Memory helps language agents improve future decisions under cost limits
The Value of Memory in Language Agents
Artificial IntelligenceComputation and Language
Summary
Language agents use memory to remember important parts of past experiences to make better decisions later. This paper shows a way to measure how useful memory is by looking at how it reduces mistakes and helps recover important information when needed. The authors study different challenges, like uncertain future requests and how feedback changes what’s important to remember. They use examples and simulations to show when memory works well and when it can cause errors in decisions or evaluations. Their framework helps designers balance remembering useful details and the costs of storing and using that memory.
What this means in practice
- •For ai system builders: Design memory modules to store and retrieve information that most reduces errors within resource limits for language agents handling evolving tasks.
- •For automated customer support teams: Improve chatbot decision quality by managing memory content to balance the cost of storing details with benefits in follow-up interactions under uncertain requests.
A theory result. No direct application yet.
Authors
Zehao Lin, Erenkai Zhu, Helan Hu, Xixuan Hao, Chunyu Li, Kai Chen
Abstract
Memory enables language agents to carry useful distinctions from past experience into future decisions. Its value depends on which distinctions survive, whether a bounded agent can use them, and what can be recovered or revised as demands evolve. We develop a decision-theoretic framework that follows this value from raw history through persistent state, operational use, and costly recovery and revision. Original and retained value are expressed in units of avoided task loss; risk-cost frontiers describe attainable recovery and preservation across uses. Unknown demand yields value profiles, while interactive feedback changes the distribution of future work. A cost-aware simulation condition and five minimal counterexamples expose the roles of information timing, retrieval access, component interaction, feedback and demand uncertainty. In particular, a pointer with zero mutual information about the answer can still reduce the cost of exact recovery. Diagnostics on LoCoMo, a two-project update task and longitudinal cases exhibit failures of preservation, use and measurement. At fixed retrieved evidence, a saved ledger changes subsequent answers; at fixed agent outputs, an evaluator revision changes scores. The framework connects these distinctions to operational risk bounds and testable principles for allocating memory, preserving recovery options and revising evidence dependencies.