In recent years, the use of AI in administrative fields has been vigorously discussed in Japan as well. However, there are many instances where introducing AI itself has become the goal, reversing the means and the ends.

The use of AI in public and administrative sectors should be approached with caution. Especially when it involves decisions that affect individuals’ rights and obligations, the top priority should be “rule-based automation grounded in laws and regulations (algorithmized using explicit if–then rules)“.

If procedures are too complex to achieve rule-based automation, the first step should be to review and simplify the systems and procedures themselves. Introducing AI hastily at this stage risks exacerbating problems. Fundamentally, it is necessary to transform the way laws are written into machine-processable formats—this is the true foundation of a digital nation.

Current laws are written in natural language premised on human interpretation and application, making them prone to ambiguity and context-dependency. When attempting automated processing by machines, some form of “supplementation” inevitably becomes necessary. However, using machine learning (ML)-based AI as a tool to fill these gaps is a short-sighted approach and entails significant risks, as AI carries unavoidable risks such as “black-boxing,” “bias amplification,” and “loss of accountability.”

The combination of “natural language laws × AI supplementation” should be actively avoided. This is not a denial of AI itself, but a warning against getting the order and roles wrong.

Looking at Estonia’s example, by building the necessary public databases and enabling real-time cross-referencing via a mechanism like X-Road, dynamic rule application that responds to real-time data updates becomes possible. On top of that, AI should be used only supplementarily in tasks separated from rights-and-obligations judgments, such as fraud detection or operational efficiency improvements.

In the future, reforming the legislative process itself—for instance, by drafting bills in structured data formats in parallel with natural language—would allow most administrative decisions affecting individual rights to be automated in a rule-based manner, significantly reducing the black-box risks of AI.

To summarize the proper order:

1. Clarify the purpose of the system (whose rights and obligations are being determined, and for what reason)
2. Simplify laws and procedures, and formalize knowledge (thorough inventory of exceptions and discretionary elements)
3. Maximize rule-based automation (determine how far automation can go without using AI)
4. Develop data interoperability infrastructure (X-Road type)
5. Carefully consider AI only for the remaining parts
6. Finally, classify and manage risks (if necessary)

Rather than forcing AI in for the sake of convenience, adhering to this order will enable Japan to realize a fair and trustworthy digital nation as well.