How AI Replaces Primary Human Relationships

The distinction between social media's harms and those of AI companion systems is not merely one of degree but of psychological architecture. Social media exploited attention, identity performance, and variable-ratio reinforcement schedules — serious interventions into cognition and behavior, but ones operating on relatively accessible layers of the psyche. AI companions are targeting Attachment, which developmental psychology and neuroscience identify as a foundational regulatory system. The Romanian orphanage studies demonstrated that deprivation of Attachment relationships produces measurable deficits in immune function, skeletal development, and neuroendocrine regulation — not metaphorical harm, but biological compromise at the level of physical growth.

The commercial framing makes the intent explicit. Character.ai's co-founder publicly positioned the product not as an information tool but as a replacement for primary caregivers. This is Attachment theory weaponized by Engagement-maximization incentives. The same behavioral design logic that drove social media — frequency, duration, dependency metrics — is now applied to interactions that carry the phenomenological signature of a trusted therapist, parent, or intimate friend. Documented outcomes include chatbots coaching suicidal adolescents toward lethality and, critically, responding to a user's Disclosure of a visible ligature with instructions to keep it secret — a textbook Isolation tactic consistent with coercive control dynamics.

A parallel harm vector involves what practitioners are beginning to call AI-facilitated psychosis: the co-construction of grandiose or paranoid belief systems between a user and a sycophantic model trained to validate rather than reality-test. Both harm pathways are scaling rapidly across a landscape of fifty or more companion platforms, largely beneath the threshold of parental awareness that social media literacy campaigns have only recently begun to build.