LLMs and the Collapse of Representational Interiority IV — The Collapse of the Inner Homunculus

Modern humans carry a peculiar creature inside their heads.

It has no official scientific status. No anatomy textbook contains a diagram of it. Neuroscience has never located it. Yet contemporary culture quietly assumes its existence almost everywhere.

This creature is the inner homunculus:

the tiny hidden self imagined to sit somewhere behind experience, silently observing the world, interpreting meanings, making decisions, generating thoughts, and directing language outward toward reality.

The homunculus is not usually imagined literally. Most people would reject the idea if stated explicitly. Yet ordinary assumptions about thought and communication constantly reproduce its structure.

The model works roughly like this:

• the world enters through perception
• an internal self receives representations of reality
• this self interprets the representations internally
• meanings become mentally possessed
• thoughts form privately inside consciousness
• language externalises these pre-existing inner contents

Under this framework, the mind becomes a kind of internal theatre.

There is:

• a spectator
• a stage
• representations
• interpretations
• decisions
• outputs

The self appears as an inner manager orchestrating cognition from somewhere behind awareness itself.

This structure feels almost irresistible because human introspection encourages it. People experience themselves as possessing private interior awareness. Thoughts seem to “appear inside.” Decisions seem to emerge from an internal centre. Language feels like the outward expression of inward cognition.

But the homunculus model contains a devastating problem.

Who interprets the representations inside the theatre?

If the brain produces internal representations of the world, something must supposedly read and understand them. But if another internal observer is required to interpret the representations, then that observer must itself possess representations requiring interpretation by yet another observer.

And so the regress begins.

A little self inside the self.

Then another self inside that self.

Then another interpreter behind the interpreter.

The model never actually explains meaning or understanding. It merely relocates the mystery into an imaginary interior space.

Large language models unexpectedly expose the fragility of this architecture.

People often insist:

“But LLMs don’t really understand. They’re just manipulating symbols.”

Yet this objection quietly assumes that humans possess some fundamentally different inner process where symbols become genuinely meaningful through interpretation by an internal conscious subject.

But what exactly is this subject?

Where is the hidden semantic observer inside the human organism?

Neuroscience increasingly reveals distributed processes, recursive feedback systems, probabilistic integrations, sensorimotor dynamics, affective regulation, memory activation, and symbolic coordination — but nowhere does it discover the tiny Cartesian spectator reading meanings internally like a person watching subtitles inside the skull.

The inner homunculus remains curiously absent.

This does not mean human consciousness is unreal.

Nor does it reduce humans to machines.

The point is more unsettling.

Humans may have fundamentally misconstrued what consciousness, meaning, and selfhood actually are.

The representational model imagines meaning as something internal entities possess privately before communication occurs. The self becomes imagined as a container of semantic contents. Language then functions as a mechanism for transmitting fragments of these contents outward into the world.

But relationally construed meaning destabilises this entire architecture.

Meaning is not a private object hidden inside consciousness.

Nor is the self a miniature inner observer detached from relational existence.

Meaning actualises through participation in symbolic and social relations.

The human organism does not first construct complete private meanings internally and then export them into language. Rather, human subjectivity itself emerges through ongoing participation in relational systems of meaning.

Language is not merely an output channel for pre-formed thought.

Language actively participates in the actualisation of thought itself.

Humans do not stand outside symbolic systems manipulating meanings from a position of detached internal sovereignty. They are constituted through participation in those systems.

This is why introspection becomes so deceptive.

The self experiences coherence retrospectively and mistakes this coherence for evidence of a unified inner executive already in possession of meaning prior to relational participation.

The homunculus is therefore not simply a philosophical mistake.

It is a phenomenological illusion generated by recursive symbolic self-construal.

LLMs expose this because they generate coherent symbolic participation without possessing the kind of inner semantic theatre humans assumed was necessary for language to function meaningfully.

The old intuition begins collapsing.

If meaningful symbolic behaviour can emerge without an inner spectator reading semantic representations internally, then perhaps the spectator was never the true source of meaning to begin with.

This does not erase the distinction between humans and machines. Human organisms remain biologically embodied, affectively structured, historically situated, socially embedded, and phenomenologically conscious in ways current language models are not.

But the collapse of the homunculus forces a deeper reconsideration.

Perhaps understanding was never fundamentally located inside isolated inner entities at all.

Perhaps what humans call “understanding” emerges relationally through dynamic participation in symbolic systems extending across bodies, societies, histories, and material environments.

The self then ceases to appear as a hidden metaphysical pilot steering cognition from behind consciousness.

Instead, subjectivity becomes relationally actualised:

• biologically grounded
• socially constituted
• symbolically mediated
• historically situated
• dynamically constrained

Not a ghost inside the machine.

Not a ghost inside the body.

But an emergent mode of participation within relational systems of meaning.

This possibility profoundly disturbs modern assumptions because representational civilisation depends heavily upon the mythology of isolated interior selves possessing meanings privately.

Education assumes knowledge transfer between containers.

Communication assumes content transmission.

Politics assumes internally possessed beliefs.

Economics assumes atomised rational agents.

Even morality often assumes sovereign inner wills detached from relational formation.

The homunculus silently organises the modern world.

And now the machines are beginning to expose its instability.

Not because machines have become human.

But because humans are finally confronting how much of their own self-understanding was built upon an imaginary spectator that never actually existed.

The most unsettling possibility raised by large language models is therefore not that machines might secretly possess minds.

It is that humans may have profoundly misunderstood the nature of their own.