Dissertation Defense: LLM-based Long-Term Life Task Planning to Reduce Human Uncertainty
Ben Wang, PhD Candidate, SLIS
Committee: Drs. Jiqun Liu (Dissertation Committee Chair), June Abbas, Yong Ju Jung, Andrew H. Fagg (Computer Science)
When: Thursday December 4th 10am-12pm CST
Location: SLIS Conference Room, Bizzell Library Room 120
401 W Brooks St. Norman OK 73019 or join via Zoom
Abstract: In long-term life tasks, people often face challenges from uncertainty in tasks and information-seeking, which can create difficulties in decision-making and task completion. Recent advancements in Artificial Intelligence (AI), especially in Large Language Models (LLMs), offer transformative capabilities in domain-specific task planning and problem-solving. Despite these innovations, there is limited understanding of how such technologies can be applied to assist humans in long-term life tasks. This dissertation work seeks to address this gap by exploring how human-AI collaboration, mediated through LLM-based agents, can improve long-term life task planning and uncertainty management. To achieve this, this dissertation first proposes the long-term life task type and investigates how people may use AI tools to assist them in planning long-term life tasks and cope with uncertainty. Secondly, it proposes the Goal Oriented Long-term liFe planning (GOLF) framework that integrates LLMs to emulate human-AI collaboration in diverse long-term life task domains. The framework facilitates task decomposition and iterative planning while evaluating uncertainties throughout the process. Thirdly, the study introduces a GOLF benchmark to evaluate and generalize the effectiveness of LLMs in long-term planning scenarios. This study operates at the intersection of cognitive science, information science, and human-AI interaction, and makes multifaceted contributions. Theoretically, by introducing the concept of long-term life tasks and proposing the GOLF framework, it explores the potential of human-AI collaborations for long-term planning and uncertainty management. Methodologically, through a simulation-based approach and the LLM benchmark, it enables systematic evaluation of LLMs in long-term planning scenarios across domains. This work bridges the gap between human cognitive strategies and AI planning capabilities, providing a robust foundation for future advancements in AI-assisted task management and goal achievement.
