Koda wants to ensure that its members’ copyrighted content does not become part of the development of artificial intelligence (AI) without their explicit consent. This includes but is not limited to language models, foundation models, generative AI systems, and recommendation algorithms.
Koda expressly reserves the right to engage in text and data mining in accordance with Article 11 b (2) of the Danish Copyright Act (implementing article 4 in Directive (EU) 2019/790). Consequently, any utilisation of members' works for text and data mining requires Koda's authorisation, unless otherwise stipulated by law. Koda emphasises that the rules in Article 11 b of the Copyright Act and the underlying EU law, in Koda's view, do not provide access to text and data mining for the purpose of AI development. However, Koda is aware that this view is not shared, especially by international tech companies. According to the above, these rights are explicitly reserved regardless of this disagreement.
Requests for the use of works managed by Koda for text and data mining purposes should be directed to Koda at [email protected].
Background for reservation of text and data mining
The advancement of artificial intelligence (AI), including generative AI, is heavily reliant on substantial volumes of data. For effective training of an AI, access to comprehensive and diverse datasets is imperative. These datasets allow the AI to identify, learn, and comprehend intricate patterns.
When developing an AI capable of generating music and lyrics, these training datasets could, for example, consist of:
- Music (audio) – including mixed recordings as well as individual tracks or stems
- Lyrics
- Digital representations such as MIDI files
- Graphical representations in the form of sheet music/notes, chords, notations, tablatures
- Text-based descriptions of the music in the form of metadata and tags
- Data from composition databases, linking specific songs with specific songwriters.
By analysing these patterns, advanced models can be created that predict, simulate, and generate new data. These models – commonly referred to as large language models, foundation models, or generative AI systems – form the basis for many of the AI services we currently use and will use in the future.
This involves a comprehensive mapping of the sound and structures of music, enabling users to prompt an AI and have it generate music and lyrics.
The development of such services is contingent upon the contributions of music creators who, across generations, have enriched our musical culture and heritage.
In order to include copyrighted works in the training and/or development of AI, permission is required.
