A Comprehensive Survey of Self-Evolving AI Agents [pdf]
11 days ago
- #AI Agents
- #Machine Learning
- #Self-Evolving Systems
- Recent advances in large language models have increased interest in AI agents for complex tasks.
- Most existing agent systems are static post-deployment, limiting adaptability in dynamic environments.
- Research is exploring agent evolution techniques to enhance systems via interaction data and feedback.
- Self-evolving AI agents bridge foundation models' static capabilities with lifelong adaptability.
- A unified conceptual framework for self-evolving agents includes System Inputs, Agent System, Environment, and Optimisers.
- The survey reviews self-evolving techniques targeting different agent system components.
- Domain-specific evolution strategies in biomedicine, programming, and finance are examined.
- Evaluation, safety, and ethical considerations for self-evolving agents are discussed.
- The survey aims to support the development of adaptive, autonomous, lifelong agentic systems.