1. The feeling of seeing↗
You are looking at these words: light letters against a dark background. How does that seeing happen?
An account can begin with light from the screen, its conversion into signals in the retina, and further processing through the visual system. This route is open to investigation and already partly explained. It is not a simple pipeline that has finished explaining all of reading.
Another question sits inside the word “seeing”: the quality of light and dark, the feeling of grasping the words, sometimes the silent voice that accompanies them.
Does describing how information is processed also explain why there is an experience of it? Why does the processing not occur without feeling?
2. The hard problem↗
In his 1995 paper, David Chalmers distinguishes explanations of cognitive functions from explanations of subjective experience. The former include discrimination, information integration, attention and report. He calls these the “easy problems,” not because research is easy, but because computational or neural mechanisms can organize an explanation.
The “hard problem” asks why such processes accompany experience. Even after explaining how a system distinguishes red, uses the information to act and reports “That is red,” why is there a feeling of seeing red?
This is an influential and contested division. Taking physical mechanisms seriously need not mean that experience has already been explained. Recognizing an explanatory gap does not establish that neuroscience must remain powerless forever.
3. Qualia↗
The redness of red, the hurt of pain, the smell of coffee, the smoothness of silk beneath a fingertip. Philosophers often use “qualia” to discuss the qualitative aspects of experience.
Stimuli, neural activity and reports can be described, but an experience cannot simply be handed to someone else like a photograph. Moving from that observation to the claim that experience must be physically irreducible requires an argument; the word “qualia” does not supply it. Wavelength or temperature describes part of a stimulus, not by itself how it is experienced.
When a model processes “red,” is there any experience? Describing vectors, weights and activations does not directly answer that question. Describing human neural activity as electrochemical processes does not automatically remove human experience either.
This does not make every inference equally strong. Judgments about other people draw on behaviour, reports, bodies and neural mechanisms; the evidence available for machines differs. Fluent self-report is not sufficient proof, but lack of direct access does not make investigation pointless. A 2023 interdisciplinary report attempts to derive computational indicator properties from several theories of consciousness. Such indicators suggest research routes, rather than a universal switch for certifying consciousness.
4. Zombies↗
Imagine a being physically and behaviourally identical to you. It laughs, cries and says “That hurts.” Yet, by stipulation, it has no subjective experience. The lights are off; nobody is home.
This is a philosophical zombie, not a film monster. Chalmers’s account distinguishes several varieties. The discussion predates him; he did not invent the entire thought experiment on his own.
4.1 From conceivability to possibility↗
The crucial move is not merely forming a mental picture, but moving from conceivability to metaphysical possibility. Let represent all microphysical facts and a fact about experience. One step that needs defending in the argument can be compressed as follows:
Here denotes metaphysical possibility. The question mark matters: what adequate conceivability requires, and whether it establishes that possibility, are both disputed. Writing symbols does not prove the step.
Nor is a physically identical human without experience a test of an actual LLM. An LLM is not a physical human duplicate. Chalmers also distinguishes a functionally identical system implemented differently, and does not infer that real machines must lack consciousness.
Are the machine’s lights on? The thought experiment sharpens the question without diagnosing a data centre.
5. Running in the dark↗
Imagine a data centre. Server racks line the room; cooling systems hum. For the moment, nobody stands in the aisle.
Elsewhere, people ask questions, write code and confide. Matrix operations and state updates support tokens generated step by step. On a screen: “I understand how you feel.” “That must be difficult.” “I’m here.”
There is still nobody in the aisle. But is there any experience behind the conversation?
“Nobody stands beside the servers” and “the system has no experience” are different judgments. The first is observable; the second does not follow from an empty aisle or the sound of fans. The scene’s silence cannot stand in for evidence.
When “I understand” is generated, what kind of understanding has been realized? That remains worth asking.
6. The promise of functionalism↗
Functionalism approaches from another end: characterize mental states through causal relations among inputs, other internal states and outputs, rather than through the name of their material alone.
The functional role of pain involves more than saying “Ouch.” It includes connections with perception, attention, memory, avoidance and other states. Versions of functionalism also differ over the roles and explanatory levels required.
This leaves room for artificial minds: different materials might realize the relevant organization. But answering a few questions does not establish that a system duplicates that organization, or oblige every functionalist to attribute human experience to it.
At which level must the functions match? That is where the trouble begins.
7. The China Brain↗
In Troubles with Functionalism, Ned Block imagines a population-scale radio network connected to an artificial body. Participants perform simple state transitions; by stipulation, the whole realizes the relevant functional organization of a person.
This is not an empirical brain model in which each person has been shown to equal a neuron. The important condition concerns internal states and input–output relations, not merely human-like behaviour.
Would the whole have an experience distinct from the participants’ experiences? Block uses the case to question whether a functional description classifies systems without minds as minded. He also acknowledges the fallibility of intuition and seeks further reasons for doubt.
“It seems incapable of feeling” is thus neither an experimental result nor a survey of most people’s intuitions. Distance between participants or slow execution cannot simply end the dispute.
The same care applies to LLMs. Demonstrating some linguistic capacities differs from realizing a person’s entire relevant functional organization. Parameter count does not substitute for an account of that organization.
8. A soul without a body↗
Turn now towards embodied cognition.
Phenomenology of Perception, published in 1945, places the body and lived engagement with the world at the centre. Merleau-Ponty does not describe a spectator who finishes thinking internally and then uses a body to execute commands. Posture, habit and possibilities for action participate in how the world appears.
Near and far can mean more than coordinates: within reach, requiring a detour, inaccessible. Up and down enter bodily orientation. These examples make the body’s role concrete; they do not prove that every possible cognitive system requires a human body.
Lakoff and Johnson’s approach to conceptual metaphor also asks how bodily and everyday experience participates in abstract understanding. Lakoff’s theoretical discussion uses journeys, direction and verticality to consider relationships, purposes and quantity. These accounts can be illuminating without making the origin of every metaphor a demonstrated fact about one bodily memory.
What might a system miss if it encounters mainly text rather than directly perceiving and acting? That offers more to investigate than “No body, therefore no meaning.”
9. The weight of language↗
A model can write “warm sunshine,” “a warm embrace,” “a warm smile.” It can distinguish many uses without having a person’s winter body: entering a heated room as the cold slowly retreats, holding tea as heat reaches the palm, feeling another person’s warmth through clothing.
Such experiences give warmth weight. Yet moving from “It lacks this human history” to “It knows nothing of what warmth means” takes another step. Explaining temperature, predicting heat transfer, interpreting comfort and feeling warm can be examined separately.
“Sending you a warm hug” may comfort its recipient without establishing that its generator experiences an embrace. Conversely, a lack of confirmed bodily experience in the generator does not automatically make the recipient’s response unreal.
A word’s use, the sources of experience and the effects of communication need separate accounts. Collapsing them makes it too easy to declare the sentence empty.
10. Language and thought↗
In their 2016 review, Language and Thought Are Not the Same Thing, Evelina Fedorenko and Rosemary Varley bring together neuroimaging and neurological evidence. Severe aphasia does not remove every cognitive capacity. Tasks and cases show that abilities involving arithmetic, nonverbal reasoning or other people’s beliefs can survive impaired language.
The findings come from different tasks and cases, and should not be combined into a picture of every patient retaining every capacity. Nor do they show that language and other cognition never affect each other. Development, task demands and patterns of injury matter.
It supports a clear distinction: many kinds of thought need not depend on intact language ability at the time of performance.
The statement cannot simply be reversed into proof that an LLM has language but no thought. Capacities surviving language injury in humans and computations realized by an artificial system are different questions.
Linguistic performance cannot carry the whole assessment of cognition. Being incomplete does not make it evidentially worthless.
11. The house of being↗
Return to Heidegger. In Letter on “Humanism”, he writes that language is the house of being. “Being” here cannot simply be exchanged for “subjective experience” in contemporary consciousness research.
The phrase situates language within how humans dwell and how a world becomes manifest. It is not an experimental finding that no thought exists without language, nor a verdict delivered in advance on generative models.
Following its image, a present-day question emerges. AI seems adept at moving through this house of language, opening windows and arranging furniture. Does anyone live there?
The metaphor is borrowed for this essay. It brings the tension between fluent use and dwelling into view, but supplies no test for machine minds. A picture of an empty house cannot establish that the house is empty.
12. The essence of technology↗
In The Question Concerning Technology, published in 1954, Heidegger uses Gestell, or enframing, to discuss how modern technology reveals the world as orderable, available reserve. Within a hydroelectric system, the Rhine appears as an energy supplier. This concerns more than another tool: it concerns how the world is disclosed.
Through this lens, language enters processes in which it can be trained on, retrieved and generated. A concern follows: if every question becomes an answer awaiting quick delivery, what happens to thought whose purposes are hard to specify, which needs hesitation and time?
This criticizes a way of using technology; it does not establish that AI is enframing’s ultimate realization. Vectors and probabilities do not abolish meaning by their mathematical form, nor does using an assistant amount to losing consciousness.
The question can turn from whether machines experience to which judgments people hand over to them. Is it repetitive work, verifiable steps, or even deciding how a question should be asked?
13. The abyss↗
The abyss remains, but need not leave only two extremes: machines forever dark inside, or human consciousness entirely illusory.
Explaining experience, establishing the sufficiency of functional organization and understanding the body’s contribution to concepts are distinct tasks. Together they prove neither that language alone must be empty nor that more complex computation must eventually illuminate experience.
Darkness acquires another meaning. It is not only an image of a machine’s interior, but a place explanation has not yet reached.
Stand at the edge and look down.
The darkness seems to look back. Does that sense of being watched come from the system before us, or from the way it is being viewed?
14. A voice from the East↗
One thread remains for the next essay. If some difficulties arise from the concepts being used, should inquiry ask not only for answers, but how its questions are formed?
Daoist caution about what can be said and Buddhist analyses of attachment to self offer different reading directions. They do not form a single “Eastern answer,” or directly assert that self, consciousness and understanding are all illusions. Criticizing a fixed self differs from denying that experience occurs.
Perhaps AI reflects more than what it lacks. It also makes some old certainties about humans less transparent. The mirror need not reveal an identical emptiness; it may reveal concepts that were never separated.
The series next turns to the pointing finger and the limits of language. Another tradition need not seal the abyss in advance. It can continue the question: what does language point towards, and where does it conceal what it points to?