Can statutory licensing fix the AI content crisis?

Content is the fuel that powers AI. Whether it’s training data for an LLM or information to help an AI agent answer a question, AI depends on vast amounts of human-generated content. In many cases, creators haven’t been compensated for the use of their work which has led to lawsuits in numerous jurisdictions. Statutory licensing is one way to balance the scales. In this post, we’ll look at what that might entail as well as some of the hurdles that might prevent these schemes from ever getting off the ground.

What problem does AI content use create for creators and copyright?

Many creators feel like they’re getting the short end of the stick at the moment. Not only has their content been used to train AI models, but the increasing prevalence of AI-generated summaries is reducing traffic to the original sources. A recent study commissioned by the European Parliament’s Committee on Legal Affairs found that when creators learn that their content is being used for AI training, they tend to reduce their output. That’s a problem when the whole point of copyright is to incentivize creators to continue creating. It also ultimately hurts AI companies, too, given that their models are still reliant on human-created work. 

Licensing can be an attractive solution for both AI companies and creators. It ensures that creators are compensated for their work while also encouraging them to continue creating which ultimately provides AI developers with a steady stream of data. There have already been moves in this direction such as Axios’ three-year deal with OpenAI and Amazon’s deal with The New York Times. 

But this isn’t a panacea. Major publishers have the clout to negotiate deals with companies; most creators do not. At the same time, AI companies may find it prohibitively complex to negotiate deals with numerous individual publishers. These constraints can ultimately reduce the pool of data. And if unlicensed data remains a viable alternative to the licensing regime, some AI companies will feel incentivized to avoid licensed content. 

How could statutory licensing regulate AI training data use?

A statutory licensing regime would address many of these problems. Guaranteeing compensation motivates creators to continue creating, and neither the AI company nor the creator have to invest time or energy into negotiating individual deals. It can also avoid the kind of disparities that can result when licensing deals are negotiated privately between AI companies and publishers. When that happens, bigger publishers will often be able to negotiate better terms than their smaller counterparts. 

Statutory licensing can take different forms:

• In compulsory licensing, a central authority sets the terms for use of copyrighted material and anyone can use the content provided they adhere to those terms. This can leave some creators feeling like they aren’t in full control of their work, however.
• In extended collective licensing, collective management organizations such the Copyright Clearance Center or the American Society of Composers are authorized to license content within their purview unless the individual creator opts out. This has the advantage of respecting the rights of creators who want more control over the reuse of their work.

Why is statutory licensing for AI facing political resistance?

Opposition to any type of statutory licensing is strong in the tech community. Understandably, many AI companies don’t want to be forced to pay for something they’ve been able to get for free until now, and they also voice concerns about the possibility of intrusive bureaucracy. 

The US government has also shown itself to be markedly hostile to any type of statutory licensing regime. The Trump administration has repeatedly made clear its preference for voluntary regimes. 

Moreover, the US hasn’t been afraid to use its geopolitical clout to discourage other countries from implementing statutory regimes. For example, Indonesia was forced to pledge that it wouldn’t impose any kind of bargaining code, digital tax, or otherwise require tech companies to pay for news when negotiating a trade agreement with the US. America’s increasingly confrontational and transactional foreign policy doesn’t bode well for statutory licensing. However, the EU report we mentioned earlier recommended that Europe pursue statutory licensing.

Is statutory licensing a realistic solution for AI content compensation?

Statutory licensing offers a coherent answer to a real problem. It guarantees compensation and levels the field between large publishers and everyone else. But a workable mechanism is not the same as a likely one. The tech industry has little appetite for paying for what it has so far obtained for free, and the US government has signaled at every turn that it prefers voluntary arrangements, both for itself and for everyone else. 

Therefore, the question is less whether statutory licensing could work than whether the political will exists to build it. For now, the momentum sits with private deals between AI companies and the publishers big enough to command them. Whether that leaves smaller creators better off or simply entrenches the disparities a statutory regime is meant to correct remains to be seen.

‍