As the development of AI technologies has become an international trend, demand for relevant legal requirements has gradually increased. Training AI models requires a massive amount of existing copyrighted works in the process of machine learning. However, AI tool developers face significant challenges to obtain individual authorizations from each copyright owner. On the contrary, excluding copyrighted works from AI training data in order to avoid infringing others’ copyrights may both hinder the development of AI technologies and harm the overall social well-being. As a result, many countries have reviewed the defense of fair use or fair use of copyrighted works one after another, though most of the countries lack legislation applied to extensive reproduction of existing copyrighted works for AI training. 

 

 

In Taiwan, the Executive Yuan originally planned to draft Artificial Intelligence Basic Law in September 2023. However, after taking into account the recent developments in generative AI technologies, the launch of the draft is postponed to 2024. With that, among the guidelines for the use of generative artificial intelligence within the public sector promulgated by the Executive Yuan for the Executive Yuan and affiliated agencies (institutions), the self directions established by respective ministries/institutions are given precedence. In addition, considering various and different drafts that have been proposed by legislators since 2019, as well as the draft legislation explanations over the years that have been provided by the National Science and Technology Council under the Executive Yuan in the past years, the drafted AI basic laws in Taiwan focus on ethical principles, national policies, protection of privacy and personal information, regulatory sandboxes, alongside assisting mechanism to supervise the innovative AI technology. However, they did not cover the regulations associated with copyright. 

 

As a result, the following questions remain to be addressed: how to cope with industry’s demand to utilize the existing copyrighted works to enhance the AI technologies in Taiwan; is there room for fair use; whether a legalized authorization system is needed to obtain permission from copyright holders. 

 

 

Japan’s Copyright Act has enabled the AI industry to utilize copyrighted works for the purpose of information analysis or machine learning under certain conditions. Singapore's Copyright Act also permits users to make a copy of and communicate a work to which they have lawful access for the purpose of computer data analysis such as machine learning.

 

As the European Parliament and the European Council reached a provisional agreement in December 2023 for the world’s first comprehensive law that regulates AI, namely the Artificial Intelligence Act (the “AIA”), this has gradually formed the standards of AI-specific law. Though the preliminary versions of the AIA did not include the regulations related to copyright, however, as the tensions between copyright holders and AI model developers have emerged due to launches of the generative AI technologies, amendments of the AIA now involve the transparency obligations where AI developers are required to disclose a comprehensive listing of the copyrighted materials used for training their algorithms, accompanied by precise identification of rightsholders, as an attempt to balance both industry development and interests of copyright holders. 

 

 

In response to emergence of an innovative AI technology, and given the far-reaching characteristics for online applications of AI technologies, every country should find it necessary to stipulate relevant regulations. With amendment of the copyright laws that establish exceptions for copyright infringement in Japan and Singapore, and on top of the lawsuits regarding copyright infringement in the U.S. and gradual formation of the world’s first AI Act “the European Union's Artificial Intelligence Act,” there is increasingly clear standards of fair use of others’ works. On the industry front, AI developers, when developing and perfecting new technologies, should pay continued attention to development and density of regulatory in various countries, and accordingly adjust technical content and authorizing procedure, in order to lower the risk of copyright infringement.