Leveraging Dual Process Theory in Language Agent Framework for Real-time Simultaneous Human-AI Collaboration

• DOI: 10.48550/arxiv.2502.11882
• Publication Date: 2025-02-17
• Authors: Zhang Shao; Xihuai Wang; Wenhao Zhang; Chaoran Li; Jing Song; Tingyu Li; Lin Qiu; Xuezhi Cao; Xunliang Cai; Wen Bin Yao; Weinan Zhang; Xinbing Wang; Ying Wen
• Keywords: Dual language; Dual purpose

Agents built on large language models (LLMs) have excelled in turn-by-turn human-AI collaboration but struggle with simultaneous tasks requiring real-time interaction. Latency issues and the challenge of inferring variable human strategies hinder their ability to make autonomous decisions without explicit instructions. Through experiments current independent System 1 2 methods, we validate necessity using Dual Process Theory (DPT) tasks. We propose DPT-Agent, a novel agent framework that integrates for efficient collaboration. DPT-Agent's uses Finite-state Machine (FSM) code-as-policy fast, intuitive, controllable decision-making. Mind (ToM) asynchronous reflection infer intentions perform reasoning-based decisions. demonstrate effectiveness DPT-Agent through further rule-based agents collaborators, showing significant improvements over mainstream LLM-based frameworks. To best our knowledge, is first achieves successful autonomously. Code can be found https://github.com/sjtu-marl/DPT-Agent.