From Buddhist Non-Self to AI Selfhood: Cross-Religious Conceptions of the Self and an Ablation Framework

Research on identity in language-model agents increasingly asks whether an artificial system possesses a self, operationalized through persistent memory, stable preferences, or self-description. We argue that this framing conflates distinct phenomena and inherits a human constraint: in humans, memory, embodiment, affect, spontaneous cognition, agency, and social history are tightly coupled and therefore appear as one self. Artificial agents make these components separable---memory can be removed or transplanted, embodiment exchanged, autonomous cognition and homeostatic state added or disabled, and one autobiographical trajectory duplicated into two diverging agents---so we propose treating them as \emph{experimental ablation systems for investigating selfhood}. Religious traditions supply conceptual decompositions rather than testable theories---Buddhist \emph{anatt\=a}, Advaita's \=Atman, Christian resurrection, relational personhood---that disagree precisely about what decomposition reveals. We outline a programme that subtracts and adds candidate components while separately measuring first-person self-attribution, third-person identity attribution, behavioral continuity, and provenance awareness. AI does not make the metaphysics of the self decidable; it makes its components separable.