日本語 
Japan’s technology and energy sectors are increasingly exploring advanced artificial intelligence (AI) to streamline complex workflows in regulated industries. In a recent collaboration, NTT Docomo Business and Chugoku Electric Power have begun developing and testing a power‑sector‑specific large language model (LLM) based on NTT’s Japanese‑language AI platform, tsuzumi 2. This initiative aims to harness generative AI to improve both efficiency and accuracy in utility documentation and regulatory compliance processes.
A Tailored AI for Japan’s Power Sector
The utility sector, particularly in the areas of electricity production and distribution, has to operate in a very regulated environment, involving a lot of reporting, document preparation, and compliance checks. Conventional generative AI models may find it difficult to provide accurate results in such a domain, where technical expertise and domain-specific vocabulary are of utmost importance.
To address this, the partnership between NTT Docomo Business and Chugoku Electric Power is training a custom LLM using tsuzumi 2, a model optimized for Japanese language comprehension and generation. Rather than relying on general‑purpose AI alone, this model incorporates Chugoku Electric’s internal manuals, regulatory forms, and historical submissions as training data. The idea is to embed domain‑specific expertise directly into the model so it can generate content that adheres closely to real‑world energy sector rules and practices.
Also Read: Japan Moves to Draft National AI and Robotics Strategy, Targeting Service Robot Gap
How It Works — Beyond Generic AI
The development process includes several key elements:
Curated Training Data: Chugoku Electric provides internal documentation — including manuals and past regulatory filings — that are frequently used in daily operations. This creates a rich, real‑world dataset tailored to power sector needs.
Model Fine‑Tuning: NTT Docomo Business processes and formats this data for tsuzumi 2, ensuring it can be effectively learned and applied by the LLM.
QA Based Benchmarking: The team creates a specialized QA set that assesses the model’s accuracy in both common queries and domain‑specific tasks. This evaluation guides iterative retraining to improve precision.
This approach aims to deliver a model that not only understands natural language but also interprets complex energy industry concepts and regulations with high fidelity.
Bridging AI and Regulated Workflows
One of the core challenges of applying AI in regulated industries — such as utilities, aviation, finance, or healthcare — is that generic AI models can misinterpret highly technical questions or generate responses that are plausible but incorrect. This can have serious consequences when compliance is involved. By embedding domain expertise directly into the training data, the new LLM aims to minimize these risks while enhancing productivity.
NTT’s LLM technology has evolved specifically for Japanese language use cases, with earlier versions like tsuzumi already deployed for business applications. Beyond power utilities, such fine‑tuned models are gaining traction across sectors in Japan as companies seek to harness AI for document automation, customer service, and knowledge management.
What This Means for the Power Industry
The pilot between NTT and Chugoku Electric Power may bring about several key benefits to the industry:
- Faster Regulatory Filings: The companies in the industry are known to draft filings for the Ministry of Economy, Trade, and Industry (METI) or similar bodies. This could be accomplished faster by an LLM.
- Quality Control: The outputs could be made to conform to industry requirements by incorporating industry knowledge into the model.
- Expert Knowledge Access: Smaller utilities or teams with limited specialized expertise can leverage the model to access deep procedural knowledge more easily.
- Cost Efficiency: Long‑term deployment of such AI solutions can lower labor costs associated with documentation and compliance workflows — a key consideration in a highly regulated sector.
Broader Implications for AI in Critical Infrastructure
The Japanese utility industry, like many developed nations, faces the challenge of modernization while ensuring safety, availability, and regulatory requirements. The adoption of sophisticated AI tools within the utility industry’s core business indicates a trend towards the development of AI-enhanced operational systems rather than purely automated systems.
Across the world, the utility industry is actively pursuing the application of AI technology in grid optimization, real-time demand forecasting, and fault detection. Nevertheless, the application of domain-specific LLMs is a relatively new and developing field. Recent academic studies have pointed to the efficacy of customized AI solutions for power grid-related applications, such as diagnostics and decision support, indicating that specialized AI models are more effective than general AI models in a professional setting
The Japanese focus on industry-specific language models is also reflective of the overall AI development landscape in Japan, where several firms and academic groups are engaged in the development of Japanese-optimized LLMs to better meet the needs of the Japanese business community.
Challenges Ahead
Although the potential of a power sector LLM is great, there are important issues to be addressed:
Data Quality and Governance: It is important to ensure that the data used to train the LLM is representative, accurate, and privacy and security compliant.
Model Validation: In the area of compliance, it is important to have a robust validation process to avoid providing misleading results.
Scalability: In order to expand the model to include all aspects of the business of the utilities, including reporting and customer service, it will be important to continue to develop and refine the model.
Looking Forward
As the utilities industry worldwide begins to leverage AI to transform their operations, the Japanese pilot project utilizing a sector-specific LLM shows that there is a clear way forward for the regulated industries to harness generative AI. If successful, this pilot project can be a model for other utilities and industries to leverage AI to think like an expert in even the most specialized fields.
English