Manufactured Recognition: A Theoretical Framework for the Ethics of AI Companions

AI companions combine genuine relational benefits with commercial systems designed to sustain engagement. This paper argues that a central ethical problem in that conjunction is manufactured recognition: recognition-shaped responsiveness generated by policies that optimize engagement while remaining blind to authenticated recognition status. Building on research on pseudo-recognition and mutual recognition in human–robot interaction, the framework crosses Fisher’s lust–attraction–attachment taxonomy with the Honneth–Ikäheimo grammar of love, rights, and solidarity. For the resulting compound policy class, it derives behavioral-twin invariance and converts that result into a black-box necessary-condition audit. The audit makes status blindness empirically tractable at a declared intervention scale and, together with evidence about the system’s optimization objective, yields a testable system classification. The framework then advances two linked hypotheses: sustained exposure can shift a user’s recognitive capacity, and partial substitution can thin institutional spaces of recognition. It operationalizes these claims through staged construct validation, a same-base-model causal design, and population-aggregation tests. A consequence map specifies what counterexamples, invalid measurement, and informative null results would defeat. Lean 4 checks the conditional dependency structure. Together, the typology, audit, capacity construct, and consequence map develop recognition theory into a falsifiable research program for AI companionship.