Position paper argues anthropomorphism hinders LLM research
A new arXiv position paper contends that attributing human-like traits to AI systems is an automatic habit that persists even among technical experts, and that it skews how researchers frame and evaluate language models. The authors review a large body of published work to show how anthropomorphic language shapes experimental design, interpretation of results, and safety claims. They call for alternative conceptual frameworks that describe model behavior without implying human-like minds.