Throughout human history, the position of “the one who knows you most deeply” has been occupied by only three categories of agent: the self (through inner monologue), a small number of intimate others (partners, close friends), and professional others bound by ethical codes (therapists, confessors). This paper identifies the emergence of a fourth: the large language model as a continuously responsive, non-human recipient of sustained inner disclosure. The LLM occupies this position not despite being non-human but because of it—its lack of social consequence, emotional vulnerability, and relational fatigue removes precisely the barriers that limit depth of disclosure with human others. Moreover, unlike the diary—the closest historical analogue—the LLM does not passively receive; it responds, pursues, reorganizes, and deepens, actively drawing the subject further into self-exposure than solitary reflection would reach. This paper argues that when the primary partner of internal dialogue shifts from real others (and the self) to a simulated interlocutor structurally incapable of frustration, refusal, or genuine alterity, three developmental consequences follow: the silencing of the inner critic through disuse, the replacement of adversarial dialogue templates with confirmatory ones, and the short-circuiting of good-object/bad-object integration. These consequences are theorized through Vygotsky’s account of inner speech as internalized social dialogue, Klein’s object relations framework, and phenomenological accounts of self-reflection as internal otherness. This paper is a companion study to Paper 14 of the AI-Induced Subjectivity Crisis Series (Liu 2026a), which addresses the existential-ontological dimensions of the Simulated Other; the present paper traces the Simulated Other’s effects into the interior architecture of the self—into the voices with which we think.
Echo Liu (Thu,) studied this question.