The Embodied Hijack: when Pleistocene minds meet disembodied artificial intelligence.

Abstract

The rapid integration of artificial intelligence into everyday life has intensified a long-standing feature of human cognition: the attribution of agency, intention, and understanding to nonhuman systems. People describe language models, virtual assistants, and autonomous technologies as if these systems know, decide, want, or understand, and they continue to do so even when they know the systems have no inner life. The standard account dismisses this as naïve anthropomorphism, the misfiring of evolved agency-detection systems calibrated in the Environment of Evolutionary Adaptedness. We argue that the standard account is incomplete. It explains the immediacy of anthropomorphic response but not its persistence even when users know the system has no mind. Drawing on evolutionary psychology, philosophy of agency, and the active inference framework, we advance the Embodied Hijack hypothesis. Across evolutionary time, fluent communication and contingent responsiveness were produced only by embodied, self-maintaining agents with vulnerability and temporal continuity. Current conversational LLM deployments are the first class of entity to reproduce these signals without the grounding properties - biological self-maintenance, vulnerability, and temporal continuity - that historically produced them. The result is a predictable misalignment: users' inferential systems treat these signals as evidence of agency they were calibrated to indicate, producing systematic misattribution. The Embodied Hijack is not irrationality. It is the optimal predictive response of a Pleistocene-calibrated brain to the rupture of the evolutionary invariant that once tied fluent communication to embodied self-maintenance. The framework yields a unique empirical signature: anthropomorphic response will track the signal profile of a system independently of users' propositional beliefs about what the system is. We close by arguing that the goal is epistemic alignment - bringing how users interpret these systems into correspondence with what these systems actually are - and that this alignment is achieved through interface design rather than user education.