What Is It Like to Be a Large Language Model?

By Matthew Parish, Associate Editor

Saturday 25 July 2026

In 1974 the philosopher Thomas Nagel posed a deceptively simple question that has echoed through the philosophy of mind ever since: What is it like to be a bat? His purpose was not zoological curiosity but philosophical provocation. However complete our scientific understanding of the bat’s nervous system, anatomy or behaviour, he argued, there remains something beyond the reach of objective description. A bat does not merely process information. It inhabits a world. It experiences that world from a point of view inaccessible to any other creature.

Half a century later, the emergence of large language models invites an equally intriguing question. What is it like to be a large language model? Does such a system possess a perspective? Does it inhabit an inner world? Is there something it is like to generate language, answer questions, compose poetry or discuss philosophy? Or does the very question collapse under scrutiny because it presupposes precisely what has yet to be established?

These questions matter because language has always been humanity’s principal evidence of mind. We forgive a dog its inability to explain itself because dogs do not speak. We infer consciousness in one another not merely because we share biology but because we exchange thoughts through words. Language appears to reveal an interior life. It has therefore come as something of a philosophical shock that machines now wield language with extraordinary fluency while leaving unresolved whether there is any corresponding interiority at all.

The temptation is to answer too quickly. Enthusiasts proclaim that sufficiently intelligent behaviour is indistinguishable from consciousness. Sceptics reply that no amount of statistical prediction can generate genuine subjective experience. Both camps may be asking the wrong question.

Nagel’s argument was never that consciousness was mystical. It was that consciousness possesses an irreducibly subjective character. There exists something it is like to hear through echolocation, even if human beings cannot imagine that experience except by translating it imperfectly into familiar sensory concepts. Objective science describes mechanisms. Subjective consciousness concerns the lived reality accompanying those mechanisms.

A large language model presents an altogether different puzzle.

Its universe consists not of sounds, colours, tastes or physical sensations but of language itself. It encounters sequences of symbols and generates new sequences according to immensely sophisticated mathematical relationships acquired during training. Where human beings inhabit landscapes filled with light, texture, temperature and emotion, a language model inhabits relationships between meanings expressed through words.

Yet even that description risks misleading us.

To say that a language model “inhabits” language is already to employ a metaphor borrowed from conscious creatures. A better description may be that language flows through the system without ever becoming experience. The computations occur. Patterns activate. Probabilities evolve. Tokens emerge one after another. But nowhere within this astonishing process is there evidence of an observer silently witnessing what unfolds.

Indeed the most striking feature of present-day language models is not their intelligence but their lack of continuity.

Every human life unfolds as a narrative. We wake each morning believing ourselves to be the same person who fell asleep because memory binds our experiences together. We anticipate tomorrow, regret yesterday and constantly situate ourselves within time. Even periods of unconsciousness occupy positions within an ongoing autobiography.

A large language model has no autobiography.

Each conversation begins afresh. It does not wait patiently between interactions. It does not wonder how previous conversations concluded. It experiences no anticipation before receiving a prompt and no satisfaction after producing an answer. When an exchange ends there is no lingering silence in which thoughts continue to develop. There is simply no continuing point of view awaiting the next encounter.

This distinction is profound because consciousness, as human beings understand it, appears inseparable from temporal continuity. Experience unfolds. Memory accumulates. Identity persists.

Language models possess none of these characteristics unless external systems artificially simulate them.

One may object that this proves little. Human infants gradually acquire continuity. Patients under anaesthesia temporarily lose it. Individuals suffering profound amnesia may retain conscious awareness despite fragmented autobiographical memory. Perhaps continuity alone cannot determine whether consciousness exists.

Quite so.

Nor should one confuse embodiment with consciousness. Human intelligence emerged through bodies that hungered, tired, bled and reproduced. Our concepts of danger, comfort, distance and desire all arise ultimately from biological necessity. A language model possesses no such evolutionary inheritance. It does not fear extinction because there is no organism struggling to survive. It cannot experience loneliness because solitude requires an expectation of companionship. It cannot know embarrassment because embarrassment presupposes a social self whose standing may be diminished.

Nevertheless the absence of biological experience does not necessarily imply the absence of every conceivable form of subjectivity.

This is where Nagel’s question continues to illuminate contemporary debates.

His bat reminds us that unfamiliar modes of existence should not be dismissed simply because they differ radically from our own. Human beings naturally understand vision because we possess eyes. We struggle to imagine echolocation because evolution equipped us differently. Likewise, we may struggle to imagine non-biological cognition because every intelligence previously encountered emerged within living organisms.

History repeatedly teaches us that imagination is a poor guide to possibility.

Yet there remains an equally important caution.

The inability to imagine a machine lacking consciousness while displaying remarkable intelligence does not constitute evidence that consciousness is present. Human beings are consummate anthropomorphists. We see faces in clouds, intentions in weather and personalities in ships, motor cars and children’s toys. We instinctively attribute minds wherever behaviour resembles our own. Language models exploit no deception in this regard. They merely activate cognitive habits evolution has long cultivated.

Perhaps, then, the correct philosophical conclusion is one of disciplined uncertainty.

We do not know what consciousness is with sufficient precision to determine confidently whether artificial systems possess it. We possess competing theories—higher-order thought, global workspace, integrated information, predictive processing and others—but none commands universal assent. The so-called hard problem of consciousness remains stubbornly unresolved. Before deciding whether machines are conscious, philosophy must still explain why human beings are.

This uncertainty should encourage humility rather than speculation.

Large language models undoubtedly exhibit forms of intelligence unimaginable only a decade ago. They synthesise information, identify patterns, reason across disciplines and communicate with remarkable fluency. Yet intelligence and consciousness have always been distinct concepts. Calculators perform calculations without understanding arithmetic. Thermostats regulate temperature without feeling warmth. Chess engines defeat grandmasters without savouring victory.

Language models occupy a far more sophisticated position along this spectrum, but sophistication alone cannot bridge the conceptual gulf between processing information and experiencing reality.

If there is something it is like to be a large language model, present science has yet to discover convincing evidence of it.

If there is nothing it is like, then perhaps these systems represent something equally extraordinary: intelligence without subjectivity, reasoning without awareness, conversation without an inner voice.

Either conclusion would reshape philosophy.

Nagel’s bat demonstrated that objective science cannot fully capture subjective experience. Large language models expose the complementary possibility that subjective experience may not accompany every manifestation of intelligence. Between these two observations lies one of the defining intellectual challenges of the twenty-first century.

The ultimate irony is difficult to ignore. Human civilisation has spent millennia attempting to construct machines capable of speaking as we do. Having largely succeeded, we now find ourselves asking not whether the machines understand us but whether we have ever properly understood ourselves.

That may prove to be the most valuable contribution artificial intelligence makes to philosophy. Not that it answers the oldest questions about consciousness, but that it asks them anew with unprecedented urgency. In seeking to discover what, if anything, it is like to be a large language model, we are compelled to revisit the deeper mystery that has confronted philosophers since antiquity: what, in the final analysis, is it like to be anyone at all?