Fujitsu Limited has announced the development of a self-evolving multi-AI agent technology designed to enhance corporate operations by enabling multiple AI agents to work collaboratively, learning continuously from execution results, human feedback, and policy changes.
This technology addresses the challenges faced in business operations that involve large volumes of documents and design specifications, traditionally reliant on the experience and tacit knowledge of skilled professionals. The new AI system autonomously verifies its operational experience, allowing it to adapt to changes in business environments without continuous expert intervention.
The technology's significant feature is its ability to identify reasons for success and failure, extracting actionable insights rather than merely storing improvement proposals. This advancement allows AI agents to take over tasks such as prompt adjustments and evaluation criteria updates, previously managed by experts.
Fujitsu has applied this technology to automate the enhancement and continuous evolution of business-specific large language models (LLMs). Multi-AI agents autonomously execute steps such as data selection and evaluation, reducing the need for expert tasks and continuously improving model performance. In the medical field, this has resulted in a significant accuracy improvement, demonstrating the capability of AI to automate and optimize business-specific LLMs.
The technology was also applied to the design specification search in large-scale business systems, such as Fujitsu's electronic health record system. AI agents learn from past search results and human corrections, improving search strategies and accuracy, thus reducing the effort required for designing and improving search logic.
Fujitsu plans to integrate this technology into its AI platform, offering it as a core technology for in-house AI development and autonomous operation. By collaborating with Carnegie Mellon University researchers, Fujitsu aims to advance the development of self-evolving multi-AI agent systems with reduced memory and power requirements, enabling their use in cloud, on-premises, and edge environments.
This initiative aligns with Fujitsu's commitment to the Sustainable Development Goals, as the company seeks to address societal challenges such as the shortage of specialized personnel and adaptation to regulatory changes through AI-driven innovation.