Navigating a new and uncertain landscape
At the recent Publishers Licensing Services conference, RightsZone’s Clare Hodder was part of a session called Asking the Right Questions: A Practical Guide to Licensing content for AI. Based on informal research carried out with colleagues in the industry, the talk posed a series of questions that rights professionals might want to consider when thinking about embarking on AI licensing deals.
In this blog, we share some of the key take aways from that session, as well as questions rights teams should be asking.
AI licensing – an exciting opportunity?
Amongst all the hype, fear and lawsuits surrounding Artificial Intelligence (AI), there is actually something quite exciting happening for rights professionals. We find ourselves at the forefront of developing a new rights sales channel. Whilst we wait to hear the outcomes of the various legal cases against those companies that have taken intellectual property without the consent of rights holders, many other companies (and even some of those in the court rooms) are actively seeking licences to include published works in their AI projects. Whilst a potential new channel for rights sales is exciting. At the same time, it presents some challenges.
The licensing challenges we face
This type of licensing doesn’t quite fit the usual rules. Concerns about how to protect the creative content that we are the proud custodians of in this new era are real. Nevertheless, if we don’t find a way to make this new licensing channel work, the arguments of those calling for exceptions to copyright law may be strengthened. Perhaps, there is a way to engage with tech companies and license content to them responsibly, with appropriate guard rails and with adequate remuneration, much as we do with other licensing partners.
For the rights and licensing professionals exploring this new opportunity, the route ahead is uncertain. There are no precedents to follow. So, how do we decide which direction to go in?
Here are some questions to consider that might help you plot a course:
Do we have the right rights?
As with any new rights deal, establishing whether or not you have the rights being asked for is a critical first step. That means checking contracts with authors and other agreements that may be relevant. The particular rights needed may vary depending on the requirements of the licensee, the rights you have may change with reversions, acquisitions and divestments. It’s important to be able to check and re-check as circumstances change.
A clear difference between AI licensing and more traditional rights licensing deals is the need to provide content, as well as rights. Identifying which formats you have content available in, where it’s held, how it can be extracted and delivered are all important questions to consider.
AI licensing deals move quickly so being able to supply lists of available content and files at scale and at speed is a big shift for many rights teams. It is likely to require new workflows and collaboration across teams.
How exactly will our content be used?
The way in which your content will be used will vary depending on the particular use case. In the creation of Large Language Models (or LLMs), pre training requirements are different to those needed for Fine Tuning or Model Evaluation. Meanwhile, companies creating Retrieval Augmented Generation (or RAG) models will use your content in a completely different way.
Talk to those you are considering licensing your content to. Get a good understanding of exactly what will happen to the content. These insights will help you to identify potential areas of risk and what sorts of guardrails you could introduce to mitigate those risks.
All of this is new to those developing AI tools too. Choosing partners who you can collaborate with effectively, learn from and co-create solutions with is beneficial to both parties and a good foundation for building the long term, trust-based relationships on which licensing thrives.
What’s the value of our content?
The value of your content to one potential licensee may be several magnitudes more or less, than its worth to another.
Licensing for a large LLM training data set might see all content valued equally and a premium being paid for quantity and speed of delivery, rather then the relative quality or type of content being licensed. Whereas a specialist RAG model, focused on a very niche area, might rely on a smaller amount of highly relevant, expert-created material. However, this is true of all licensing deals, not just AI licensing.
Valuation is dependent on things like:
- context of use
- availability of alternate sources
- risk to cannibalisation of sales
- the particular quality of a work
- unique character of a work, or that of its creator.
In traditional licensing, it is common to trade contractual terms for an increase or decrease in fee. For example, varying rights granted, term of agreement, particular obligations etc., and the same sorts of negotiation tactics can be used when negotiating AI licensing agreements. Through dialogue with your customer, understand how the content will be used. You can then consider how to negotiate a price that maximises value and mitigates risk.
Why are we doing this AI licensing deal?
There are a number of reasons why your organisation might want to enter into AI licensing deals, not least to try and regain control of what is happening with the content that you manage. We know that content is being taken from a variety of sources, often without permission. So, entering into a licensing arrangement means that you can exert some influence in terms of the partners you want to work with, what content they have access to and what they do with that content.
Working in this way also enables your organisation to learn. Your teams can build experience and understanding in how content is being used in the artificial intelligence space, which will help inform future licensing deals, as well as broader strategy. Of course, AI licensing deals also bring in a welcome new additional revenue stream.
Being able to control, manage and seek remuneration for the content being used is just as important for any creators, on whose behalf you are managing some of these rights. It can bring creators additional opportunities too. In our research, we heard of an author who obtained paid consultancy work, as a result of a limited RAG licence that was entered into for a series he had authored.
In opposition, your organisation may have good reasons not to license AI companies. This is also understandable. Not every prospective licensing partner or opportunity will be a good fit. You might not want to licence all of your content and you might not want to license for every use case. The AI licensing route will be different for everyone.
High quality input = higher quality output … that makes a difference
It is worth considering the potential benefits of licensing high quality published works for use by AI companies, that goes beyond our organisations and the creatives we work with. If models don’t train on published works, they are training on whatever they can scrape from the internet. Training models on high quality input is likely to lead to higher quality output.
There are some very exciting projects happening in the world. These are making a positive impact on humanity, thanks to advances in AI. See for example this McKinsey report about AI projects in relation to the UN Sustainable Development Goals). This interview with Jane Goodall about how AI is helping the Born Free foundation to protect wildlife and support local communities too. And finally, this article which outlines how AI is helping museums to unlock and share more stories from the past). All of these rely on models trained with high quality data.
As we touched on earlier, licensing enables us to evidence that there is a healthy and functioning rights market. It gives us leverage to push back against claims that copyright is “unclear” or too difficult to navigate. New copyright exceptions are not required, if there is no market failure to correct.
AI licensing – learning and navigating our way together
There is a lot of uncertainty about how AI will evolve and what that means for published content. The relative risks and rewards of licensing or not licensing will vary from case to case. But, in many ways that’s not a situation that is very different to any other kind of licensing agreement that rights and licensing professionals deal with.
We know how to nurture relationships with customers.
We know how to evaluate an offer.
We know how to protect our content in agreements.
We, instinctively, know how to make deals that work for our organisations and our creators.
Even as we navigate through an uncertain landscape, we can be confident that our skill and experience as rights professionals will guide the journey ahead.
And if you want support or to talk about this more, we are only an email away.