AI, LLM training: DPIIT mulls govt-fixed royalty rates for content creators

The government has proposed that technology companies using content creators’ data to train their artificial intelligence (AI) and large language models (LLMs) pay royalties to these creators at a rate set by the government or a court.

 

The new framework, proposed by the Department for Promotion of Industry and Internal Trade (DPIIT), under the Ministry of Commerce and Industry, observes that payment for this blanket licence should be collected by a unified industry body, which would then be responsible for distributing the proceeds among content creators.

 

DPIIT released these guidelines as part of a working paper late Monday night, giving stakeholders 30 days to submit their views. In its draft report, the DPIIT rejected a voluntary licensing mechanism, arguing that such a framework would impose a “significant compliance burden on those seeking licences, due to the complexity of negotiations and the associated uncertainty”.

 

A government- or court-fixed rate under a statutory licensing mechanism would, instead, cut costs and create a “predictable environment for licensees of works”. “While this model takes away the power of copyright owners to refuse licensing or negotiate a fee, it guarantees them fair compensation,” the DPIIT report said.

 

The statutory licensing framework will also ensure that AI models are free of bias or hallucinations, as the maximum amount of content would be available at reasonable rates to all companies training AI models and LLMs, the working paper suggested.

 

The framework should also allow the parties — both companies that use content to train AI and LLMs, and content creators — to challenge or seek a review of the pre-determined rate before a judicial forum.

 

Experts, however, believe the working paper’s premise — that fair compensation for content creators can be practically and accurately administered through a mandatory collective mechanism for AI training — is flawed.

 

“The very mechanism intended to simplify licensing and reduce transaction costs for AI developers creates an intractable ‘black box’ problem in the royalty distribution phase. Contemporary generative AI systems process massive, diverse datasets in iterative and non-deterministic ways that inherently do not preserve work-level traceability or contribution,” said Jameela Sahiba, associate director at technology policy advocacy body The Dialogue.

 

While the hybrid model may initially increase companies’ compliance costs, it could lead to better record-keeping, cleaner licensed datasets, and standardised ways to track data sources, said Neelima Vobugari, cofounder and chief operating officer at AiEnsured.

 

Other experts believe that, though the intent behind the framework is positive, it could create significant issues if implemented in its current form. “The regulatory load will be enormous, and expecting a central agency to manage filings, verify claims, set royalties, and act as the middleman between creators and AI companies feels unmanageable. Almost every decision will still be open to judicial review, so who will realistically take that risk?” said Pawan Prabhat, cofounder of Shorthills AI.

 

Objectives of the framework 

• Availability of lawfully accessed content for AI training as a matter of right
• Fair compensation to copyright holders
• Rate setting via a quick, transparent process, affected parties can challenge predetermined rates
• Facilitate both big players, startups, and individuals in the AI tech
• Upholds the basic principles of copyright and rewards human creativity