METI Minister Eager to Employ ChatGPT for Diet Responses

The Yomiuri Shimbun

13:53 JST, April 12, 2023

Economy, Trade and Industry Minister Yasutoshi Nishimura said on Tuesday ChatGPT could be used for the preparation of answers to Diet questions, if concerns are resolved. His remark will likely arouse controversy, given the number of issues surrounding this conversational AI model, such as the possibility of copyright infringement.

“If concerns such as the handling of confidential information are resolved, we would like to pursue the possibility of using AI to reduce the workload of public officials in the national government,” Nishimura said at a press conference after a Cabinet meeting.

Citing Diet questions as an example of possible use, he added: “We have tried a few things and received sentences that sound suitable as answers to Diet questions. So there is a possibility that the work burden could be considerably reduced if the answers were made based on the latest data.”

To improve the accuracy of conversational AI, it is necessary to train it on a large amount of varied data. However, there are concerns that some of the data collected may be ambiguous, confidential or personal, or that it may infringe on copyrights. The text produced may also not always be accurate.

“We would love to use this technology, but there are several issues that need to be cleared up at present,” said digital minister Taro Kono.

A senior economy ministry official described the technology at its current level as “not suitable for government work.”

Overseas, Italy has temporarily banned the use of ChatGPT due to concern that a vast amount of personal information may have been collected illegally.

Meanwhile, Internal Affairs and Communications Minister Takeaki Matsumoto said Tuesday at a press conference that Japan wanted to discuss AI during the upcoming digital and tech ministers meeting of the Group of Seven advanced nations, adding, “There is a need to balance development and regulation.”

Safety concerns

Experts have voiced concerns about the security of information used on generative AI and questioned the effectiveness of its introduction.

“Bureaucrats would have to scrutinize the answers composed by ChatGPT,” said Takao Komine, a visiting professor at Taisho University.

Komine, a former bureaucrat, said there were many advantages to using ChatGPT for government work, such as in researching foreign systems, but that the AI also “can generate forced answers out of publicly available information, making it is dangerous to use it unquestioningly.

“It is impossible to wholly entrust to ChatGPT the Diet answers, which represent the government’s official view,” he added.

Ichiro Satoh, a professor at the National Institute of Informatics, said, “There is a risk that bureaucrats will enter undisclosed information in preparing the answers, which will be learned by ChatGPT and can then be leaked.”