Japan's education ministry plans to implement new guidelines allowing elementary, junior high and high schools limited use of generative artificial intelligence, such as ChatGPT, to help formulate ideas to facilitate classroom discussions, among other uses, sources close to the matter said Thursday.
However, the ministry will not allow blanket use of the tool, with the draft guidelines saying generative AI should not be used in exams that measure students' academic performance or be used freely by students without them knowing the tendencies and limitations of the technology.
It also said its use to create poems or haiku or in artistic activities without careful consideration is inappropriate.
The Ministry of Education, Culture, Sports, Science and Technology is set to release the guidelines as early as July after hearing stakeholder opinions and making necessary revisions, the sources said.
The draft says it is important to "nurture abilities to thoughtfully use generative AI" but recommends a "restrictive" introduction, noting the potentially negative impact on students' critical thinking skills and creativity, as well as risks of personal data leaks and copyright infringements.
Teachers are also required to instruct students that it will be considered cheating if they submit essays written by generative AI as classwork or to contests as their own, according to the sources.
The guidelines emphasize that teachers and students should be mindful not to enter personal or confidential information into generative AI programs while being aware of the risks of copyright infringement if generated texts or images are made public outside of the classroom, such as on school websites, the sources said.
They encourage using generative AI to reduce teachers' workloads, such as creating draft programs for sports days, making exam questions or writing notices to parents. However, even in those cases, teachers need to edit and check the content, they said.
ChatGPT, developed by U.S. tech firm OpenAI, and other generative AI programs are trained using massive amounts of data from the internet and can process and simulate human-like conversations with users or create images based on users' instructions.