Japan's Government Drafts Diet Responses Using AI. Experts Question Whether Bureaucrats Understand the Risks.
Japan's government has begun using generative AI to draft responses to lawmaker questions in the Diet, raising concerns about whether officials grasp the stakes involved in automating policy communication.
The Digital Agency launched a pilot program allowing 180,000 central government employees to use an AI platform called Genai. Drafting Diet responses is a major function of the system.
The move addresses a real problem. Preparing answers to Diet questions has driven long working hours for bureaucrats for years. They must cross-check past responses against laws, regulations, and statistics to ensure consistency. Using AI to gather data and verify facts could free time for actual policy work.
But Diet responses are not routine administrative tasks. They shape how Japan's government addresses emerging challenges and set the nation's direction. They carry weight beyond explanation-they are acts of policymaking.
The Limits of Machine-Generated Text
Generative AI produces text based on patterns learned from internet data. Diet responses require something different: careful judgment about people's lives, conflicts of interest, national priorities, and context-specific meaning.
Past Diet answers often contain subtle diplomatic nuances. A 1972 response by then Foreign Minister Masayoshi Ohira on China-Taiwan relations illustrates the point. He said Japan hoped for "a peaceful resolution" and saw "no possibility of escalating into an armed conflict." This phrasing signals Japan's understanding of China's position on Taiwan while conditioning support on peaceful methods. An AI system cannot make such distinctions.
The government has said humans make policy decisions. But if officials rely on AI not just for data gathering but for drafting actual responses, that boundary blurs. The technology will influence outcomes whether or not anyone acknowledges it.
A Structural Problem
The editorial suggests lawmakers could reduce bureaucratic burden by submitting questions further in advance. This addresses the root cause rather than masking it with automation.
There is a larger risk. If government officials use AI heavily for responses, lawmakers may eventually do the same when formulating questions. If both sides of the Diet depend on the technology, the logic follows: why maintain the institution at all?
Learn more about generative AI and LLM capabilities and how they apply to AI for government use cases.
Your membership also unlocks:
