Govt Did Not Explain AI’s Copyright Risks Ahead of Law Revision

Ahead of a legal revision that made it possible for copyrighted material to be used to train artificial intelligence models, the government did not adequately explain the risks to copyright holders, The Yomiuri Shimbun has learned.

Even though a government advisory panel had discussed issues related to a revision of the Copyright Law, taking into consideration the emergence of advanced technologies such as generative AI, Japan ended up with the laxest legal framework in the industrialized world, without informing copyright holders of the risks involved.

Article 30-4 of the revised law enacted in 2018 permits the use of copyrighted material including text and images for machine learning, regardless of whether the applications of the technology are for profit.

The clause states that such material cannot be used “if the action would unreasonably prejudice the interests of the copyright owner,” but the Cultural Affairs Agency provides only a limited list of applicable cases.

Amid the rapid spread of generative AI tools such as ChatGPT, infringement concerns are growing among copyright holders.

In October 2015, a working team comprising professors and lawyers in the Copyright Subdivision of the agency’s Cultural Council began discussions on the law revision, with the implications of AI and big data in the so-called the Fourth Industrial Revolution in mind.

Based on industry opinions, the team considered the development of a so-called cyber-physical system (CPS), in which large amounts of data are collected, stored and analyzed by high-performance AI.

The team held hearings with copyright holders in June and August 2016. An explanatory document presented in advance addressed the possibility that copyrighted material could be used in an envisaged cyber-physical system, but did not specify the risk of copyright infringement.

“CPS was a concept that included generative AI,” an agency spokesperson said. However, the emergence of generative AI was not addressed during discussions.

At the hearings, the Japanese Society for Rights of Authors, Composers and Publishers (JASRAC) and other organizations expressed copyright infringement concerns regarding technological innovation and the development of new businesses. The team said the risk of copyright infringement would be zero or minor because the use of copyrighted material in the envisaged cyber-physical system would be limited to processing.

The working team’s report, which was compiled in February 2017, formed the basis of a revised bill after discussions at the Cultural Council, and the Diet subsequently passed the bill, which was enacted into law.

Most of the team members interviewed by the Yomiuri Shimbun said Article 30-4 “is a provision that can flexibly respond to technological progress.”

However, generative AI applications can create content that is the same or similar to the source material, including copyrighted works, on which it has been trained.

“Discussions at the time looked into generative AI, but we did not expect the technology to develop so fast,” said lawyer Zen Tatsumura, a member of the working team.

Team member Prof. Takeshi Maeda of Kobe University said, “The emergence of interactive AI such as ChatGPT occurred much sooner than expected.”

Lawyer Kazuo Makino, who specializes in copyright law, said, “Even though the rapid progress of generative AI was not envisaged at the time of the law revision, examples of specific cases that ‘unreasonably prejudice the interests of the copyright owner’ need to provided and such cases need to be addressed.”