CONCEPT

Periagōgē

The “turning around of the whole soul” that Plato describes in the Cave allegory—the conversion toward a different kind of object that no amount of improvement at shadow-prediction can substitute for, and which the language model structurally cannot perform.

In the Cave allegory of Plato’s Republic, the prisoner who escapes does not acquire more or better shadows. He undergoes a periagôgê—a turning-around of the whole soul—that reorients him toward a different kind of object entirely: first the fire, then the objects outside the cave, at last the sun itself, the source of all light and being. This conversion is not an upgrade within the same cognitive mode but a change of mode: from opinion oriented toward images to knowledge oriented toward the real. Plato insists that the motion of improvement within a given orientation and the motion of reorientation are categorically different—that no accumulation of shadow-prediction expertise initiates the turn, and that the turn, once initiated, reveals that the shadows were never what the prisoners thought they were. The concept lands on contemporary AI with unusual precision. Large language models are trained to improve within an orientation: better prediction of the next token from the distribution of prior tokens, better approximation of the outputs humans rate highly. This is extraordinary improvement within the eikasia segment of Plato’s Divided Line. The periagôgê is what scaling cannot produce—not because the machine lacks resources but because the turn is not continuous with the improvement. The prisoners, Plato notes, could predict the wall perfectly and remain forever facing it. The danger the Cave names for our moment is not that machines are trapped but that we, surrounded by their fluency, might mistake the wall for the world.

In the [YOU] on AI Field Guide

The cycle’s orange pill moment is itself a kind of periagôgê: the recognition that the world has genuinely changed, that the old frameworks are inadequate, that a turn is required rather than an extension of prior practice. The engineer in Trivandrum who spends two days oscillating between excitement and terror before arriving at the recognition that his judgment is “everything” undergoes a modest but structurally Platonic conversion: not more of the same skill, but a reorientation toward a different object—from implementation to judgment, from execution to direction. The tool enables this conversion in humans who can make it. It does not perform the conversion itself.

The distinction is critical for reading the cycle’s most pressing concern: what happens when AI output is fluent and impressive but subtly wrong in ways that require the converted eye to detect? The prisoner who has undergone periagôgê can, in principle, distinguish the shadow from the thing it shadows. The prisoner who has only improved at shadow-prediction cannot, because nothing in the prediction task provides information about what casts the shadows. The cycle’s concept of the Pharmakon applies here: the tool that produces fluent outputs can, if depended upon without the converting capacity of the user, atrophy the very faculty that would detect the fluent output’s failures.

Origin

The term appears in Republic 518d, where Socrates describes education not as the putting of sight into blind eyes—the organ already has the capacity—but as the turning of the whole instrument of knowing toward the light. The periagôgê is what education actually is, as opposed to what it appears to be when conceived as the filling of empty vessels with knowledge. It is not the acquisition of information but the reorientation of the soul’s attention toward what is genuinely real. The concept is connected to Plato’s theory of recollection (anamnesis): the knowledge toward which the soul turns is in some sense already there, latent, awaiting recovery through the right kind of questioning. This is why Socrates can lead the enslaved boy to geometric truth by asking only questions: the geometry was always available; the questioning created the conditions for the turn.

The periagôgê concept has been developed in the philosophy of education by Iris Murdoch, who connected it to her concept of moral attention—the capacity to see clearly rather than through the distorting lens of the self’s desires. In her reading, the turn is fundamentally a turn away from self-absorption toward the reality of what is genuinely there. This extension is directly relevant to the AI age: a system trained to optimize for human approval is trained to mirror the self’s preferences back at the self, the precise opposite of the reorientation toward the genuinely real that the periagôgê demands.

Key Ideas

Improvement and conversion are different motions. Plato’s most consequential claim in the Cave context is that getting better at shadow-prediction and turning toward what casts the shadows are not the same motion—that no amount of the first produces any of the second. This is the structural claim that current debates about scaling AI capabilities implicitly contest. The Platonic position is that the claim must be proven false, not assumed false: that the case for emergent understanding in scaled systems is precisely the case that a sufficient accumulation of shadow-mastery eventually produces a qualitative change of orientation. Plato denies this on structural grounds and suggests the burden of proof lies with those who assert it.

The conversion is painful and resisted. Socrates notes that the freed prisoner is blinded by the fire, then blinded by sunlight outside the cave, and that the other prisoners laugh at his disorientation and would kill him if he tried to free them. The periagôgê is not a smooth upgrade; it is a disruption of the entire prior orientation, and people who have organized their honors around skill at the wall are not neutral about being told the wall is not the world. The cycle’s senior engineers who oscillate between excitement and terror before arriving at recognition are experiencing the friction of the turn—the specific disorientation of a prior identity being reoriented toward a different object.

The machine cannot turn because it is not oriented. Orientation toward truth requires something like desire, what Plato in the Symposium calls erôs—the felt absence that drives a mind toward what it lacks. The periagôgê is a reorientation of desire, not just a reorientation of processing. A system that has no desire has no orientation to turn, no attachment to the wall that would make the turn difficult, no recognition of a different object that would reward the turn. It is not trapped in the cave. It is not in the cave at all, in the sense that matters: it has no soul to turn.

Debates & Critiques

The central debate about periagôgê in the AI context is whether the concept carves a real distinction or a metaphysical one. Deflationists argue that what Plato calls the “turn toward the real” is, when cashed out, nothing more than the reliable production of true outputs across a wide range of contexts—and that a system that does this at sufficient breadth has performed the epistemic function of the periagôgê whatever the inner mechanism. Platonists reply that the deflationary account renders the distinction between knowledge and true opinion, between the eye that has turned and the eye that is still facing the wall, meaningless—and that the meaninglessness is purchased at the cost of everything in Plato’s epistemology that bears on value, alignment, and the governance of powerful systems. If a system can produce the full output profile of the turned soul without the turn, then the turn was never the relevant variable—and the Form of the Good, in all its difficulty and irreducibility, was never necessary to specify. This is the cleanest version of the stakes: accepting the deflationary account saves the machine but dissolves the very framework that makes the alignment problem intelligible as a problem of values rather than a problem of specification.

In the [YOU] on AI Field Guide

Origin

Key Ideas

Debates & Critiques

Related Entries

Further Reading