AI watermarks erase human authorship twice
Models absorb human creativity without consent, then label human-directed work as machine-made, creating a troubling double erasure of authorship.
Artificial intelligence systems imprint invisible markers on human-created content, classifying it as machine-generated. This double erasure strips the original author of credit during both the training phase and the final output. Models learn from human writing, sometimes without consent or compensation, before later labeling their own work as non-human.
These text watermarks, mandated by regulators, bias the generation process to embed a statistical fingerprint, making it impossible to discern the extent of human input from AI influence. The same tools designed to detect synthetic content inadvertently flatten the spectrum of human involvement into a single "machine-made" label, regardless of payment or effort.
Payment does not exempt users from this classification, leaving the system fundamentally unsuited for crediting human authors. Even if detection tools become available, they cannot definitively prove that an AI was involved, only that a model touched the text. The watermark is intended for machine systems, not human readers, and thus remains invisible to most audiences.
Light editing can remove the mark entirely, while thorough rewriting may not. The technology primarily flags those least capable of circumventing it, rather than effectively addressing the issue of synthetic content. Legal disputes surrounding AI training on copyrighted material remain unresolved, with many cases pending. Companies have relied on fair use to justify using human-created works for model training, despite the lack of compensation.
The law is still evolving, with no definitive rulings on whether training AI on copyrighted material constitutes infringement.
Written by urgent.news from The Jakarta Post's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
