1 Introduction
The difference between a system with a first-person subjective
point of view and a system without one is the explanatory gap between
our physical theories and what it is to have a mental life (Godfrey-Smith 2019,
1–2). There is undeniably something it is like to be a person
working on an essay, and seemingly nothing it is like to be the chair
they are sitting in. Phenomenal consciousness, henceforth
p-consciousness, is the name for this experience of something it is
like to be something (Block 1995, 230).
Humans are confident in one another’s p-consciousness through verbal
reporting. We communicate in language, and the similarity between our
descriptions of experience convinces us that the others have something
like our own. We remain reasonably confident about the p-consciousness
of other mammals on the basis of neural similarity; mammalian brains
share the same key features as ours, only smaller (Birch 2020, 288). The
methodology breaks down for non-mammals, who may or may not possess
structures homologous to the mammalian brain. Writing off every other
animal class as a result ignores the possibility of multiple
realizability, where different underlying mechanisms perform the same
functions.
This essay examines the work of Simona Ginsburg and Eva Jablonka, who
have developed an approach that links certain forms of enhanced
learning to p-consciousness. They argue that any organism capable of
these forms of learning is p-conscious, and so has a subjective
experience of its own. The learning types are grouped under the label
unlimited associative learning (UAL): an organism can learn about its
environment and itself in an open-ended manner (Birch et al. 2020,
56). The constraints on UAL still allow that it is functionally
impossible for all of the possible associative links to be exhausted in
the finite lifetime of a real organism (Birch et al.
2020, 56).
To identify UAL with consciousness, Ginsburg and Jablonka use a cluster
of enhanced associative learning capacities that, in humans, are
believed to be facilitated by p-consciousness (Birch 2020, 291–92). These are:
(1) compound stimuli, where a stimulus is
built from more than one perceptual feature, in either a single sense
modality or several (Birch et al. 2020,
56); (2) novel stimuli, where the
conditioned stimulus can be novel to the organism;
(3) second-order conditioning, where the
organism can build up open-ended chains of association between stimuli
and actions (Birch et al. 2020,
56); (4) trace conditioning, where
there can be a time gap between the conditioned and unconditioned
stimuli; and (5) flexible, easily rewritable
associations with value, where stimuli can be positive or negative
and the organism can revise their value in response to environmental
change (Birch et al.
2020, 56).
These five capacities suggest how UAL might serve as the basis for a
research programme, once we have a clearer picture of how UAL relates
to a minimally sufficient functional description of p-consciousness.
That relation is the topic of the next section: is UAL p-consciousness,
or only an indicator of it? After that, the essay turns to the
implications of UAL for p-consciousness, and finally to some of the
support for, and criticism of, the proposal.
2 Natural or nominal kind
This section sets out UAL as Ginsburg and Jablonka and their
collaborators understand it, places it alongside a separate framework
due to Nicholas Shea and Tim Bayne, and uses that framework to assess
the relation between UAL and p-consciousness. It then considers what
that relation, whatever it turns out to be, implies for the wider
research programme.
Ginsburg and Jablonka’s theory is evolutionary, and proceeds in stages
marked by three transitions: from non-life to life, from automata to
p-consciousness, and finally to the rational-symbolic (Ginsburg and
Jablonka 2020, 158). An evolutionary transition marker signals
each transition. Once the diagnostic marker is present, the transition
has occurred, and the organism has gained a new mode of being not
possessed by its progenitors (Ginsburg and Jablonka 2020,
157–59). The relevant transition for this essay is the second,
from automata to p-consciousness; UAL is Ginsburg and Jablonka’s
proposed marker for it. The authors maintain that UAL is diagnostic in
the sense that it positively marks the presence of p-consciousness in
an evolutionary lineage (Ginsburg and Jablonka 2019,
226), and not necessarily anything more. If UAL is a good
minimal description of what is functionally required for
p-consciousness, however, the obvious question is whether it does not
essentially constitute being p-conscious.
To address this question, the essay turns to Shea and Bayne’s ‘natural
kind methodology’. The framework is concerned with the status of
p-consciousness itself: is p-consciousness an underlying natural
property that entails its perceived ‘symptoms’, in which case it is a
natural kind (Shea and Bayne 2010, 470); or is
it just a family of cognitive abilities with no underlying natural
property, i.e., a nominal kind like the concept of dirt (Shea and Bayne
2010, 470)? If p-consciousness is a natural kind, to use Shea
and Bayne’s example, it is more like hepatitis C than dirt (Shea and Bayne 2010,
470). Hepatitis C can be diagnosed without a near-perfect
detection method, provided enough symptoms are found and other causes
can be ruled out. The natural kind methodology takes the same line: if
p-consciousness is a natural kind, it can be detected with a diagnostic
toolkit. Identifying a cluster of properties does not by itself
guarantee that there is an underlying natural kind that best explains
the cluster. Even so, if p-consciousness is a natural kind, the
methodology supports more investigation than the pre-theoretic
approaches we already possess, such as verbal reporting (Shea and
Bayne 2010, 470). On that assumption we can collect a wide
array of evidence and assemble nomological clusters: sets of evidential
properties that are ‘instantiated together better than chance’ and
where ‘subsets of the cluster support induction to other elements of
the cluster’ (Shea and Bayne 2010,
471).
Ginsburg and Jablonka’s UAL seems to satisfy these criteria for a
nomological cluster. The five features of UAL overlap rather than
standing as separate capacities, and they have co-evolved with one
another (Birch et al. 2020,
56). For Birch, Ginsburg and Jablonka to treat UAL as a good
diagnostic, they must take it to require the functionally coupled
systems that result in p-consciousness (Birch et al. 2020,
57). In other words, they have identified a natural cluster of
evidential properties whose best explanation is an underlying natural
kind property, i.e., p-consciousness. UAL is therefore not equivalent
to p-consciousness; it is a nomological cluster best explained by the
existence of p-consciousness. Ginsburg and Jablonka are
p-consciousness realists, and intend their chosen cluster to provide
evidence of p-consciousness as a natural kind.
A consequence of formulating their evolutionary transition marker
within Shea and Bayne’s natural kind methodology is that the
pluralistic nature of the nomological cluster becomes explicit.
Ginsburg and Jablonka’s central thesis for UAL is that some forms of
learning are facilitated by p-consciousness (Ginsburg and Jablonka 2007,
219). Within the natural kind methodology, further evolutionary
transition markers (further nomological clusters) could be found and
added in support. The synthesis of the two frameworks gives a
comprehensive research programme: we are licensed to look for UAL in
non-human animals, and we are licensed to look for additional
evidential properties best explained by p-consciousness as a natural
kind. More interrelated evidential properties means more nomological
clusters, and so more empirical traction.
The next section turns to support for, and criticism of, Ginsburg and
Jablonka’s methodology and their realist stance on
p-consciousness.
3 Support and criticism
The first piece of supporting evidence for UAL comes from the fossil
record. Ginsburg and Jablonka argue that in the Cambrian record
(542–485 million years ago (Ginsburg and Jablonka 2019,
407)) we can see relevant features in some vertebrate organisms,
i.e., a nervous system (Ginsburg and Jablonka 2007,
222–23). Neural structure is strongly associated with UAL in
modern humans and other animals, and supports the five criteria of the
UAL nomological cluster. From an evolutionary perspective, Ginsburg and
Jablonka argue, the appearance of open-ended associative learning in
the Cambrian is unsurprising. The same period saw a sudden
diversification of life and a sharp increase in the number of species,
known as the Cambrian explosion (Godfrey-Smith 2016,
11). Ginsburg and Jablonka attribute that explosion to the
development of UAL (Ginsburg and Jablonka 2019,
419–20). UAL endowed organisms with the ability to adapt their
behaviour within a single lifetime, which in turn drove morphological
diversity: an animal that discovered and exploited a novel food source
was likely to stay and reproduce in that environment (Ginsburg and Jablonka 2019,
420–21). Its offspring were likely to exploit the same food
source, and from there the pressures of natural selection opened up
further morphological niches (Ginsburg and Jablonka 2019,
421). This is a compelling hypothesis, and its explanatory power
adds weight to Ginsburg and Jablonka’s broader evolutionary approach.
It also leans on the continuity principle of the broader biogenic
approach: that there is a continuous, unbroken lineage from the simpler
cognitive forms of the past to our most complex cognition today (Lyon 2006, 15). Locating the
origins of p-consciousness deep in the past in this way also leaves
room for a wide distribution of p-consciousness in the present.
The first criticism worth examining is the application of UAL as an
indicator of p-consciousness to artificial intelligence. It is fairly
conceivable that an AI system, either now or in the near future, could
fully demonstrate UAL. Would such a system count as p-conscious? The
position of Birch, Ginsburg and Jablonka is simply no (Birch et al. 2020,
56). They argue that UAL is an evolutionary transition marker
for biological systems, not for computer systems (Birch et al. 2020,
56). I find this a somewhat weak argument. It implies that what
matters is biology, and the fact that some evolution has occurred; both
premises deserve scrutiny. The premise of biological necessity sits
awkwardly with the weight they place on functionalism. Ginsburg and
Jablonka treat UAL as a marker of p-consciousness precisely because it
is multiply realisable, so that non-mammalian animals with very
different neural structures can become candidates for p-consciousness.
If what is essential about p-consciousness is the functional
organisation captured by UAL, it is hard to see why AI should be ruled
out as a candidate for realising that organisation in a different
material substrate. The appeal to evolution is, if anything, easier to
push back on. Many AI systems already are evolutionary algorithms;
large numbers of candidate AIs are generated in discrete generations
and a form of Darwinian selection is applied (Mirjalili et
al. 2020, 3), with the ‘fittest’ AI as the end product. So the
evolutionary requirement is not, on its own, a principled barrier to AI
being assessed under the UAL transition marker. From a realist
position, settling the case would require knowing more about what sort
of natural kind p-consciousness actually is. Theory-light frameworks,
those that do not operate with a fully specified theory of
consciousness (Birch 2022, 140–41),
such as Ginsburg and Jablonka’s and Shea and Bayne’s, deliberately
avoid commitments that would invite overt anthropomorphism. It is
therefore inconsistent to reject artificial cases out of hand when they
have undergone evolution and functionally realise UAL.
A further criticism concerns the markers themselves. As an indicator of
p-consciousness, UAL cannot deliver verdicts on organisms that do not
display the five criteria listed in the introduction, which raises a
risk of false negatives: animals that are p-conscious may be labelled
as not p-conscious. The worry is that a criterion such as
(4) trace conditioning may genuinely indicate
p-consciousness in humans, but it is not clear that its absence in a
non-human animal entails the absence of p-consciousness (Shevlin 2021, 308). An
organism that relies heavily on one sense has little reason to develop
or retain a capacity for multisensory learning (Shevlin 2021, 308). The
connection between learning and p-consciousness can be questioned more
broadly. Henry Shevlin notes that many paradigm examples of
p-consciousness in humans, such as perceptions and bodily sensations,
seem unconnected to our capacity for learning (Shevlin 2021, 308). The
point has merit, but UAL does at least provide a basis for assessment
and diagnosis, where the human evidence about perception and bodily
qualia rests on verbal report and so cannot be applied to non-human
animals. The right response is probably to use UAL only positively, to
ascribe p-consciousness to animals capable of it, and not to rule out
organisms that lack it. Identifying further nomological clusters could
later strengthen the negative case to a tolerable level of
uncertainty.
The final criticism is to ask whether p-consciousness is a natural kind
at all. The natural kind methodology that Ginsburg and Jablonka’s work
feeds into relies on this premise. If p-consciousness is in fact a
nominal kind, the methodology will not pick out p-consciousness; it
will only pick out unlimited associative learning. Daniel Dennett
proposes a qualia-irrealist stance that equates p-consciousness with
access consciousness (Dennett 2015, 2). Access
consciousness, henceforth a-consciousness, is the availability of a
perception for reasoning and the rational guidance of action (Block 1995, 227).
Dennett argues that although it seems intuitive that the qualia of
subjective experience exist, this is in fact only an intuition (Dennett 2004,
60–61). He further argues that we do not know the limits of
sufficiently detailed physical theories, so it is at least possible
that a piece of supposedly non-scientific knowledge, such as the
‘redness’ of red, could be determined scientifically (Dennett 2004,
60–61). There is therefore a real possibility that
p-consciousness is not a natural kind. If it is in fact a nominal kind,
Ginsburg and Jablonka’s research programme is in trouble: UAL testing
would only tell us about varying capacities of a-consciousness, and
would not provide the definitive diagnostic of p-consciousness it is
intended to be.
4 Conclusions
Ginsburg and Jablonka’s evolutionary transition marker theory fits the
description of the natural kind methodology set out by Shea and Bayne,
which also makes it natural to read Ginsburg and Jablonka as
p-consciousness realists. Shea and Bayne’s framework lets us address
the question of whether UAL is equivalent to p-consciousness. The
answer here is that it is not. UAL is a nomological cluster of
evidential properties, best explained on the realist position by the
existence of p-consciousness as a natural kind. The fit between the
two frameworks also opens the door to a more comprehensive research
programme along natural kind lines.
Ginsburg and Jablonka’s evolutionary perspective offers a compelling
hypothesis about the early origins of p-consciousness and its role in
the Cambrian explosion. It implies that the distribution of
p-consciousness across the animal phyla may be extensive, with
p-consciousness realised in many different ways across many different
neural architectures.
I also resist the move to dismiss UAL out of hand when it is applied to
AI, given the functionalist commitments Ginsburg and Jablonka rely on
elsewhere. They appeal to multiple realisability so that animals with
neural structures very different from those of mammals can count as
candidates for p-consciousness. AIs are already produced using
evolutionary algorithms, so excluding them from the evolutionary
transition marker framework looks more like biological chauvinism than
a principled distinction.
UAL is also susceptible to false negatives, and should be supplemented
with further nomological clusters to bring that risk down. The
diagnostic should be applied with care to exotic organisms that differ
significantly from us; some UAL criteria are not obviously reasonable
for creatures in environments where the relevant form of learning would
be of little use.
Dennett’s critique leaves open the possibility that p-consciousness is
not a natural kind. Even so, the prospect of further empirical research
under the natural kind methodology is too useful to abandon at this
stage. Researchers into non-human p-consciousness can both apply the
UAL diagnostic to animals and search for additional nomological
clusters to reduce uncertainty in their assessments. On that basis UAL
and the natural kind methodology that underpins it are a worthwhile
investment of time and resources going forward.