When AI Becomes Hard to Understand: Cognitive Demands in Real-World Human-AI Conversations
研究分析了84,000多条ChatGPT和Gemini对话，探讨AI响应在实际对话中何时变得难以理解，并提出了一种对话复杂度预算的概念来优化用户、任务和交互的响应配置。
0 citationsRead paper
Selected work
Representative Papers
From Information to Delegation: Mapping Human-AI Financial Decision Making
This study addresses the need to quantify human information demands and delegation behaviors in financial decision-making involving artificial intelligence, thereby elucidating the mechanisms underlying human–AI allocation of decision authority. To this end, we propose a novel measurement framework that integrates intent recognition with actual delegation behavior, shifting the analytical focus from conversational content to observable authorization actions. Leveraging 1.5 million real-world interaction logs between users and high-autonomy AI systems—specifically ChatGPT and Gemini—we establish an empirical behavioral benchmark in contexts where AI exhibits substantial autonomy. Our findings reveal that users predominantly rely on AI for information acquisition and judgment support, yet rarely delegate actual execution of financial transactions, offering both empirical grounding and methodological innovation for understanding collaborative human–AI decision-making.
0 citationsRead paper
Recent publications
Latest Papers
When AI Becomes Hard to Understand: Cognitive Demands in Real-World Human-AI Conversations
研究分析了84,000多条ChatGPT和Gemini对话，探讨AI响应在实际对话中何时变得难以理解，并提出了一种对话复杂度预算的概念来优化用户、任务和交互的响应配置。
0 citationsRead paper
From Information to Delegation: Mapping Human-AI Financial Decision Making
This study addresses the need to quantify human information demands and delegation behaviors in financial decision-making involving artificial intelligence, thereby elucidating the mechanisms underlying human–AI allocation of decision authority. To this end, we propose a novel measurement framework that integrates intent recognition with actual delegation behavior, shifting the analytical focus from conversational content to observable authorization actions. Leveraging 1.5 million real-world interaction logs between users and high-autonomy AI systems—specifically ChatGPT and Gemini—we establish an empirical behavioral benchmark in contexts where AI exhibits substantial autonomy. Our findings reveal that users predominantly rely on AI for information acquisition and judgment support, yet rarely delegate actual execution of financial transactions, offering both empirical grounding and methodological innovation for understanding collaborative human–AI decision-making.
0 citationsRead paper