Artificial intelligence
Our position on generative AI
When generative artificial intelligence (GAI) systems emerged, we published our AI principles, setting out key considerations for the development of policy in the field of AI and intellectual property. In its regulatory approach, the UK has the opportunity to develop a regime which promotes growth for both the technology and creative industries, and in the process ensuring authors are treated fairly in the development of GAI systems.
Context
The emerging market for GAI services and systems trained on content is highly commercial. We see this with OpenAI adding a profit-making subsidiary supported by Microsoft with a multibillion-dollar investment, and later switching entirely to a for-profit model, with the company valued at £118 billion in October 2024. The reliance of these systems on English language works places UK authors at the heart of the debate around the use of their works to train GAI systems.
Copyright protects an author’s skill, effort and resourcefulness in choosing and presenting the words in their works, it protects writing. The value of works for training GAI systems similarly relates to written material, as opposed to using random or isolated words. Chief executive of OpenAI Sam Altman echoed this appreciation for uniquely expressed content: “it would be impossible to train today’s leading AI models without using copyrighted materials.”
Clarity and certainty
The use of copyrighted works to develop and train artificial intelligence is distinct from uses such as Text and Data Mining (TDM) which in their scope and definition predate the widespread emergence of GAI systems. In the UK there is a TDM copyright exception for non-commercial research and a functioning and developing licensing market for commercial uses – but these are separate issues from the use of content to train GAI systems. The EU has conflated TDM and GAI in its policy approach, leading to a now uncertain and contested system.
Current EU legislation relies on an opt-out system for copyright owners which is fraught with difficulties and won’t work in practice, as has been well set out by Ed Newton-Rex, former Stability AI senior executive and founder of Fairly Trained, in an extensive analysis. Among many concerns, an opt-out system would place a significant burden of enforcement on authors to monitor uses beyond their control and core profession. The existing technical options for opting-out are not fit for purpose, either being specific to URLs or file metadata, from which a creator’s work may often be separated.
Choice
The current GAI situation has been imposed on authors, whose content has been used by commercial operators without permission. In a survey of our members, receiving 13,500 responses, we found that only 7% of those who knew their works had been used to train AI gave permission for this use, while 91% felt that they should be asked for permission. In a short time, the use of these systems has become commonplace and commercially valuable with significant further profit potential. This is not a market that authors sought, and it poses wider issues from competition to impact on culture, society and the environment. If the market is to develop in a sustainable way, it must be based on choice not compulsion.
Transparency
Many authors are unaware as to the extent of the use of their works, the ALCS survey found that 77% of authors don’t know if their works have been used to train AI or not. We need a workable, regulated approach to create systems and data to identify with sufficient specificity the works of individual authors that have been used within GAI systems.
GAI training rights are a relatively recent and novel concept. It should not be assumed that authors previously signed over these rights in contracts agreed when such uses were not conceived. Greater transparency in authors’ contracts regarding past and future uses of GAI rights will also support the development of a well-functioning market.
Compensation for past use
Authors whose works have been used without permission should be compensated. Our survey reveals that 92% of our members would want to receive compensation for any historic use of their work for training GAI. The Science & Technology Committee shared this view describing as ‘inevitable’ the ‘agreement of a financial settlement for past infringements by AI developers, the negotiation of a licensing framework to govern future uses.’
Licensing markets for future use
Regarding future uses, market-based solutions are the quickest route to empower development of GAI that does not come at the expense of the UK’s authors and creative industries. This market is developing based on a range of licensing approaches. The foundation of this market is transparency for works used and works available to be used, to make legitimate GAI development and fair licensing possible.
Conclusion
New legal regulations to establish transparency for the use of copyright works for training generative Artificial Intelligence systems is essential to support a licensing market to develop and grow based on certainty, clarity and choice. Such an approach will simultaneously benefit the UK’s world-leading creative and technological industries, avoiding the disputes and lethargic progress that inevitably follows where one is set against the other.
Our principles on authors and AI
We have outlined a set of principles that must be considered in any future policy decisions. The principles are intended to be general and not restricted to any particular technology, which means they will continue to be relevant as existing technologies develop and new technologies emerge.
Human authors should be compensated for their work. New technologies must respect the established precedent of no use without payment. Licensing is an effective way of ensuring authors are paid for use of their work.
Technological developments must not be used to undermine authors’ rights. New developments in technology cannot be used as an excuse to erode the established fundamental rights of authors.
Licensing terms should be carefully considered to ensure they include every use of an author’s work. New technologies are capable of analysing an existing work and using the content to create many other works. Authors should be compensated every time their work has been used to create something new, not just when their work is first analysed.
Stakeholders must acknowledge the limitations of AI. AI is trained on data created by humans, and so has a tendency to repeat and reinforce human biases and misconceptions.
Policymakers and industry should seek applications of AI that support creators. If AI is used to diminish the role of creators on the creative process, it will have devastating cultural and economic consequences.
You can read the document in full here.
Our member survey on AI
Between June and August this year, we surveyed our members on their feelings towards artificial intelligence and two potential licensing options that would enable ALCS to compensate writers when their works were used. The survey was completed by 13,574 members and revealed that authors are aligned with the key themes of choice, recognition, transparency, and remuneration outlined in our Principles for AI and Authors.
You can find out more here.
