Door Opens for AI-Assisted Music Royalties, but Verifying ‘Human Creation’ Remains a Challenge

Music created with the help of artificial intelligence can now be registered with the Korea Music Copyright Association (KOMCA) and generate copyright royalties if humans played a substantive and leading role in writing, composing or arranging the work. The problem is that objectively distinguishing what was created by AI from what was created by humans remains difficult. In practice, the system may have to rely heavily on creators’ own disclosures and honesty. . / Image generated by ChatGPT According to KOMCA, new reporting and registration standards for music created with AI assistance took effect Aug. 3. Works can be registered even when AI was used as a supporting tool, as long as a human substantially led the creative process. Songs generated entirely by AI based only on prompts are excluded. Registration, however, is a procedure for the trust management of economic copyrights and the collection and distribution of royalties. Registration with KOMCA does not in itself determine whether a work qualifies for copyright protection under law or establish who ultimately owns the rights. The registration process begins with declarations and assurances from applicants. Creators must disclose which parts of the work involved AI, what tools were used, how they were used and what creative work they personally performed. KOMCA initially reviews applications based on those disclosures. When necessary, it may request supporting materials such as project files and sheet music and conduct technical verification or an internal review. The association also plans to establish grounds for suspending royalty payments or recovering improperly received royalties when false declarations are identified. The central issue is where to draw the line for “substantive and leading” human creation. Some cases are relatively straightforward. A person may write the lyrics and melody and use AI only to supplement the accompaniment. At the other extreme, a user may simply enter prompts and have AI generate an entire song. But the distinction becomes far less clear when a person modifies an AI-generated melody or selects and combines several AI-generated outputs. Project files and edit histories can only show processes that were actually recorded. Copyright experts, including KOMCA and the Digital Intellectual Property Institute, believe AI music detection programs can identify, to some extent, works generated entirely by AI. But once a person modifies AI-generated material, it becomes difficult to determine the original AI-generated portion and the extent of human changes from the finished song alone. If creators do not disclose their use of AI, reconstructing the entire production process after the fact is virtually impossible. If false declarations cannot be detected, creators could delegate most of the work to AI while claiming that they led the creative process and still receive royalties. Improper royalty payments for AI-generated material could undermine confidence in the copyright royalty distribution system. Verification is likely to become even more difficult as AI functions become deeply embedded in ordinary creative software. When AI tools become standard features within existing programs, such as Adobe Photoshop’s Generative Fill, it becomes difficult to determine from the finished product whether AI was used and to trace the extent of human contribution. There are also limits to how far laws and regulations can close such gaps. South Korea’s AI Basic Act, which took effect in January, requires providers of generative AI products and services to indicate that outputs were generated by AI. It does not, however, require ordinary users who employ AI as a creative tool to disclose the scope of AI use or the extent of their own contribution. For artistic and creative content that is difficult to distinguish from reality, the use of AI may also be disclosed in ways that do not interfere with the audience’s experience. But even when a work carries an AI disclosure, consumers may still have no way of knowing precisely which parts were created using AI or how extensively it was used. The same problem could emerge across the broader content industry. A person could create a fictional universe and its characters, use AI to generate dialogue for a particular scene and then revise the output. From the finished work alone, it would be difficult to separate the contributions made by AI and the human creator. Similar practices are possible across film, broadcasting, advertising, games, webtoons and web novels, where AI-generated images, videos, dialogue and sound can be subsequently edited by humans. “Only the creator can know exactly where and how extensively AI was used,” a copyright expert said. “If someone creates a draft using ChatGPT or Claude and then revises it, determining where and how AI was used ultimately depends on the honesty of the creator who used it.” jubar@chosunbiz.com