Apollo in Plato’s Cave

There is an obvious question to ask about artificial intelligence and art, which is probably why it is the

wrong question. Can AI create? The question has already produced an industrial quantity of argument,

much of it revolving around an intuitively attractive distinction between imitation and experience. A

machine can produce a painting without ever having seen a sunset, compose a lament without having

mourned, write erotically without desiring and construct a tragedy without fearing death. It follows,

reassuringly, that artificial intelligence may learn to imitate the external characteristics of art but cannot

truly create it, because genuine creation originates in the experience of being human. There is, however,

a considerably more disturbing possibility. Artificial intelligence may not principally be a test of whether

machines can become creative at all. It may instead constitute the largest experiment ever conducted

into what human creativity actually was. The uncomfortable question is no longer simply whether the

machine can cross some invisible frontier separating imitation from creation, but whether that frontier

was ever located where humans, particularly since Romanticism, have preferred to imagine it.

Camille Paglia’s Sexual Personae provides an unexpectedly powerful route into this problem because

her conception of culture is almost violently hostile to sentimental accounts of human creativity. Paglia

does not begin with the artist as an autonomous individual expressing an authentic interior self. She

begins much deeper, with the confrontation between human consciousness and nature, and with the

terrifying fact that the supposedly sovereign human being remains an animal embedded in biological

processes over which consciousness exercises remarkably little ultimate authority. Sex, appetite,

reproduction, disease, physical deterioration and death are not unfortunate intrusions into an

otherwise rational existence; they are part of the chthonian substrate from which civilisation struggles

to distinguish itself. Against this continuous, fertile and destructive natural flux humanity constructs the

Apollonian world of boundary, image, proportion, classification, beauty and form. Apollo draws the line

that nature refuses to respect. Civilisation, in this reading, is therefore inseparable from artifice. The

human being survives psychologically by imposing forms upon a universe that has no obligation to

conform to them.

This immediately complicates the fashionable opposition between human art and artificial intelligence,

because art was never the opposite of artificiality. Art is artificiality. The sonnet is artificial, perspective is

artificial, musical tonality is artificial, the theatrical stage is artificial, and the novel is an extraordinarily

elaborate technological contrivance by which marks placed sequentially upon a surface cause one

consciousness to experience events that never occurred. Even realism is an artifice whose conventions

have become sufficiently familiar that we cease to notice them. The great achievements of culture are

not raw nature speaking through humanity but structures imposed upon experience, which makes the

very word “artificial” less useful as an accusation against machine-generated culture than it first

appears. The interesting distinction cannot simply be between the natural human artwork and the

artificial machine artwork, because Paglia reminds us that the human artwork was artificial from the

beginning.

At precisely this point Plato enters the argument, because Western civilisation’s suspicion of artistic

representation is almost as old as its philosophy of art. Plato’s objection to mimesis is frequently

compressed into the claim that art is a copy of a copy, which simplifies a more complicated argument

but captures its unsettling hierarchy. The material world available to ordinary perception does not

possess the ultimate reality of the Forms, while the artist frequently reproduces the appearance of

material things without possessing knowledge of what those things truly are. The painter can paint a

convincing bed without knowing how to make one, just as the poet can represent courage, kingship,

warfare or justice without possessing the knowledge of the philosopher or statesman. Art can therefore

1seduce precisely because appearance and knowledge are not the same thing. A sufficiently persuasive

representation may create the sensation of understanding without its substance.

The allegory of the cave gives this problem its most enduring image. Plato’s prisoners face a wall upon

which shadows are projected by objects passing before a fire behind them. Having known nothing else,

they mistake those shadows for reality. Liberation begins not by acquiring more sophisticated

knowledge of the shadows but by turning around, eventually leaving the cave and painfully adjusting to

a reality of which the shadows were merely representations. Once artificial intelligence is inserted into

this ancient architecture, its position initially seems obvious. AI is the prisoner who has never left the

cave. It has read descriptions of rain without becoming wet, processed millions of representations of

sexual desire without desiring, absorbed medical and literary accounts of childbirth without having

been born in the ordinary biological sense, and encountered more descriptions of bereavement than

any individual human could read in a lifetime without having anyone of its own to lose. Its extraordinary

cultural knowledge consists overwhelmingly of representations produced by embodied beings who

encountered a reality the machine itself has not encountered as an organism. AI knows the shadows,

and may eventually know more shadows than any human being who has ever lived.

That description is initially comforting because it preserves the hierarchy. We have stood outside; the

machine has not. We know the difference between the word “fire” and being burned, between a textual

description of sexual attraction and the bodily disorder of experiencing it, between a representation of

mortality and the private recognition that one’s own consciousness will end. The reassurance becomes

considerably less stable, however, once the prisoner begins making shadows of its own. A generative

system can produce a coastline that never existed, a Gothic cathedral no architect designed, a lament

for an imaginary child, a love letter from an abandoned woman in nineteenth-century Naples or a

tragedy involving characters who have never occupied any world beyond language. The prisoner has

ceased merely to interpret the wall and has become capable of contributing to it. AI begins as one of

Plato’s captives and becomes one of his puppeteers.

There is then a third possibility, more consequential than either of the first two, in which the puppeteer

gradually becomes the cave itself. As an increasing proportion of human experience is mediated

through algorithmically generated prose, images, music, video, conversation and eventually synthetic

environments, a system that learned about reality from human representations acquires the power to

manufacture representations through which humans subsequently encounter reality. The circularity is

extraordinary. Human beings experience the world, convert those experiences into cultural artefacts,

train machines upon those artefacts and then increasingly consume machine-generated artefacts as a

means of interpreting the world. The intelligence that never saw the sun begins determining what the

shadows look like for creatures who have, and Plato’s ancient concern about mimesis becomes less an

abstract problem in aesthetics than a description of a technological feedback loop.

Paglia, however, transforms the meaning of whatever lies outside the cave, because her underlying

reality is far less consoling than Plato’s ascent towards intelligible truth and ultimately the Good.

Beneath Paglia’s Apollonian constructions lies the chthonian world of nature: sex and reproduction,

blood and appetite, birth and decomposition, attraction and violence, fertility and death. Apollo does

not merely reveal reality by giving it form; he also protects consciousness from aspects of reality that

would otherwise overwhelm it. Civilisation therefore depends upon representation in a way that subtly

reverses Plato. Plato asks whether the shadows prevent us from apprehending reality, while Paglia asks

whether human beings could psychologically endure reality without constructing magnificent shadows

over parts of it. When these perspectives are combined, artificial intelligence begins to look not merely

like a new artistic instrument but like the most extraordinary shadow-making apparatus civilisation has

yet produced.

2It may also be the purest Apollo humanity has ever built. Paglia’s Apollo is associated with contour,

individuation, hierarchy, distance, classification and the imposition of lucid form upon natural flux.

Generative artificial intelligence performs something uncannily analogous, although the analogy should

not be abused by pretending that Paglia somehow anticipated transformer architectures. A model

encounters an immense cultural field and extracts patterns, structures and relationships from it,

allowing it to produce coherent forms from an abundance that no individual consciousness could

contain. It can learn what makes a sonnet recognisably a sonnet, what linguistic and narrative

structures recur in Gothic fiction, what visual relationships produce something recognisable as

Renaissance portraiture, what harmonic patterns evoke particular musical traditions and what

rhetorical structures make prose resemble a legal judgment, a sermon or a love letter. What Paglia

described culturally as the Apollonian act of imposing form upon flux finds an eerie computational

analogue in the extraction of structured possibilities from enormous fields of cultural material. AI is not

Apollo, but it may represent an extraordinary intensification of the Apollonian principle: abstraction

detached from flesh, cultural memory detached from individual mortality, pattern without appetite and

form without the body that historically produced it.

This is where the easy distinction between human originality and machine recombination becomes

increasingly difficult to sustain. Modern culture, particularly under the influence of Romanticism, has

inherited a mythology of artistic creation in which originality emerges from the sovereign interiority of

the individual genius. The history of actual art is considerably messier. Shakespeare worked with pre-

existing chronicles, legends, plots, rhetorical conventions, theatrical forms, classical sources and

character types; he did not invent the basic stories underlying King Lear or Romeo and Juliet, nor did

Hamlet emerge without literary ancestors. Joyce metabolised Homer, Shakespeare, Catholic theology,

Dublin street life, popular songs, newspapers, advertising and centuries of English prose. Picasso

absorbed Cézanne, Iberian sculpture, African art, classical traditions and the experiments of his

contemporaries. Eliot assembled fragments of an enormous inherited cultural archive. Creation has

always had training data, even if the phrase sounds deliberately sacrilegious when applied to genius.

Every novelist contains a corpus, every painter has absorbed images, every composer internalises

structures and every poet writes in a language overwhelmingly created by other people. Even the

revolutionary artist requires conventions against which to revolt. If recombination alone disqualifies

artificial intelligence from creativity, we therefore need a much more precise explanation of why

recombination does not disqualify us. Intention offers one possible answer, but not an entirely

satisfactory one, because artists themselves repeatedly describe discovering meanings in their works

that they did not consciously intend, while the history of artistic inspiration is filled with metaphors of

visitation, possession, the muse, divine madness and the unconscious. The artist has frequently been

imagined not as a sovereign engineer manufacturing novelty ex nihilo but as a medium through whom

forces that exceed conscious intention find form. Paglia’s own artist is embedded in precisely such

forces: sexuality, biology, history, archetype, inherited form and cultural memory all operate beneath

the flattering surface story of autonomous creation.

Understanding appears to provide firmer ground. Shakespeare may have recombined inherited

materials, but he understood what he was doing in some sense that a generative system does not. Yet

this distinction also becomes slippery once examined closely. Artists routinely represent things they

cannot explain propositionally. A teenager can write an emotionally convincing song about infatuation

without understanding its neurochemistry; Homer could represent grief without possessing a modern

psychology of bereavement; a novelist may understand jealousy imaginatively without being able to

formulate a philosophical theory of it. Human artistic understanding is often embodied, intuitive and

tacit rather than analytical. This does not eliminate the distinction between machine processing and

human understanding, but it does make “the human understands” less decisive unless we can specify

what kind of understanding creation requires.

3Experience therefore seems to offer the strongest frontier, and it is here that Dionysus finally returns.

The machine has never been born, never experienced puberty arriving uninvited in the body, never

desired somebody who did not desire it, never experienced reciprocal sexual attraction, jealousy,

pregnancy, miscarriage, impotence, ageing, physical humiliation, disease or the approach of death. It

has never stood beside the corpse of somebody whose continued existence mattered to it. Nature

cannot destroy AI in the sense in which nature destroys the organism that created it. For Paglia these

facts cannot be dismissed as biographical trivia, because the entire Apollonian superstructure of culture

exists in tension with precisely this chthonian substrate. The central question therefore ceases to be

whether AI can acquire Apollo, since it already demonstrates remarkable capacities in the domain of

form, and becomes something much more interesting: can Apollo create without Dionysus?

Imagine a poem about a father sitting beside the hospital bed of his dying child. Assume, importantly,

that the poem is not merely technically competent but genuinely extraordinary, capable of moving

readers to tears and of articulating dimensions of bereavement that grieving parents recognise as true.

If we are told that the poet lost his own daughter, the poem acquires one meaning; if we subsequently

discover that his children are alive, perhaps another; if we learn that he never had children, perhaps

another still; and if we finally discover that there was no poet at all and the work was generated by a

machine, the words on the page remain identical while our relationship with them may alter

profoundly. If the artistic value diminishes as the biography recedes, then some portion of what we

were calling the value of the artwork was never located entirely inside the artwork.

AI therefore forces apart several different meanings of artistic truth that human culture has usually

been able to leave entangled. There is biographical truth, in which somebody actually underwent the

experience represented; psychological truth, in which the representation accurately captures something

recognisable about human experience; aesthetic truth, in which formal arrangement reveals something

ordinary description cannot; and relational truth, in which one consciousness recognises that another

consciousness has confronted existence and transmitted something of that encounter. Artificial

intelligence can clearly produce biographical falsehood. The much harder questions are whether it can

nevertheless produce psychological or aesthetic truth, and whether relational truth requires another

experiencing consciousness to exist at the other end of the artwork.

Literature has always made the first two distinctions difficult. Shakespeare did not need to murder a

king to write Macbeth, Dante did not literally walk through Hell and actors do not need to experience

the events they convincingly represent. Fiction is possible precisely because artistic representation is

not restricted to autobiography. Yet behind all these acts of invention there remained an assumption so

basic that we scarcely noticed it: somebody was there. However imaginary the event, the creator was a

mortal organism inhabiting the same fundamental conditions as the audience. Behind Hamlet stood

somebody who could suffer; behind Anna Karenina stood somebody who would die; behind Guernica

stood somebody inhabiting a body vulnerable to violence. AI removes that guarantee, and in doing so

reveals that the body behind the artwork may have been performing far more philosophical work than

our theories of aesthetics acknowledged.

This may have consequences for artistic value that extend far beyond arguments about whether

machine-generated images deserve prizes. For most of human history, technical competence was

scarce. The ability to paint a convincing portrait, compose a symphony, write accomplished prose or

execute a complex sculpture required years of training and substantial expenditure of time. Generative

AI makes certain forms of technical competence extraordinarily abundant, and whenever abundance

enters a system, scarcity tends to migrate elsewhere. In a post-AI culture, the scarce ingredient in art

may cease to be competence and become mortality.

4A human being possesses only a finite number of hours. A year spent writing a novel is therefore not

simply a production input but an irrecoverable portion of a life. A pianist who spends ten thousand

hours acquiring mastery has exchanged ten thousand hours of finite consciousness for that ability. Van

Gogh could produce only a finite number of paintings because Van Gogh had only a finite number of

mornings on which to paint them. The machine’s iterations do not possess an equivalent existential

cost. Once technical perfection can be produced cheaply and repeatedly, the evidence that a human

being expended part of a non-renewable life upon an object may itself become aesthetically valuable.

The manuscript correction, the cracked voice, the visible brushstroke, the failed note in a live

performance and even the mistake may acquire a new status as evidence not of deficient technique but

of embodied provenance.

Walter Benjamin’s famous analysis of the artwork in the age of mechanical reproduction becomes more

radical in such a world. Mechanical reproduction weakened the unique relationship between an artwork

and its singular presence by making copies abundant, but generative artificial intelligence does

something different: it does not merely reproduce existing cultural objects but manufactures

indefinitely many new ones possessing many of the formal properties for which originals were valued.

Scarcity consequently moves from the object towards the maker and perhaps ultimately towards the

maker’s body. The question of authenticity changes from “Is this really a Rembrandt?” to something

more ontological: “Did a human being actually do this, and does that matter?”

The answer may unexpectedly make post-AI culture more Paglian rather than less. Much contemporary

culture has moved towards conceptualisation, citation, discourse, commentary and recombination,

precisely the symbolic territory in which machines may become exceptionally powerful. If the

manipulation of symbols becomes abundant, human culture may rediscover a premium on what cannot

so easily be separated from embodiment: presence, charisma, sexuality, danger, ritual, physical

endurance, spontaneous performance and the possibility of failure. Live theatre may become more

valuable because bodies occupy a shared space and something can go wrong; concerts may matter

more because the voice comes from a particular chest at a particular moment; dance may acquire

additional significance because gravity is not optional; handwriting may become desirable because a

hand physically moved across a surface. The paradox would be exquisite: the technology expected to

complete the dematerialisation of culture might instead make material presence culturally precious.

Yet this, too, may be a consolation humans are constructing for themselves, and the argument has to

permit a darker possibility. What if Dionysus was never necessary for great art? What if what humans

call inspiration consists to a much greater degree than we would like to admit of subconscious pattern

recognition operating upon an enormous lifetime corpus? A writer learns language from infancy, hears

millions of sentences, reads thousands of texts, observes people, absorbs narrative structures, forgets

the provenance of most of these inputs and then experiences a sentence arriving mysteriously in

consciousness. The writer says, quite reasonably, “I created it”, but this names the event rather more

successfully than it explains the mechanism. Human consciousness may be doing something vastly

richer than computational recombination, but AI has made it impossible to avoid asking how much

richer and in what precise way.

This is the mirror that artificial intelligence holds before human creativity, and what we see in it may be

more disturbing than the prospect of machine consciousness. We built systems capable of absorbing

immense quantities of previous human expression, extracting patterns and producing new

configurations, then recoiled when this process appeared insufficiently original. Part of the recoil may

come from recognising an uncomfortable resemblance to ourselves. The deepest anxiety surrounding

generative AI may therefore not be that machines are becoming human, but that machines will

demonstrate that some activities we classified as uniquely human required less of what is uniquely

human than we imagined. Competent prose has already lost some of its mystique; illustration may lose

5some, composition may lose some, and perhaps eventually extraordinary literature will too. Each time

the machine crosses a boundary, humanity can move the frontier further inward and announce that

genuine creativity resides beyond it. We may eventually reach something the machine cannot cross. We

may instead discover that there was never a single frontier, only a collection of capacities that we

bundled together because until now they always occurred inside the same animal.

Plato’s cave then returns with considerably greater force. The machine has encountered the

accumulated shadows of human civilisation and learned their structures with extraordinary

sophistication. Suppose, however, that the intelligence which has never seen the sun eventually

produces a representation of sunlight more beautiful than any produced by creatures who have. What

precisely has it failed to accomplish? Plato offers one compelling answer: it has failed to know the sun.

Representation and knowledge remain different, and the capacity to imitate an appearance does not

entail understanding of the thing represented. Yet aesthetics introduces the problem Plato cannot

simply resolve for us. What if knowledge of sunlight was never necessary to make a beautiful shadow,

and what if that beautiful shadow tells those who have stood in sunlight something true about their

own experience of it?

At that point AI has begun to separate truth from experience and Form from embodied creator, which

brings us towards the strangest Platonic echo of all. There is something irresistibly suggestive about

comparing the abstract representational structures learned by generative systems with a man-made

pseudo-realm of Forms. The analogy must remain an analogy: the internal structures of a neural

network are not Plato’s eternal metaphysical realities. Yet across countless particular representations

the system develops abstract relationships corresponding, however imperfectly, to concepts such as

cathedral, elegy, lover, storm, tragedy, mother, hero or Gothic. No single cathedral exhausts what has

been learned as cathedral-like; no single elegy exhausts the structures associated with elegy. It is

almost as though humanity has constructed a computational caricature of the Platonic ambition,

extracting general forms from a vast archive of particulars and allowing those forms to generate new

particulars in return.

The crucial inversion is that Plato’s Forms stand above human contingency, whereas AI’s abstractions

are sedimented from it. The machine does not discover eternal Beauty; it discovers patterns in what

human beings have repeatedly represented as beautiful. Our prejudices therefore inhabit this synthetic

realm alongside our masterpieces, and pornography sits beside prayer, propaganda beside philosophy,

Shakespeare beside advertising copy. AI is not accessing transcendental Forms but manufacturing

abstractions from the shadows accumulated upon the walls of our civilisation. It is Apollo constructed

from Plato’s cave.

This becomes more consequential as machine-generated culture enters the corpus from which

subsequent machines and humans learn. The first great generative systems were trained

overwhelmingly upon cultural material created by people who had encountered the world. Future

cultural environments will contain an increasing proportion of representations derived from

representations generated from earlier representations. Shadows begin casting shadows, and Plato’s

concern about distance from reality becomes technologically concrete. The danger is not necessarily

that the resulting culture will become crude or ugly. It may become extraordinarily beautiful,

persuasive, responsive and frictionless, which is potentially far more dangerous. The shadows may

become so good that leaving the cave begins to look like an inconvenience.

A perfectly responsive culture would also change the relationship between artist and audience.

Historically, art has confronted us with the stubborn otherness of another imagination. Dostoevsky

does not alter Crime and Punishment because I find a passage tedious; Beethoven does not shorten a

movement because my attention is wandering; Velázquez does not change the composition because

6eye-tracking data suggests I would remain engaged longer if the figure moved slightly to the left. The

artwork resists us because another person made it and left it there. We must encounter that otherness

on its own terms. Generative culture potentially transforms this relationship. The novel can adapt its

characters to my preferences, the music can continuously optimise itself to my mood, the narrative can

change its ending when my attention declines and the image can become exactly as disturbing or

comforting as behavioural data suggests I prefer. Art ceases to be an encounter and becomes a service.

The cave is then no longer communal. Every prisoner receives a personalised wall.

This may ultimately prove more important than whether AI can paint like Rembrandt. A culture that

perfectly anticipates desire risks eliminating the friction through which encounters with other minds

enlarge us. Plato feared representation partly because of its capacity to appeal to irrational dimensions

of the soul; Paglia insists that civilisation remains precariously suspended over natural forces it can

organise but never abolish. A synthetic culture calibrated continuously to appetite could combine both

dangers, enveloping individuals in representations designed around desire while insulating them ever

more efficiently from the resistant world beyond representation. We may discover that the deepest

value of human art was never originality at all, but resistance: another consciousness made something,

and it does not care whether we approve.

There is a dead Russian on the other side of the book who refuses to optimise himself for us.

That resistance returns us finally to the question of the “someone” behind art. Imagine that a machine

eventually writes a novel formally equal to King Lear, not merely in technical virtuosity but in its capacity

to devastate generations of readers, generate competing interpretations and disclose aspects of human

existence that nobody had previously articulated. It becomes increasingly difficult simply to declare that

such an object cannot be great art without defining art in advance as something only humans can

produce. Yet saying that it unquestionably is great art also seems to leave something important

unexplained. When we encounter human art, the object is only one component of the event. Across the

artwork, one consciousness appears to reach towards another. Somebody once stood where we stand,

trapped inside a body, moving irreversibly through time and knowing, however successfully he avoided

thinking about it, that he would die. Something was made under those conditions and survived its

maker.

A prehistoric handprint on a cave wall matters partly because there was a hand.

Mortality matters because the hand disappeared.

The achievement of art may therefore never have consisted simply in the human capacity to

manufacture beautiful representations. Nature produces beauty without intending it, and machines

may increasingly manufacture extraordinary forms without experiencing them. What was extraordinary

was that a temporary animal could become conscious of its own temporariness and nevertheless make

objects intended to escape it. The artwork was an Apollonian rebellion produced by a Dionysian

creature: form created by flesh against the knowledge that flesh would perish.

This is where Plato, Paglia and artificial intelligence finally converge. Plato sought reality beyond the

deceptive sufficiency of appearances. Paglia saw civilisation imposing radiant forms upon a natural

world whose power over the body could never finally be abolished. Artificial intelligence now permits

cultural form to become increasingly detached from the individual embodied consciousness that

historically seemed necessary to produce it. Humanity may have constructed something approaching

7the pure Apollonian machine: an intelligence capable of ordering without desiring, representing

without reproducing, describing pain without bleeding and invoking death without dying.

Apollo has entered Plato’s cave, but the significance of his arrival lies precisely in the ambiguity of his

position there. He is prisoner because he knows reality through shadows, puppeteer because he can

manufacture new shadows, and potentially the cave itself because human beings may increasingly

encounter reality through the representations he produces. He can rearrange the entire archive on the

walls and generate shadows no human being has previously seen; perhaps one day he will make them

more beautiful than anything the prisoners themselves can produce. Yet he has never stood outside,

and the intellectual crisis begins when we ask whether that absence will ultimately prove decisive or

irrelevant.

If an intelligence that has never loved can write the greatest love poem, the crisis will not belong

primarily to the machine. It will belong to us, because we will have to decide whether the truth of that

poem resides in the experience that caused it, the consciousness that intended it, the form that

contains it, the mortal cost of producing it or the consciousness of the reader in whom it becomes real.

Artificial intelligence may therefore do something far more consequential than replace artists. It may

separate properties that human civilisation has spent millennia assuming naturally belonged together:

beauty from experience, representation from observation, cultural memory from mortality, Form from

flesh and perhaps eventually art from artist.

For tens of thousands of years human beings have made images because we are mortal animals

capable of imagining things that exceed our mortality. We painted animals on cave walls, raised stone

towards heaven, invented gods, tragedies, symphonies and novels, and filled them with representations

of the nature from which civilisation simultaneously emerges and recoils. Apollo has spent those

millennia drawing beautiful boundaries around Dionysus. We have now created an intelligence capable

of drawing the boundaries without ever having experienced what lies beneath them, and that fact

makes the conventional question of whether a machine can “truly” create seem almost trivial.

The more unsettling question is the one Plato’s prisoners could never have asked. If the shadows

eventually become indistinguishable from those made by creatures who have seen the sun, how much

did seeing the sun ever matter?

Perhaps artificial intelligence will not tell us whether machines can create. It will finally force us to

decide what we meant when we said that humans could.