8, 159-182.
Turing, Alan M. (1950). Computing machinery and intelligence. Mind,
59, 433-460.
Udell, David Billy, and Eric
Schwitzgebel (2021). Susan Schneider’s proposed
tests for AI consciousness: Promising but flawed. Journal
of Consciousness Studies, 28 (5-6), 121-144.
Wason, P. C. (1968). Reasoning about a rule. Quarterly
Journal of Experimental Psychology, 20, 273-281.
Whiten, Andrew, Nicola McGuigan,
Sarah Marshall-Pescini, and Lydia M. Hopper (2009). Emulation,
imitation, over-imitation and the scope of culture for child and chimpanzee. Philosophical
Transactions of the Royal Society B, 364,
2417-2428.