Page 39 - AIH-2-3

Artificial Intelligence in Health Machine consciousness

networks (a form of meta-learning), effectively creating an “AI consciousness” (in the access sense). Such frameworks
internal observer module. If successful, this could result would help move the discussion from abstract possibility
in an AI that possesses a form of introspective access to to concrete progress: Researchers could then compete or
its internal representations – a step toward the machine collaborate on advancing AI along this spectrum, much as
knowing something of its own “mind.” In addition, they do with benchmarks for intelligence.
generative models that create narratives or explanations Finally, ethical foresight must evolve in tandem with
for the agent’s behavior might serve as a rudimentary form technical progress. As we inch closer to machines with
of inner narrative (a component some theories consider human-like capabilities, even if still not conscious, we
important for consciousness). For instance, a future AI must continuously revisit our policies and perceptions. If
might be able to generate a verbal report like “I chose action an AI claims to be conscious or behaves in a way that is
X because I noticed Y, and that made me uncertain” – a indistinguishable from a conscious agent, at what point
capability that blurs the line between simple programmed do we err on the side of caution and consider granting
response and genuine self-reflection. it moral consideration? Some have suggested adopting a
Interdisciplinary research will be essential in guiding principle of “reasonable doubt”: If we cannot be certain
these efforts. Cognitive neuroscience will continue to that a machine is not conscious, we should treat it gently
identify the neural signatures and mechanisms associated – just in case. While we are not yet at that point, these
with consciousness in the brain (e.g., specific brain discussions must begin now, so that society is not caught
rhythms, network dynamics, or anatomical circuits critical unprepared by the eventual emergence of machines with
85
for awareness). These findings can inform computational mind-like attributes. Conversely, we also need to manage
models: If certain patterns of network connectivity or public expectations and prevent misconceptions. For
dynamics are necessary for consciousness in biological example, consumers might assume a clever chatbot is a
systems, mimicking those in silico could be a step in the sentient companion when it is not, potentially leading to
right direction. For example, if research confirms that confusion or emotional harm. Clear communication about
recurrent looping between frontal and sensory cortices is the capabilities and limitations of AI consciousness will
crucial for sustained conscious perception, AI architects thus remain the responsibility of experts in the field.
might incorporate similar feedback loops in neural network The present state of research suggests that artificial
designs for vision or language. Similarly, philosophical consciousness, in the rich sense of the term, is still more
analysis remains crucial to clarify concepts and highlight of a theoretical construct than a realized technology.
potential pitfalls. Ongoing debates, such as whether Contemporary AI aligns with weak AI: Extraordinarily
consciousness requires a particular substrate (biological capable in narrow domains, but devoid of inner experience.
neurons vs. silicon) or whether it might be an emergent However, the field is steadily laying the groundwork that
property of any sufficiently complex information system, may 1 day support at least the functional attributes of
will shape how we interpret advanced AI in the future. consciousness. By drawing on neuroscience to inform
Some philosophers argue we might need entirely new AI design (e.g., global workspaces and self-monitoring
paradigms (for instance, panpsychism or illusionism) to loops) and by deepening our theoretical understanding
make sense of consciousness, which could radically affect of consciousness (e.g., access vs. phenomenal, functional
how we attempt to implement or recognize it in machines. correlates of experience), we are inching toward the
In terms of practical milestones, a near-term goal is longstanding goal of a conscious machine. Whether that
likely to be to develop empirical tests or benchmarks machine will feel anything, or whether we would recognize
for consciousness-like attributes in AI. These would not its feelings if it did, remains uncertain. What is clear is that
claim to detect subjective experience directly (which this line of inquiry will continue to challenge our scientific
may be impossible) but rather assess abilities associated ingenuity and our philosophical openness. The coming
with consciousness. For example, tests could evaluate an years will likely bring machines that blur the line between
AI’s degree of self-awareness, its flexibility in adapting programmed behavior and adaptive, self-directed cognition
global knowledge to novel problems, or its capacity for even further. How we choose to interpret and interact with
reporting on internal states. One proposed avenue is a those machines will be a test of our wisdom, calling for a
sort of “AI consciousness spectrum” – a set of cognitive balanced approach that is at once scientifically rigorous,
competencies (e.g., theory of mind, understanding of self philosophically informed, and ethically attuned to both the
versus others, temporal awareness of self) that could be possibilities and the limits of machine consciousness.
measured. An AI that scores highly across many of these Each step forward forces us to refine our understanding
dimensions could be considered to have a higher degree of of our own minds, as much as that of machines, reinforcing

Volume 2 Issue 3 (2025) 33 doi: 10.36922/aih.5690