Working Paper on AI and LLM
Context:
A committee constituted by the Department for Promotion of Industry and Internal Trade (DPIIT) has released a working paper titled "One Nation, One License, One Payment: Balancing AI Innovation and Copyright".
The paper addresses the conflict between AI developers needing data to train Large Language Models (LLMs) and content creators demanding compensation for the use of their copyrighted work.
Key Recommendations:
Mandatory Blanket License:
The committee rejects voluntary licensing (individual deals) and proposes a mandatory statutory license.
This grants AI developers the right to use any lawfully accessed copyrighted content for training without seeking individual permission from creators.
In exchange, AI developers must pay royalties to rights holders.
No Opt-Out Mechanism:
Copyright owners cannot opt out or withhold their works from being used for AI training, provided the access is lawful.
Institutional Framework:
A new centralized body called the Copyright Royalties Collective for AI Training (CRCAT) is proposed to collect and distribute royalties.
It will include copyright societies and collective management organizations (CMOs).
Royalty Determination:
Royalty rates will be fixed by a government-appointed committee (not market negotiation) and may be reviewed every 3 years.
The paper suggests a flat rate or a percentage of the AI developer's gross global revenue.
Retrospective Application:
The obligation to pay royalties may apply retroactively to AI models that have already been trained on copyrighted data.
NASSCOM's Dissent:
The industry body for IT companies dissented, arguing that forced royalties act as a tax on innovation.
They advocated for a Text and Data Mining (TDM) exception (allowing free use of publicly available data) or at least an opt-out mechanism for creators.
There are fears that high royalty costs could create entry barriers for smaller AI start-ups compared to big tech giants.