Appointed Leader of Govt-Affiliated AI Security Institute Says Only Promoting AI is Dangerous; Stresses the Importance of Safety Assurance

The Yomiuri Shimbun

15:45 JST, February 9, 2024

The government-affiliated Japan AI Safety Institute (AISI), which is responsible for ensuring the safety of artificial intelligence, is set to launch Wednesday. Akiko Murakami, the appointed head of the new entity, said, “You can’t step on the accelerator without a brake,” in an interview with The Yomiuri Shimbun on Thursday. She added, “It is dangerous to think only about promoting AI.”

Murakami, chief digital officer of Sompo Japan Insurance Inc., had engaged in the development of Watson — the question-answering system — at IBM Japan’s research lab. In 2021, she joined Sompo Japan where she leads on the active use of AI in business operations.

“AI development and risk management are done separately in a corporation, so the risk management is having a hard time keeping up with the technological evolution,” Murakami said. “It is extremely important to create an environment where system developers can provide risk managers information and advice on where to apply the brakes.”

AI has grabbed attention for its advantages such as an improvement of operational efficiency, but it is also said to have various risks including mass generation of disinformation and furtherance of discrimination.

Murakami said it was important to control AI-caused problems with technology and expressed hope for the use of those technologies that can help reduce AI risks and ensure security, such as digital watermarks that identify AI-generated documents and images.

AISI will be responsible for formulating standards used by AI developers to assess the safety of their products.

“We will work with industry, academia and government to compile guidelines that present points where it should not be done or must be proceeded with caution,” Murakami said.