A new study published on arXiv explores how people's social norms differ when interacting with large language models (LLMs) compared to humans. Researchers conducted experiments using the Dictator Game and Ultimatum Game, finding that self-serving allocations by LLMs are judged as less appropriate than when humans make them. Additionally, rejecting offers from an LLM in the Ultimatum Game is considered more appropriate than rejecting identical offers from a human, suggesting people attribute bounded agency to LLMs in resource allocation scenarios. AI
IMPACT Suggests that human perceptions of fairness and agency differ between humans and LLMs, impacting AI system design and user trust.
RANK_REASON Research paper published on arXiv detailing experimental findings on human norms applied to LLMs. [lever_c_demoted from research: ic=1 ai=1.0]
- Dictator game
- Krupka-Weber norm elicitation task
- large language models
- Paweł Niszczota
- UK
- ultimatum game
- US
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
