Taiwan seeks mechanism to pay authors for AI training content — Taipei Times
The Taiwanese government is studying ways to compensate authors and other content providers whose materials are used to train artificial intelligence models. However, authorities have not yet determined how such content will be assessed and licensed or on what basis payments will be made, Taipei Times reports.
The issue was raised by Democratic Progressive Party legislator Michelle Lin during a meeting of parliament's transportation committee. She asked whether the Ministry of Digital Affairs would provide materials for AI training free of charge and whether it was creating a mechanism under which AI developers would pay news media for using their materials.
Voluntary licenses and the public domain
Digital Affairs Minister Lin Yi-ching said the ministry collects four types of materials: works whose copyright protection period has expired and which have entered the public domain, book descriptions from publishers, excerpts intended to introduce publications, and works whose authors have voluntarily granted permission for their use.
According to him, the ministry has not approached Taiwanese news outlets for materials to train models. Authorities want major AI companies, including U.S. ones, to pay for the content they use. For now, the government is considering only public-domain materials and voluntarily licensed works to avoid lengthy legal disputes that could slow AI development in Taiwan.
More current news is available on the UA.News Telegram channel Telegram.
Corpus for sovereign AI models
At the end of last year, the ministry began compiling a corpus of high-quality texts in traditional Chinese for Taiwan's sovereign AI models. It initially included materials from central and local government bodies. On September 15, the agency announced participation rules for private publishers and e-book platform operators, providing for voluntary participation, the granting of permissions, and the removal of works by authors who decline to participate.
As of the middle of the month, the corpus contained 3.3 billion tokens — units of text that AI models use to process and learn language. At the same time, the procedure for calculating payments, distributing revenue, and the timeline for introducing the relevant mechanism have not yet been determined.
At a July meeting of the Taiwan Intellectual Property Office, representatives of magazine publishers and broadcasters proposed creating a licensing platform where media outlets could register content for paid use by AI companies. Chiu Ai-lung, director of the legal department at CommonWealth Magazine, also called on developers to disclose the list of licensed content and the data sources used to train models.
Mei Chen, secretary-general of the Satellite Television Broadcasting Association, said some international AI companies continue collecting data from media websites even after receiving direct prohibitions from owners. Meanwhile, Chang Jung-kuei, vice chair of the Taiwan Association of Software Innovation and Services, stressed the need to preserve the possibility of fair use of lawfully obtained data if the AI-generated result does not replace original works in the market.
