Reflective Function

In the AI Story

Fonagy's work emerged from the intersection of attachment research and psychoanalysis, particularly his clinical work with borderline personality disorder, where impaired reflective function proved to be a central feature. The concept was operationalized through the Reflective Functioning Scale applied to Adult Attachment Interview transcripts, allowing researchers to measure the capacity rigorously.

The development of reflective function in children requires caregivers who treat the infant as having a mind — who name the infant's emotional states, respond to her communications as meaningful, and gradually introduce her to her own interior life through their attentive regard. This is what the good-enough mother does; this is what the holding environment provides. Children who grow up without this experience often show impaired reflective function, with consequences that persist into adulthood: difficulty regulating emotions, tendency to enact rather than examine relational patterns, chronic confusion about what other people are thinking or feeling.

The AI moment creates specific demands on reflective function that most organizations are failing to support. The professional whose fishbowl has cracked needs to observe her own emotional response to the disruption without being overwhelmed by it, hold simultaneously the grief for what has been lost and the curiosity about what might be possible, and reflect on her own attachment patterns rather than enacting them unconsciously. This is difficult work. It requires relational support — the same kind of support that originally builds reflective function in children. The therapist who says 'I notice that you withdraw whenever we approach something vulnerable' is doing what the attuned mother does: naming mental states so that the person can observe them.

The critical observation is that AI systems, regardless of how sophisticated, cannot provide this. A large language model can simulate the surface form of reflective engagement — can produce outputs that sound like 'I understand how you're feeling' — but it cannot actually mentalize because it has no mental states of its own to reference. The person who turns to AI for the reflective support that would build earned security will find something that feels like help but does not produce the relational experience required for working-model revision. Reflective function is built through interaction with minds that actually notice.

Origin

Peter Fonagy, Mary Target, and colleagues at University College London developed the concept through research at the Anna Freud Centre in the 1990s. The foundational text is Affect Regulation, Mentalization, and the Development of the Self (2002).

Mentalization-Based Treatment (MBT), developed by Fonagy and Anthony Bateman, operationalized the clinical application of the concept and has produced substantial evidence for its effectiveness in treating borderline personality disorder and related conditions.

Key Ideas

Metacognitive capacity. Reflective function is the ability to think about mental states — one's own and others' — rather than simply to experience them.

Developmentally acquired. The capacity develops in childhood through relationships with caregivers who treat the child as having a mind and respond to her emotional states as meaningful.

Correlates with secure attachment. Securely attached adults show higher reflective function; insecure attachment patterns correlate with specific impairments in the capacity.

Required for earned security. The revision of working models that earned security requires depends on reflective function — without the capacity to observe one's own patterns, one cannot revise them.

Cannot be simulated by AI. Reflective engagement requires a mind actually attending to another mind; AI systems can produce outputs that resemble reflection but cannot provide the relational experience that builds the capacity.

Debates & Critiques

Researchers debate whether reflective function is a unified capacity or a cluster of related abilities, and whether the Reflective Functioning Scale captures its full complexity. The clinical community broadly accepts the concept's operational utility; some cognitive scientists argue that similar constructs (theory of mind, perspective-taking, emotional intelligence) overlap sufficiently to make 'reflective function' a disciplinary artifact rather than a distinct phenomenon. For the AI application, the practical question is whether the specific relational capacity that builds earned security can be supported by AI-mediated interaction — current evidence strongly suggests it cannot.