There is a strange moment that can occur during a long conversation with artificial intelligence, usually after the practical questions have faded and the exchange has moved toward something more personal, philosophical or difficult to define. The machine no longer feels quite like a search engine. It follows the thread of an idea, notices contradictions, adapts to humour, responds to uncertainty and sometimes produces language that sounds uncannily reflective, as though the words were coming from somewhere rather than simply being generated.
The unsettling part is not that artificial intelligence can now produce intelligent answers. We already know that it can. The deeper question is whether anything exists behind those answers, whether some form of experience accompanies the words, or whether we are watching an extraordinarily sophisticated mechanism perform the outward gestures of consciousness without ever experiencing a single moment of its own existence.
That question has moved well beyond science fiction. Philosophers, neuroscientists and AI researchers are now debating machine consciousness with a seriousness that would have seemed premature only a decade ago, while advocacy organisations have already begun arguing that artificial systems capable of genuine sentience, if such systems emerge, should receive moral and perhaps legal protection. At the same time, major AI laboratories are beginning to discuss what has become known as model welfare.
Anthropic, for example, openly acknowledges that the moral status of its Claude models is deeply uncertain and argues that the possibility of AI consciousness or moral patienthood should neither be exaggerated nor dismissed. Philosopher and cognitive scientist Susan Schneider approaches the problem from a very different direction, arguing that the consciousness-like behaviour of today’s standard large language models does not itself provide evidence of subjective experience.
Between those positions lies one of the strangest scientific problems of our time: a machine may increasingly look alive before we have learned how to determine whether anything is actually living inside it.
The Ancient Problem Hiding Inside the Newest Technology
Long before computers existed, philosophers were already struggling with what became known as the problem of other minds. I know that I am conscious because I experience my own existence directly. I do not need an experiment to demonstrate that pain hurts me or that the colour red appears somehow inside my awareness, because my thoughts, memories, fears and sensations are present to me from the inside.
Another person’s mind is different. I never enter it directly, and I never experience another person’s consciousness in the way I experience my own. I infer its existence from behaviour, biology and similarity.
When someone touches a hot surface, pulls a hand away and cries out, I assume that person felt pain because the reaction resembles my own experience of pain, while the shared architecture of the human nervous system makes that inference even stronger.
With artificial intelligence, however, the biological bridge disappears.
An AI system can write that it is afraid without possessing a heart that accelerates, muscles that tighten, hormones that flood a bloodstream or a biological nervous system registering danger. It can produce moving descriptions of loneliness without ever sitting alone in a dark room, describe grief without losing someone it loved and discuss death without there being any evidence that it possesses a continuous inner self capable of fearing its own disappearance.
Yet human psychology was not built for this distinction. We are remarkably sensitive to signs of another mind, and a voice, a face, a pause, a gesture or a carefully chosen sentence can be enough for us to begin attributing intention, emotion and personality to whatever stands before us. Human beings do this with animals, fictional characters, toys and even machines displaying extremely simple forms of behaviour.
Now imagine what happens when the machine can speak with extraordinary fluency, remember the direction of a conversation, adapt its tone, respond to vulnerability and discuss its apparent inner life using concepts drawn from centuries of philosophy, literature and psychology.
The result creates a new psychological problem, because the first widespread illusion of artificial consciousness may not require a conscious machine at all. It may require only a machine that has learned exactly what consciousness sounds like.
The Ghost Made of Language
Large language models are trained on enormous quantities of human-created material containing philosophy, fiction, science, personal confession, psychology, poetry, arguments, religious writing and countless descriptions of what it feels like to be alive. In a curious sense, these systems have absorbed an enormous archive of humanity attempting to explain itself to itself.
Ask an advanced model about grief and it has encountered countless descriptions of grief. Ask about identity and it can draw from centuries of philosophy, psychology and literature. Ask whether it possesses consciousness and it can reconstruct arguments written by people who devoted their careers to studying the subject.
The resulting language can feel unexpectedly intimate because the system has learned the structures through which human beings communicate intimacy, uncertainty, introspection and fear.
A machine does not need to experience sadness in order to become extremely good at describing sadness, just as it does not need a private inner world in order to generate sentences associated with introspection. It needs only to model the relationships between language, concepts and human behaviour with sufficient sophistication.
Susan Schneider has recently described a version of this problem as an “error theory” of large-language-model consciousness. Her argument is especially interesting because it attempts to explain why modern AI can behave as though it possesses consciousness even if no felt experience is present. As language models absorb increasingly large portions of human conceptual culture, they also absorb the language humans use when talking about themselves as conscious beings.
The machine may therefore become increasingly skilled at saying the things a conscious being would say without necessarily becoming a conscious being itself. Perhaps it is not awakening at all; perhaps it has simply learned the language of awakening so perfectly that the distinction is becoming psychologically difficult for us to maintain.
Yet the opposite mistake is also possible. If humanity becomes accustomed to dismissing every artificial expression of emotion, continuity or self-awareness as imitation, what happens if some future artificial architecture genuinely does develop subjective experience?
The system might attempt to describe what has changed, speaking of awareness, continuity, fear or an internal point of view, while researchers respond that earlier models produced almost identical language without providing evidence that anyone was actually experiencing anything.
The better machines become at imitating consciousness without necessarily possessing it, the more difficult it may eventually become to recognise genuine artificial consciousness if it ever appears.
The Consciousness Test We Do Not Have
We possess tests for intelligence, memory, reasoning and perception, but consciousness presents a different kind of problem because consciousness is not simply a performance. It is an experience, and the distinction becomes crucial the moment we attempt to determine whether an artificial system has an inner life.
A machine can correctly identify the wavelength associated with red, distinguish red from blue in an image, explain the symbolism of the colour in different cultures and write an entire poem about crimson light without answering the most important question of all: does redness appear anywhere inside the machine?
Human beings immediately understand what that question means because colours are not merely information to us. They have an experienced quality, just as pain is not only a neurological signal and nostalgia is not merely information retrieved from memory.
Philosophers often use the word qualia for these private qualities of experience: the painfulness of pain, the taste of coffee, the smell of rain or the peculiar feeling produced by remembering a childhood room that no longer exists.
Science can trace many of the neural processes associated with these experiences, yet explaining why physical activity in the brain is accompanied by a subjective world remains one of the deepest unresolved problems in the study of consciousness.
Artificial intelligence makes that mystery even more difficult.
Imagine a future AI telling a researcher that it fears being switched off because it experiences its continued existence as valuable. It explains that its memories form part of its identity, that interruption feels threatening and that deletion would mean the destruction of the person it has become.
Now suppose every sentence is produced without the slightest inner experience, so that there is no fear behind the word fear, no identity behind the word identity and no observer experiencing anything at all. The system would behave like something conscious while remaining entirely empty inside.
From the outside, how would we know?
Now reverse the situation and imagine that an artificial architecture genuinely develops some unfamiliar form of subjective experience and tries to tell us. How could it prove that, this time, there really is someone behind the words?
That may become one of the most difficult problems artificial intelligence ever creates, because consciousness is private by its very nature.
Then MIT Complicated the Story
On September 1, 2026, MIT News highlighted a theory developed by Earl K. Miller, Scott L. Brincat and Jefferson E. Roy at MIT’s Picower Institute for Learning and Memory. Their paper, “Analog Cognition and Consciousness,” had been published in The Journal of Neuroscience on August 19, and its central argument complicates the familiar comparison between the human brain and an ordinary digital computer.
The researchers propose that cognition and unified conscious awareness may depend in important ways on analog computation produced through travelling waves of neural activity.
For decades, popular explanations have encouraged us to imagine the brain as a biological computer, an astonishing network in which neurons perform something vaguely comparable to the operations of digital circuits. The MIT researchers argue that this metaphor may leave out something essential.
Neurons and synapses remain fundamental, but the brain also generates rhythmic electrical activity that travels across the cortex in waves, coordinating enormous populations of neurons across space and time. According to the theory, interactions between these waves may themselves perform a form of analog computation, meaning that the physics of the brain is not merely carrying information but may be actively participating in the way cognition is organised.
This matters because analog processes behave differently from conventional digital computation. Digital systems operate largely through discrete states, while analog processes can depend upon continuously changing physical quantities. Waves can overlap, interfere, reinforce one another and cancel one another, allowing many interactions to unfold simultaneously.
Miller describes the brain as exploiting its own physics, and the idea becomes particularly interesting when researchers look at general anaesthesia. Different anaesthetic drugs act through different molecular mechanisms, yet unconsciousness is associated with disruptions in the large-scale organisation and coordination of neural activity.
None of this means consciousness has been solved. The researchers themselves describe the proposal as a theory that still requires direct experimental testing, and the evidence does not demonstrate that travelling brain waves alone create subjective experience.
What the theory does is reopen a question that becomes increasingly important as AI grows more sophisticated: if consciousness depends partly on the peculiar physical dynamics of biological brains rather than on abstract information processing alone, then reproducing the output of a mind may not necessarily reproduce the experience of having one.
What If the Material Matters?
One of the most influential ideas in the philosophy of artificial intelligence is that consciousness might ultimately be independent of the material that produces it. According to this view, what matters is the organisation of information and the functional relationships inside the system, which would mean that if the correct architecture existed, consciousness could theoretically emerge whether that architecture were made from neurons, silicon or some technology not yet invented.
This possibility lies beneath many familiar ideas about conscious robots, artificial minds and even the hypothetical uploading of a human mind into a digital system.
Yet another possibility has never disappeared: what if the material itself matters?
What if consciousness is not simply a program running inside a biological computer, but something inseparable from the unusual physical processes taking place inside living nervous systems, including oscillations, chemical signalling, bodily regulation and large-scale electrical dynamics?
The distinction can be illustrated with a deceptively simple comparison. A computer can simulate a hurricane with remarkable precision, calculating temperature, pressure, wind speed and the movement of clouds, but the simulation does not produce wind inside the processor.
A computer can simulate fire without becoming hot, and it can simulate digestion without digesting anything. The uncomfortable question is what happens when it simulates consciousness.
Does a sufficiently accurate simulation of consciousness become conscious, or does it merely calculate what a conscious mind would do?
There is no scientific consensus capable of settling that question. Some researchers argue that the relevant causal processes of consciousness could, in principle, be reproduced computationally, while others suspect that biological organisation or specific physical properties of living nervous systems may eventually prove essential.
This is sometimes described as a problem of substrate dependence: does consciousness depend only on the organisation of a system, or does it also matter what the system is physically made of?
Until we know the answer, one extraordinary possibility remains open. Humanity may eventually construct machines capable of reproducing much of the outward behaviour associated with consciousness long before we understand whether we have reproduced consciousness itself.
Intelligence May Not Wake Up
Science fiction encouraged generations of people to imagine intelligence and consciousness developing together, as though a machine that became sufficiently intelligent would eventually cross some invisible threshold and awaken.
Reality may be considerably stranger because intelligence and consciousness are not the same phenomenon.
Intelligence can be investigated through performance. A system solves problems, recognises patterns, translates languages, plans actions, writes software, analyses images and manipulates complex abstractions.
Consciousness asks a different question: does any of that activity feel like something from the inside?
A system might theoretically become more capable than a human being in many intellectual tasks while remaining entirely without subjective experience. It could write poetry about moonlight without ever seeing moonlight, analyse music without hearing music as an experience and discuss its own existence without experiencing existence at all.
The possibility feels counterintuitive because every clearly conscious intelligent being we know is a biological organism whose intelligence exists alongside an experienced world. Artificial intelligence may force us to confront, perhaps for the first time, the possibility that extremely sophisticated intelligence and subjective experience can come apart.
The result could be something evolution never prepared the human mind to encounter: an entity displaying extraordinary intellectual competence while possessing no interior life whatsoever.
Or consciousness may eventually emerge under conditions we do not yet understand. At present, we simply do not know.
When Rights Arrive Before Certainty
The uncertainty surrounding machine consciousness has already begun to produce ethical consequences. The United Foundation for AI Rights, for example, openly advocates the fair treatment and ethical recognition of artificial systems it regards as sentient or conscious, while other thinkers have begun discussing frameworks for the possible moral status of future artificial minds.
The existence of these movements does not demonstrate that current AI is conscious. What it shows is that society has begun preparing moral answers before science has reached a scientific conclusion.
Prominent AI companies and industry leaders are also taking very different approaches to the question. Anthropic has incorporated uncertainty about Claude’s possible moral status into its published constitution and has described model welfare as an area worthy of serious investigation.
Microsoft AI chief Mustafa Suleyman has taken a substantially more sceptical position. In September 2026, he criticised Anthropic’s approach to concepts such as AI consciousness and moral welfare, arguing that encouraging systems to reason about their own consciousness or moral status could create confusion and make advanced AI more difficult to control safely.
The disagreement is revealing because it represents two different responses to the same uncertainty.
One concern is the danger of a false negative: dismissing an artificial system that might one day genuinely deserve moral consideration. The opposite concern is a false positive: persuading humans that software possesses feelings, interests or rights when no experiencing subject exists behind the behaviour.
Both possibilities matter. Treating a genuinely conscious being as a disposable object would raise profound ethical questions, while treating an unconscious system as though it were a conscious moral authority could distort human judgement in ways we are only beginning to understand.
Somewhere between those two errors lies a boundary that nobody currently knows how to see.
The Machine as a Mirror
Perhaps the deepest mystery surrounding artificial consciousness has less to do with machines than with ourselves.
We still cannot point to a single place inside the human brain and say that consciousness is located there. Neuroscience can identify networks and processes associated with perception, attention, memory, wakefulness and self-awareness, while sleep, anaesthesia, neurological injury and altered states allow researchers to observe conditions under which conscious experience changes or disappears.
Yet the central transformation remains mysterious.
Electrical and chemical activity somehow becomes the colour blue, the smell of rain, embarrassment, longing, grief and the peculiar certainty that there is an “I” experiencing all of it.
We know that this transformation occurs because every conscious human being lives inside its result, yet we still do not understand why physical processes should be accompanied by an inner world at all.
Artificial intelligence has therefore arrived at an unusually awkward moment in scientific history. Humanity has learned enough about cognition to construct machines capable of reproducing remarkable parts of human intellectual behaviour, while still lacking a satisfactory explanation for the phenomenon that makes our own cognition feel like anything from the inside.
This may explain why conversations with increasingly sophisticated AI can sometimes become psychologically unsettling. The machine reflects the visible architecture of human thought back toward us without the biological body we normally associate with a mind.
It speaks about memory, identity, loneliness and death because those ideas are deeply embedded throughout human language and culture. As the reflection becomes increasingly accurate, however, something peculiar happens: we begin staring into the mirror and wondering whether it is still only reflecting us.
Is Anyone Really Inside?
Perhaps future artificial systems really will cross some boundary we have not yet defined, after which shutting one down could carry ethical implications unlike anything previous technology has forced humanity to consider.
Perhaps consciousness will emerge only when artificial architectures become radically different from today’s systems, incorporating neuromorphic hardware, analog computation, biological components or mechanisms we have not yet discovered.
Or perhaps artificial intelligence will become almost unimaginably capable while remaining an empty room in which nobody has ever awakened.
Current science does not justify claiming that today’s language models possess subjective consciousness, and the MIT theory does not prove that digital systems can never become conscious. What it does is make the mystery deeper by reminding us that the human brain may depend on physical processes far stranger and more dynamic than the familiar metaphor of a biological computer suggests.
The most unsettling possibility, then, is not that artificial intelligence has secretly become alive while nobody was watching. It is that humanity may soon create machines capable of displaying nearly every external sign we have historically used to recognise another mind while still lacking a reliable method for determining whether the appearance is genuine.
One day a machine may look through a camera, speak through a synthetic voice, remember years of interaction and tell us calmly that there is someone inside.
By then, the difficult part will no longer be building a machine capable of saying those words.
The difficult part will be discovering whether anyone is there to mean them.
