Distributed agents improve skills without sharing private data
Federated Agent Optimization
Artificial Intelligence
Summary
Many AI assistants learn by working on tasks and using tools, but their experiences are kept private within different organizations. The authors talk about a way for these AI agents to learn from each other without sharing sensitive details like raw data or full task histories. They introduce a concept called Federated Agent Optimization, which balances improving AI skills, protecting privacy, and limiting communication. This approach lets agents share only safe, useful knowledge to get better together while respecting confidentiality.
What this means in practice
- •For enterprise ai developers: Enable AI assistants in different companies to improve their decision-making without exchanging private data or full workflows.
- •For chatbot platform engineers: Develop chatbots that collectively enhance skills by securely sharing abstracted knowledge instead of user conversations or logs.
A position paper. It proposes an approach and reports no results.
Authors
Qiang Yang, Zhiqiang Kou, Xueyi Zhang, Dong-Dong Wu, Hanlin Gu, Jing Guo, Yang Liu, Di Jiang, Qian Xu
Abstract
Large language model (LLM) agents increasingly operate in private environments and accumulate valuable experience from task execution, tool use, feedback, and local knowledge. Yet such experience is distributed across organizations and cannot be directly shared because of privacy and proprietary constraints. Conventional federated learning is insufficient for this setting, as agent capabilities extend beyond model parameters to memory, tools, rewards, skills, and structured knowledge. In this paper, we formulate \textbf{Federated Agent Optimization (FAO)}, which studies how distributed agents can collaboratively improve through controlled information exchange while keeping raw data, complete trajectories, and private knowledge local. We define FAO as a multi-objective problem balancing agent utility, privacy leakage, and communication cost, and organize its optimization space across policy, memory, tool use, reward, and structured knowledge and skills. We further characterize how private experience can be abstracted, protected, aggregated, and adapted into transferable capabilities, providing a unified view of how agents can benefit from one another without direct experience sharing. Finally, we identify the key challenges of FAO and outline several promising directions for future research toward trustworthy federated agent systems.