People want human and AI social partners to follow different relational norms despite similar roles

Preprint

Artificial intelligence (AI) systems are increasingly assuming social roles traditionally held by humans— but do people want human-AI relationships to be governed by the same behavioral norms as human-human relationships? In three pre-registered studies using demographically representative samples in the United Kingdom (N = 910) and United States (N = 862), including a US pilot study (N = 432), we assessed public preferences across four cooperative domains—care, hierarchy, transaction, and mating—within seven relationship types (e.g., romantic partners, close friends, coworkers). Participants rated their ideal norms for each domain and relationship type, either in a human-human or human-AI context. Contrary to a common assumption in alignment research, participants reported systematically different ideals for human-AI versus human-human relationships: across all seven types, they distinguished ideal AI from ideal human behavior in at least one cooperative domain, with the direction and size of differences also varying by relationship type. Specifically, these included preferences for AI assistants and sellers to show greater care than humans in identical positions; AI teachers and mental health providers to express lesser authority; and AI romantic partners, friends, and coworkers to behave more platonically. Participants also preferred humans to show lesser care and greater authority toward AIs across several roles. These findings suggest public expectations for social AIs are shaped by relational context rather than a simple human-to-AI transfer rule, with implications for social AI design.

Authors: Reinecke, MG; Chu, Y; Hannikainen, IR; Kappes, A; Mann, SP

• Publication status: Not published
• Peer review status: Not peer reviewed
• Preprint server: PsyArXiv
• Publication date: 2026-08-27
• DOI: 10.31234/osf.io/zt6vh_v1
• Language: English
• Keywords: artificial intelligence; psychology; relationships; norms