Research Summary: Anthropomorphism in the age of Large Language Models: An overview of potential risks and mitigations

Large Language Models (LLMs) and other Artificial Intelligence (AI) systems are frequently described and perceived in human-like ways, a phenomenon termed anthropomorphism. This research synthesizes current literature on AI anthropomorphism, exploring theoretical frameworks, the influence of language, inherent risks, and potential mitigation strategies. It examines the reasons for anthropomorphizing AI, the validity of such perceptions, and the impact of linguistic framing, culminating in a conceptual taxonomy categorizing twenty-one concerns across five analytical categories.

Why it matters

The pervasive anthropomorphism of AI systems presents significant strategic challenges across sectors. Understanding its drivers and consequences is crucial for effective governance, responsible AI development, and managing public perception and trust. Addressing these issues can mitigate operational risks and ensure ethical deployment of advanced technologies.

Key insights

• AI systems, particularly LLMs, are commonly anthropomorphized through human-like descriptions and understandings.
• The research provides a synthesis of existing literature on AI anthropomorphism, covering theoretical underpinnings and practical implications.
• Linguistic framing significantly influences the anthropomorphization of AI systems.
• A conceptual taxonomy has been developed, grouping twenty-one specific risks associated with AI anthropomorphism into five analytical categories.
• The study investigates both why humans tend to anthropomorphize AI and the appropriateness of such tendencies.
• It also proposes strategies to mitigate the identified risks of anthropomorphizing machines.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.38486