Forum: AI copyright debate should distinguish training from authorship

Edmund Lam’s response (Creators seek to keep existing rights, not gain new ones, Oct 2) to my commentary (The risk of overprotecting human content creators, Sept 30) brings together several questions that should be kept analytically distinct: the protection of existing copyright, the use of copyrighted works for AI training, and the threshold for copyright in AI-assisted outputs.

My commentary did not suggest that creators should surrender rights they already possess. I expressly recognised that copyright allows owners to control certain uses of their works, including through permission and licensing. My concern was narrower: whether, as the law adapts to AI, human authorship should itself become a basis for stronger or expanded protection.

In particular, my commentary was principally concerned with when an AI-assisted output contains sufficient human creativity to attract copyright, rather than the distinct question of the terms on which copyrighted works may be used as inputs for AI training.

That distinction also frames my references to Google Books and Singapore’s computational data analysis exception. I did not suggest that Google Books and generative AI are equivalent, or that every commercial use of copyrighted material for AI training should be permissible. I expressly noted that Google displayed only limited extracts that did not substitute for the originals. My narrower point was that computational analysis of copyrighted material can generate wider social benefits.

Singapore law recognises computational analysis, including machine learning, while the Government is reviewing whether the present balance remains appropriate. Questions of training, licensing and remuneration therefore merit consideration on their own terms.

The clearest difference concerns Lam’s proposal that a human contribute at least 75% of a work’s originality. A numerical threshold may offer apparent certainty, but originality is a qualitative legal concept. A court would still need to determine what constitutes the relevant creative contribution, how human and machine inputs should be assessed, and why 75% is the appropriate dividing line.

My proposed test of meaningful human creative control addresses that inquiry directly, while recognising the need for clearer guidance on its application. The aim is not to diminish human creativity, but to identify when human creative contribution is sufficient to justify copyright without treating either human effort or a numerical threshold as a substitute for originality.

Ben Chester Cheong