Computational analysis of 100 K choice dilemmas: Decision attributes, trade-off structures, and model-based prediction.
10萬個選擇困境的計算分析:決策屬性、權衡結構與基於模型的預測
Proc Natl Acad Sci U S A 2025-04-21
Playing repeated games with large language models.
與大型語言模型反覆進行博弈遊戲
Nat Hum Behav 2025-05-09
研究發現,像 GPT-4 這類大型語言模型在自利型博弈(如囚徒困境)表現不錯,但在需要協調的博弈(如Battle of the Sexes)就比較弱。若給它們更多對手資訊,並引導用「社會性思考」推理,能提升與人類的協調能力。這有助了解 AI 的社會行為,也推動 AI 行為博弈論的發展。
PubMedDOI
Large-scale moral machine experiment on large language models.
大型語言模型的大規模道德機器實驗
PLoS One 2025-05-21
Take caution in using LLMs as human surrogates.
在將 LLMs 作為人類代理時需謹慎。
Proc Natl Acad Sci U S A 2025-06-13
雖然大型語言模型有時能模仿人類回應,但在像 11-20 money request game 這類需要複雜推理的任務上,表現常常和人類差很多,還會受到提示語和安全機制影響,出錯方式也很難預測。所以,社會科學研究如果想用 LLMs 取代人類,真的要特別小心。
PubMedDOI
Do Language Model Agents Align with Humans in Rating Visualizations? An Empirical Study.
語言模型代理在評分視覺化圖表時是否與人類一致?一項實證研究
IEEE Comput Graph Appl 2025-07-09
A large-scale replication of scenario-based experiments in psychology and management using large language models.
使用大型語言模型進行心理學與管理學情境式實驗的大規模重複研究
Nat Comput Sci 2025-07-09