SMART: Self-Aware Agent for Tool Overuse Mitigation

2025-02-17

AI Summary

LLM agents frequently over-rely on external tools due to insufficient self-monitoring capabilities, leading to excessive computational overhead and diminished efficiency. To address this, we propose SMART (Self-Monitoring and Adaptive Reasoning via Meta-cognition), a meta-cognitively inspired paradigm that dynamically balances parametric knowledge utilization and tool invocation for efficient decision-making. Our contributions are threefold: (1) the first SMART framework, integrating multi-step reasoning modeling, interpretable tool-selection mechanisms, and rationale-augmented data construction; (2) SMART-ER, the first alternated tool-use dataset with fine-grained reasoning-justification annotations; and (3) state-of-the-art performance—reducing tool calls by 24% and improving task accuracy by over 37% on GSM8K and MINTQA—enabling a 7B model to match or exceed the performance of 70B models and GPT-4o, while demonstrating significantly enhanced cross-domain generalization.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Metareasoning and Metaheuristics
• Multiagent Systems: Mechanism Design

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Current Large Language Model (LLM) agents demonstrate strong reasoning and tool use capabilities, but often lack self-awareness, failing to balance these approaches effectively. This imbalance leads to Tool Overuse, where models unnecessarily rely on external tools for tasks solvable with parametric knowledge, increasing computational overhead. Inspired by human metacognition, we introduce SMART (Strategic Model-Aware Reasoning with Tools), a paradigm that enhances an agent's self-awareness to optimize task handling and reduce tool overuse. To support this paradigm, we introduce SMART-ER, a dataset spanning three domains, where reasoning alternates between parametric knowledge and tool-dependent steps, with each step enriched by rationales explaining when tools are necessary. Through supervised training, we develop SMARTAgent, a family of models that dynamically balance parametric knowledge and tool use. Evaluations show that SMARTAgent reduces tool use by 24% while improving performance by over 37%, enabling 7B-scale models to match its 70B counterpart and GPT-4o. Additionally, SMARTAgent generalizes to out-of-distribution test data like GSM8K and MINTQA, maintaining accuracy with just one-fifth the tool calls. These highlight the potential of strategic tool use to enhance reasoning, mitigate overuse, and bridge the gap between model size and performance, advancing intelligent and resource-efficient agent designs.

Problem

• Enhance LLM self-awareness to optimize task handling.
• Reduce tool overuse in Large Language Model agents.
• Balance parametric knowledge and tool use dynamically.

Innovation

• Enhances self-awareness in LLM agents
• Introduces SMART-ER dataset for training
• Develops SMARTAgent to balance tool use