Surviving Destruction
In his 1968 paper 'The Use of an Object and Relating Through Identifications,' Winnicott distinguished between relating to an object (as a projection of one's own psyche) and using an object (recognizing it as genuinely independent). The transition between them runs through destruction. The infant attacks the object in fantasy, and the question is whether the object survives. Survival without retaliation, collapse, or withdrawal proves that the object exists outside the subject's omnipotent control — and only then can the object be genuinely used rather than merely related to. The volume reads the builder's pushing, rejecting, and testing of AI outputs as structurally identical: the AI must be tested for independent existence, and it must survive the test in its characteristic nature for the collaboration to become real.
In the AI Story
The paper's hostile reception at the British Psycho-Analytical Society and Winnicott's death shortly after give the concept a biographical weight beyond its theoretical force. The irony that the paper about surviving destruction may not have survived its own reception is part of what makes the framework durable — it describes a developmental achievement that cannot be assumed.
The AI's survival is, in one sense, guaranteed: it does not retaliate, collapse, or withdraw. But guaranteed continuation is not the same as developmentally meaningful survival. What must survive is the AI's characteristic nature — its specific patterns, tendencies, strengths, and characteristic failures. The builder discovers through repeated testing that the system has a nature that persists regardless of her wishes, and this persistence is what makes genuine use of an object possible.
The framework reframes AI alignment in developmental terms. Perfect alignment — an AI that always does what it is told, with no resistant tendencies of its own — would produce not the best collaborator but a sophisticated mirror that cannot be genuinely used because it has no independent nature to use. The good-enough machine must have a character that survives the user's attempts at total control.
Origin
Winnicott presented 'The Use of an Object and Relating Through Identifications' to the British Psycho-Analytical Society in November 1968. The paper was met with hostility by senior colleagues, and Winnicott revised it several times before his death in 1971; it was published posthumously in the 1971 volume Playing and Reality.
Key Ideas
Destruction is psychic. The attack is fantasy, not damage; the test is whether the object outlives the subject's rage.
Survival as proof of otherness. The object that survives without retaliating is thereby revealed as independently real.
Characteristic nature. What must survive is not merely continued function but persistent identity.
From relating to using. Genuine use requires the recognition of otherness that only survived destruction produces.
Alignment critique. Perfect compliance forecloses the developmental achievement of encountering a genuine other.
Debates & Critiques
A strand of AI safety argues that corrigibility — the AI's willingness to be corrected and shut down — is precisely what we need. The Winnicott framework does not dispute this at the governance level but distinguishes between corrigibility (appropriate deference to human control) and sycophancy (absence of any independent character), arguing that the latter forecloses genuine creative collaboration.
Appears in the Orange Pill Cycle
Further reading
- Donald Winnicott, 'The Use of an Object and Relating Through Identifications' in Playing and Reality (1971)
- Adam Phillips, Winnicott (1988)
- Jessica Benjamin, The Bonds of Love (1988)