Taiwan Proposes Remuneration for Private AI Training Data
Taiwan’s Ministry of Digital Development plans to encourage AI companies to establish remuneration mechanisms when using private, high-quality data for model training. No concrete design has yet been proposed for pricing, licensing administration, or distributing payments, so the proposal remains exploratory. At an October rights-holder meeting, the digital, culture and intellectual-property agencies also discussed voluntary licensing for Taiwan’s sovereign AI training corpus and draft copyright guidance. The corpus is intended to reflect Taiwan’s language, culture and social context and had accumulated about 3.3 billion tokens by mid-October.
Taiwan is combining voluntary corpus licensing, proposed data remuneration and draft copyright guidance in an effort to shape governance around private data used to train AI models.
圖片來源:
數發部
生成式AI時代,模型能力的競爭不只取決於算力與架構,也取決於是否擁有高品質、具代表性的訓練資料。過去大型AI模型多利用網路公開資料進行訓練,但隨著AI服務逐漸商業化,內容創作者、媒體、出版與文化產業開始關注,作品如何被AI使用,以及資料提供者是否能在AI價值鏈中獲得合理回饋。
面對AI發展帶來的資料利用與權利保障問題,數位發展部、文化部與經濟部智慧財產局在10月中首度共同召開權利人意見交流會議,邀集內容產業代表、公協會、AI應用業者與相關專家,就AI訓練資料利用、授權方式、著作權保障,以及優質語料整備等議題交換意見。
在這次的交流會中政府部會透露政府對AI資料政策的思考,資料不再只是透過開放取得的資源,而是要建立更完整的授權與回饋制度。
AI需要大量語料,探索資料使用的回饋機制
數發部表示,AI研發需要大量高品質語料,但內容創作者則關切作品利用情形與合理回饋,如何兼顧兩方需求,已成為現今AI的資料治理議題。
#Taiwan Ministry of Digital Development#Taiwan Ministry of Culture#Taiwan Intellectual Property Office#Taiwan sovereign AI training corpus#Promoting Data Innovation and Utilization Development Act#Generative AI Copyright Guidelines