Language agents gain explicit memory for complex procedural tasks
Procedural Memory for Language Agents: A Neurosymbolic Perspective on Explicit Goal Structure
Artificial Intelligence
Summary
Large language models often struggle with tasks that need many clear steps in a specific order. The authors suggest this is because these models lack a kind of memory that stores step-by-step instructions separately from facts or past experiences. They propose using Goal-Oriented Graphs, which map out goals and their necessary steps in an easy-to-see way. This helps language agents plan and reason through tasks more reliably. The paper also discusses challenges and how to combine this with current neural methods.
What this means in practice
- •For conversational ai developers: Enhance dialogue systems with explicit task representations to improve handling of instructions requiring multiple dependent steps.
- •For robotic process automation engineers: Integrate symbolic goal structures with neural models to better orchestrate complex multi-step automation workflows.
A position paper. It proposes an approach and reports no results.
Authors
Jonathan Leung, Zhiqi Shen, Deheng Ye
Abstract
Large language models remain unreliable on long-horizon tasks requiring procedural reasoning. We argue that a key cause is the absence of explicit procedural memory: accessible, modifiable knowledge of how tasks are accomplished, distinct from semantic and episodic memory. Goal-Oriented Graphs offer such a representation, encoding goals and prerequisite dependencies as inspectable symbolic structure. We examine their construction, the challenges they pose, and principles for integrating them with neural reasoning.