House of Lords Communications and Digital Select Committee inquiry: Large language models
The Creators’ Rights Alliance (CRA) is a collective of leading trade associations, unions and organisations representing and supporting the interests of creators in the UK. We advocate and campaign on policy issues such as fairer contract terms and working conditions, copyright, and intellectual property.
We have 19 member organisations who between them represent in excess of 400,000 creators working across the creative sectors, although their true reach far exceeds this, with millions more working in the creative industries. These include authors of original literary, dramatic, musical, and artistic works; producers or principal directors of films; performers of literary, dramatic, musical and artistic works; designers of original designs and semiconductor topography; and compilers or creators of databases.
Creator individuals and businesses contribute £115.9 billion a year to the UK’s economy.[1] Creative freelancers make up a third of the creative industry’s workforce.[2] We must all, including Government and AI developers, acknowledge the huge contribution our creators make to the UK’s creative and economic prosperity.
The CRA has answered many consultations giving our views on AI including responses to the House of Lords Governance of Artificial Intelligence (AI) enquiry, the Government’s White Paper on AI and the recent Generative AI in Education. These set out our position on the emergence of AI and the challenges and opportunities facing creators and rights holders. Copies on request contact@creatorsrightsalliance.org.
We also fully endorse other CRA member responses and that of the British Copyright Council (BCC) to this call for evidence and support all the issues they so clearly raise.
It must be emphasised that CRA members are not against new technologies. The CRA represents many creators who are increasingly working with AI technologies as both assistive and generative tools linked to the works they create. However, creators have concerns about the unregulated development and adoption of these technologies and the potential threats they represent.
1. How will large language models develop over the next three years?
Currently we have no way of knowing how Large Language Models (LLMs) will develop over the next three years; opinions conflict depending on who is asked. We already know however, that without safeguards the impact on creators and the creative industries will be devastating.
The Copenhagen Institute for Future Studies has estimated that 99 percent to 99.9 percent of the internet's content will be AI-generated by 2025 to 2030, especially if models like OpenAI's GPT-3 achieve wider adoption. Unregulated LLMs will be left to freely access what copyright protected works and creativity they want to sell on as their own, with no recourse or recompense for creators.
There is the stark possibility that without proper intervention and management these systems will develop at an unimaginable rate and with great influence. This could extend to social, creative, and even political agendas. The later we leave it, the more time and resources we will need to play catch up on practices that we know are illegal, unlawful and unethical, that are already taking place.
The House of Commons Culture, Media and Sport Committee Connected tech: AI and creative technology report has highlighted that: ‘The Government’s initial handing of the text and data mining exemption to copyright for AI development, though eventually correct, shows a clear lack of understanding of the needs of the UK’s creative industries. All branches of Government need to better understand the impact of AI, and technology more broadly, on the creative industries and be able to defend their interests consistently.’
The CRA and its members feel that there is now a real opportunity to ensure that we finally get it right. The next generation of AI and LMM systems should be created with legal and ethical compliance at their very centre. We must compel developers to make good the mistakes of earlier systems and to recall work that has been illegally accessed and assimilated.
LLMs require data to be fed as inputs. This usually takes the form of creative works, for example writing, music, art or performance. Currently many developers are accessing any creative work they wish, mainly through the illegal process of scraping creators’ websites to use their creative works to populate their systems as input data. The Government has yet to acknowledge these practices as illegal, unlawful or unethical. They infringe our carefully balanced copyright framework which is designed to encourage innovation whilst protecting and rewarding rightsholders. Acknowledging that there is a problem with this practice is the first step in addressing it.
Rightsholders’ work must be protected, and we must give them back control over their own work and how it is or is not used.
We agree wholeheartedly with the CMSC AI report that the Government should consider how creators can ensure transparency and, if necessary, recourse and redress if they suspect that AI developers are wrongfully using their works in AI development. Creators are already poorly paid and receive little recognition or protections from both Government and the wider industry.
Once work is ingested, it will stay in circulation in many guises across many cross-contaminated systems for many years. This leads to an additional difficulty posed by these infringements which AI and LLM systems cannot ‘unlearn’. As creative works have already been ingested illegally to populate these systems, the priority for LLM and AI developers must be to develop technologies enabling creative works to be undigested and ‘forgotten’, not only from theirs, but other systems across the AI/LLM multiverse.
It does not make good economic sense to risk the earnings, employment, and work opportunities of the creative individuals and businesses that contribute £115.9 billion (almost 6 per cent of the UK’s entire Gross Value Added (GVA) in 2019) and employ over 2.3 million people across the sector (Communications and Digital Committee, ‘At risk: our creative future’ (17 January 2023) HL Paper 125). By comparison, the AI industry is estimated to have contributed only £3.7 billion GVA to the UK economy last year, employing barely 50,000 people (AI Sector Study 2022, DSIT, 2023). Government must acknowledge and protect our creators if we are to deliver on its plans to grow the creative sector by a further £50 billion and a further million jobs by 2030. It must encourage and protect human endeavour and not sell it out. It should not give advantages to one sector at the cost of another. Government must protect and champion those workers who are at the creative coalface: those who create the work we enjoy, the music we hear, the images we look at or the cultural identity we share.
There are other impacts on other sectors of society where creativity is vital, for example universities and conservatoires which offer creative vocational courses (music, acting, singing, photography, design, dance, creative writing, film making, animation, games design, fine art etc); subject areas vulnerable to competition or replacement by generative AI technologies also appear to be absent from discussions. Switching from human creators to AI will have a massive negative impact on the ability of a very large number of people to earn a living; we already see creators being cut out of the creative process as a short-term way of cutting costs, and – even where they are involved – creators are being excluded from the chance to earn a living from their skills. If they cannot earn a living from their skills, they and the materials they create will disappear.
Intellectual property is fundamentally valuable to AI and LLM companies looking to collect and exploit large datasets of words, sounds, music and images in order to train their own AI platforms and commercialise them. These companies must be made to agree licensing terms with creators as do others in the publishing, advertising, and creative industries. It is the Government’s responsibility to protect and uphold the law, to protect the interests of human creative endeavour; these creators ensure the economic viability of our creative industries. Their intellectual property should not be simply given away freely and without acknowledgment for the commercial gain of others.
We must not let AI developers grow in influence and use their finances and position to stifle representation from individual creators who possess limited financial resources and do not have the access to comparable legal and lobbying representation that these larger, richer companies possess. Government has already illustrated that they are more attracted and influenced by AI and LLM companies, already taking high level representation from these groups and companies and not following up with the same level of interaction with creators and their representative organisations. Without creators these AI/LLMs would simply not exist.
Copyright-protected works have already been used unlawfully in the development of LMM systems that are already in the marketplace. There is ample evidence that copyright-protected literary works were ingested by AI systems to train LLMs; likewise entire catalogues were unlawfully copied from image libraries. There must be transparency of the lineage of information used for these systems. This must be published and accessible by those whose work is forming the basis of these systems, as well as regulators, so that they can ensure that the practice of assimilation is both legal and ethical. Developers must be compelled to publish where all their data inputs have come from, and how it was sourced; this is a crucial first step to building public and rightsholders’ trust.
a) Given the inherent uncertainty of forecasts in this area, what can be done to improve understanding of and confidence in future trajectories?
“These systems are in their own little data worlds and it is up to us to anchor them in in reality” Tom Chatfield, Chair, ALCS.
We should halt access to all ingested information that has been illegally sourced and ensure all current and future data inputs comply with copyright and related laws.
AI systems including image generators, large language models and audio generation tools have been widely reported to be trained on copyright-protected works without authorisation, consent or remuneration – a clear infringement of copyright law. There is an inherent challenge in pursuing redress for AI-related harms as they relate to copyright infringement since, in many cases, data has been scraped without rightsholders’ consent or knowledge, to develop many AI systems which have now entered the public realm.
Often there are thousands or millions of individual infringements which are not cost-effective to pursue but which collectively add up to significant wholesale infringements, conferring unfair market advantage and erosion of creators’ rights.
Copyright protection - No Text and Data Mining exceptions. The Government must commit to no Text and Data Mining exceptions whereby creators’ own work is used or ‘scraped’ by platforms for use in LLM datasets without their prior express permission; and licensing agreements should exist that clearly set out terms of usage and ensure that creators are remunerated at an appropriate level. They must be compensated for any misuse of their work.
LMM and AI developers must ensure that AI systems are designed so that they cannot be used to infringe copyright law, nor to create unlawful content. Given the territorial nature of copyright regulation, such assurances must be in line with the copyright laws of the territory in which the user is accessing the AI system.
A register of models and where they source data. UK Music have proposed an ‘input register’ for artificial intelligence companies, so that the use of works covered by our existing intellectual property framework can be properly monitored, and remuneration provided via collective agreements. This registry should record where and how these organisations plan to access the information that will populate their systems.
Open and truthful disclosure of where and how these systems are populated is essential if there is to be any confidence in the development of LLM and implementations of AI systems. If words, images, and other works generated by machine-learning systems are circulated without any declaration of their origins and validity, we face a time when none of us can trust that anything we see and hear is legal, true or authentic. This will have a huge impact on trust in news reporting or academic research. Artistic creative works must also be clearly distinguished from AI pastiches.
Opting-In: Opting-out is impossible, for one would need to know that there is an intention to access one’s work. Once it has been ingested by an LLM, it is lost forever. It will become cross-contaminated by countless other systems in a matter of seconds. To recall it is impossible and a waste of time. These systems are too flawed to provide any assurance that work can be recalled. Opt-in is the only logical and fair way.
No single licensing solution. Equity has highlighted that we must also recognise that a one size fits all licensing system will not work for all rightsholders, as actual impacts change sector by sector. For freelance creators and performers the ultimate control over whether to participate in licensing their work and rights for use in data sets should rest with the individual and only the individual. There is a very real possibility that large publishers, broadcasters, platforms and distributors – upon whom freelance creators and performers rely for work – will coerce creators and performers to license their work in data sets at the point of being engaged and on a ‘take it or leave it’ basis, in order to obtain licence fees for their organisations. Historically exploitative practices from the music, film, television, advertising and publishing sectors should not be replicated into new technology sectors.
Fair recompense for legal use and appropriate compensation for misuse. Those creators who allow their work to be used by LMMs and AI systems should be recompensed for any ingestion of their work. For those whose work is unlawfuly accessed, there should be appropriate compensation, including a penal element. The cost of such, and any system to obtain it, must be completely funded by AI/LMM developers and reflect not only the value of the work, but the time and impact on individuals of chasing such transgressions.
All distinct characteristics of individual performers and artists should also be protected. The Government must implement the Beijing Treaty, which the UK has been a signatory to since 2013 and also the recommendations of the recent CMSC report, which provides “moral rights” to performers. There must also be an acknowledgement and redress of existing unlawful misuse of artistic creativity which is driving the commercial viability of many of these AI functions. The same level of support and guidance to develop and nurture these systems should be extended to creators to ensure that they and their businesses are supported and protected from the exploitative practices.
Labelling. There should be a clear definition of what constitutes solely AI generated work, and work made with the intervention of creators and clear legislation as to when and how such works qualify for copyright protection.
Traceability and labelling need to be introduced to provide assurances to both human creators and the wider public, and to distinguish between real works and synthetically generated content.
Protect Livelihoods and Skills. Livelihood and human skills need to be given special protections, or they will be lost forever. Authors, artists, composers, designers, directors, editors, illustrators, journalists, musicians, songwriters, performers, photographers, translators, visual artists, and other creative professionals provide immeasurably important work that both the public and businesses enjoy and use. These professions and their skills cannot simply be left to cease, because technology can generate it more cheaply.
There must be recognition of the impact of AI/LLMs on potential future commissioning opportunities for creators, and of the impact on rates of pay and opportunities for those starting out to get important on the job experience.
An early impact assessment by OpenAI (https://arxiv.org/pdf/2303.10130.pdf) indicated that the exposure risk to poets, lyricists and creative writers was amongst the highest, at 68.8%.
In Writers and AI the WGGB reports that 61% of respondents agreed that: “The increased use of AI could replace writers in their craft area.” 82% agreed that: “AI developers should be transparent about what data they have used in creating AI systems, including where they have used writers’ work.”
In a briefing note for the House of Lords debate on AI, DACS are already reporting that artists are concerned that by promoting their work online they are vulnerable to having their works scraped for text and data mining purposes. There is a risk that creators will no longer make their works publicly available for fear of uncontrolled and unremunerated uses of their work. It is very clear that there is an imbalance in negotiating powers, and infringers are likely to have much deeper pockets than creators, rendering enforcement through the courts impossible.
Developers and Government must acknowledge the unequal impact that AI/LLM will have on individuals and groups who already experience challenging barriers whilst trying to access meaningful roles and work positions within the creative sector. Opportunities will hugely diminish when AI and LLMs will be able to produce content cheaper and simpler. To invest time, talent and money is a huge investment and challenge. We need to foster human creativity and encourage those from less represented groups, who already experience real barriers to careers in the creative industries. Careers in the creative industries are already low paid, undervalued, and insecure. Technology should not add to these challenges.
Independent Impact Assessment. The ability of AI firms to undercut creators’ fees by such huge margins that their careers will be potentially untenable is a cause of significant concern. An independently verified impact assessment must attempt to estimate future loss of earnings for performers and must acknowledge that not all rightsholders have equal status. This imbalance in negotiating power must be acknowledged and addressed.
Fairness and equal dealings with creators and AI/LMM developers. There is an important role for creator led trade associations, unions and groups, any proposed safeguards must have their full support. As well as staking high-level discussions with AI developers, the Government must work with creator groups to ensure any decisions and metrics used to regulate AI and LLMs are agreed across industry and sector and are not detrimental to creators and the specific business environments they work in.
Tax. There must be a level playing field in relation to the tax regime: currently tax incentives are offered to companies who invest in equipment and technology which create a clear financial incentive in favour of using technology such as AI systems to the detriment of hiring individual creators. Government needs to reconsider its tax regime to ensure it rewards and is fair to creators: tax incentives for those who employ an individual to work – which in turn provides important income to the UK economy, reducing the burden on state aid and increasing the overall wellbeing of society.
Collective Bargaining. We must strengthen collective bargaining so that trade unions, creator groups and bodies can tackle job displacement, remuneration and IP protection directly with engagers. This will encourage creators and creative businesses to continue to invest in our creative economy and encourage confidence from the next generation of creative professionals.
2. What are the greatest opportunities and risks over the next three years?
Creators will always use the best tools at hand to make more and more amazing work for audiences and businesses to enjoy.
AI can also be used to help identify where copyright infringements are already taking place and help take them down. 50% of WGGB Writers and AI respondents agreed that: “AI will be able to help identify where my work has been used.” However, 65% believed that the increased use of AI will reduce their income from writing, and 61% were worried that AI could replace jobs in their craft area(s). Though “if developed and used in an ethical, transparent and responsible way in partnership with civil society, government and trade unions it could be of benefit to writers and their careers”.
‘If LMMs and AI tools pay writers and creators fairly for the use of their work this could mean that writers and other creators are able to diversify and increase their income streams, allowing them to sustain careers. This will also mean that writing is not only an option for those who can afford to get through tough times.’ WGGB
The Creative PEC’s recent Authors’ Earning in the UK report already highlights that there has been a marked decline in the median earnings of primary occupation authors. In 2022, primary occupation authors earned a median income of £7,000 in respect of their self-employed writing activities. Assuming a maximum 48-hour working week at the £9.50 hourly rate (UK national minimum wage as of 1 April 2022) this would produce a total income of approx. £21,888 per annum before taxes. This suggests that writing cannot on average sustain an income that is consistent with a minimum wage. This has a disproportionate negative effect on those from groups which already have difficulties attaining, let alone maintaining a career in writing, especially those who have experienced long-term, intergenerational financial challenges.
We may also see the financial contribution which our creative economy makes to the UK being severely diminished or lost, as international AI/LMM technology companies may divert money away from the UK to elsewhere, in much the same way as we experience some internet companies not paying fair or proportionate amounts of tax. That will mean less spending power not only for individuals, but also for the Exchequer.
There is no possible replacement for the value of human imagination and creativity. The surge in interest in AI presents a unique opportunity to ensure the work of creators is recognised and properly remunerated and valued as work. This is in line with the #PayTheCreator campaign principles which asks that all creators are:
- Paid for the work they do, and the rights they grant, on time and reflecting their skills and contributions,
- Recognised for the contribution they make to the creative industries, the UK’s economy, and our wellbeing,
- At the heart of government policy and decision-making process.[i]
a) How should we think about risk in this context?
Above all else, support and recognition must be weighted in favour of human creativity and the right to earn a living through creative work and to retain ownership of that work. These must be the cornerstones of any future frameworks or legislation.
CRA member organisations have reported that their members are already experiencing significant misuse of their IP and loss of earnings due to the wholesale infringement of their rights by AI/LLM systems which have been trained on copyright-protected works without authorisation, consent or remuneration – a clear infringement of copyright law – and which have even sourced these works from pirate and torrent sites. AI systems cannot ‘un-learn’ the information they have been given, so works will remain part of the system for as long as it is online. There is no evidence that government has considered this harm, economic or otherwise, nor the impact of these breaches, in formulating their strategy.
KPMG research - Generative AI and the UK labour market (kpmg.com) notes that 43% of tasks fulfilled by authors, writers and translators may be automated. The new Head of the government’s AI taskforce stated earlier this summer that job losses are an inevitable consequence of AI systems. The Society of Authors have already reported that translators are informing them that they are no longer being hired to translate work, but instead, to correct machine-translated texts. Audio narrators and voice actors are having their voices simulated by AI systems without acknowledgement, authorisation, or compensation.
AI-generated artwork and photorealistic images are already being used in place of human-generated works – on book covers and in advertising – and being sold in competition with human-generated artwork on all major stock image platforms. Narrators and voice artists are having their faces, voices and likenesses simulated by AI systems in unauthorised performance synthesisation without acknowledgement, authorisation, or compensation.
Because of this free for all attitude, LLMs themselves are also at risk of being contaminated by illegally sourced data. 72,508 ebook titles were pirated in the Books3 dataset which was used to train Large Language Models like Meta’s LLaMA. Getty Images has acted in both the UK and the USA, accusing Stability AI Inc of misusing more than 12 million of its images to train its AI image generation system.
Environmental impacts in carbon emissions, material costs and subsequent damage of such technologies must also be considered; these systems use a lot of energy in their development and use.
Policymakers must hold AI developers accountable and ensure that the systems they develop do not discriminate or marginalise people – nor unfairly advantage others. We must ensure that any regulation of AI serves an anti-racist agenda, and one which also achieves equity in class, wealth, and gender, to name a few. AI developers must equally ensure that their AI systems achieve these goals.
3. How adequately does the AI White Paper (alongside other Government policy) deal with large language models? Is a tailored regulatory approach needed?
Please see the CRA response to the AI White paper attached.
The headline principles in the AI White Paper are laudable; we are however concerned that this set of cross-sectoral principles to '(i) guide regulator responses to AI risks and opportunities and (ii) describe good governance' does not provide appropriate or sufficient incentive for compliance by AI suppliers or users. In particular, any such principles must include ‘Compliance with all UK laws, including data protection and copyright’.
We continue to support the Government’s intention of ensuring coherence across the AI regulatory framework (at 13, White Paper) and whilst context-specific regulation is welcome (3.2.2.), there is a need for clear statutory obligations to ensure that rightsholders are able to contest and seek redress for AI-related harms, and it is crucial that these regulations cover both the development and the use of the technology to ensure legal compliance with existing laws.
The White Paper does not mention copyright or the impacts, either negative or positive, on creators or the rights they hold in their work.
In it paper AI and Creative Work the CRA has highlighted its concerns using the Government’s own five key principles:
Safety, security and robustness:
Transparency and explainability
Fairness:
Accountability and governance:
Contestability and redress:
Enforcement
CRA members are concerned as to how the Government’s proposed voluntary regulation will deal with enforcement. We are particularly concerned as to how individual rightsholders will be able to enforce their rights, given that the damages involved for individual uses are likely to be low value but add up to millions when aggregated to benefit the user (whether the AI developer or the user of AI systems).
AI developers and users must be reminded of their legal obligation to adhere to current copyright and related rights protections. This need for legal compliance should be stressed regardless of the solution adopted by government to regulate AI. There should be open disclosure of the datasets and works that have been assimilated by the AI system; how these were sourced; what rights were granted; and how and what recompense was provided for use of such works. A statutory duty for regulators merely to ensure parties have ‘due regard’ to the principles does not go far enough and does not send a strong enough signal to developers regarding their legal and ethical obligations in both the sourcing of their input data for the learning stage, and in the outputs and end products.
We share the view that any statutory duty should be strengthened with clear guidance set out by each regulator, and that this information must be easily accessible in both explanation and source. Copyright, moral rights and related rights protection should remain and there should be an obligation to abide by the current regulatory copyright framework across AI development and use. There should be the full implementation of moral rights in UK law, in line with the Berne Convention and international best practice.
It is important that any new regime be opt-in only, and there must be safeguards to ensure that any refusal by rightsholders to grant licences for use by LLMs and AI systems is respected. How will an individual performer’s rights be protected, and their privacy and personality? How do we deal with Deep Fakes?
The EU Directive on Copyright in the Digital Single Market requires all Member States to put in place transparency obligations, requiring parties to whom authors and performers have licensed or transferred their rights to provide information on the use of their works including revenues generated. The UK was one of 19 EU countries that supported the Copyright Directive in the final European Council vote. The government should not lose sight of the fact that it played a key role in developing and agreeing to the many necessary provisions within the Directive, which would make a significant difference to the rights of often precarious workers in the entertainment industry.
a) What are the implications of open-source models proliferating?
It is well known that once creators’ work has been ingested by these systems, it is there forever, almost impossible to recall. That is why it is so important that we ensure from the start that what is being used as data is legally and morally sourced. Without clear boundaries of how and by what processes data can be accessed and assimilated the implication is that creators, their work, opportunities and even our entire sector will be decimated by cheap, low quality and potentially unethical work and information. Who will ensure that outputs from these systems are truthful, moral, or ethical? Solutions must be developed where work that has already been ingested illegally can be recalled not only from the original offending system, but across all systems.
We are not calling for LLMs or AI to be stopped, but for safeguards and a robust framework to protect creators’ copyright and moral rights to be acknowledged and upheld.
We believe government has a central role to play in providing and funding intellectual property education and training for developers as well as consumers on the protections which rightsholders have in their work. Some of this can be delivered in partnership with organisations such as rightsholder groups like the Creators’ Rights Alliance and its members who already have resources and knowledge in these areas.
4. Do the UK’s regulators have sufficient expertise and resources to respond to large language models?[5] If not, what should be done to address this?
Compliance should not be seen as an obstacle or burden but an opportunity for developers to bring to market better systems that are assuredly safe and fair.
We agree with the CMSC that ‘all branches of Government need to better understand the impact of AI, and technology more broadly’ especially for those departments who will regulate it. It must have a better understanding of the impact on creators and their work. We also agree the Government ‘should provide a substantive update on its direction in managing the impact of AI on the creative industries and any discussions on these matters by the end of 2023’ and that “The Government should improve protections for creatives to prevent misuse of their likeness and performances by emerging technologies such as generative AI. At minimum, this should involve bringing forward ratification of the Beijing Treaty on Audiovisual Performances by the time it responds to this report.’’
The IPO has extensive experience of working with creators and of understanding the challenges they face within business. But the IPO and all Government departments must remain fair and balanced, and not biased unfairly towards LLM and AI agendas.
We must urgently implement the recommendation of the Impact of Covid-19 on DCMS sectors report for the formation of a ‘Creators Council’. This will enable the Government to better engage with representatives of the creative and cultural workforce, such as the CRA, to boost confidence across the sector and provide a voice and input during important and once in a lifetime events such as the development of LLMs and AI and the recent Covid pandemic. Such a body would function as a dialogue mechanism to better represent the views of creative individuals – many of whom are freelance – and the challenges they face. As businesses in their own right, and there being no current Creator Council, creators should have a mechanism for their voices to be included and heard at the very heart of policy making that directly affects their livelihoods and working conditions.
Compliance can only be achieved by close government co-ordination and the involvement of creator groups such as the CRA and legislative initiatives to create a level playing field between developers and the work they exploit in the development of their systems.
Regulation and guidance should be driven by the current copyright and IP framework. This already falls within the remit of the IPO. It is important that the IPO, DSIT, OFCOM and DCMS work closely to ensure all guidance is coordinated and consistent, and that it champions the protection of the rights of creative professionals against unlawful exploitation, be they authors, artists, composers, designers, directors, editors, illustrators, journalists, musicians, songwriters, performers, photographers, translators, or visual artists for example.
Any new framework or changes to legislation must strengthen provisions for creators, performance synthesisation and image rights in light of LLM and AI developments, not diminish them.
5. What are the non-regulatory and regulatory options to address risks and capitalise on opportunities?
Regulation must also address contracts and address how their professional contribution may be exploited due to unclear contract provisions, unspecified usage, and a lack of auditing rights. Artists should have the legal right to audit and access information about any uses made by AI systems. Producers and engagers should also comply with this right by reporting relevant information in good faith and ensure all commercial partners follow the same obligations.
The Government should also Implement the CRA Fair Terms for Creators: CREATOR bill which makes provision for important protections for creators against bad practice and sets simple standards for any future challenges they may face.
Fair Contract Terms for Creators highlights the seven areas of contract, engagement and usage terms which need to be included in all contractual and non-contractual use of creators and their work to protect all of them, including members of the public, when they sign a contract or their work is being used. These seven areas form the acronym CREATOR. They are all already in use elsewhere in the world or mirror existing consumer protection rights. Legal protections are needed to ensure that creators work in an environment which is fair and ‘good’, and legislative recognition of fair terms in line with CREATOR would achieve this.
Clarity: Clear contracts, in written form, which transparently set out the exact scope of the rights granted/assigned/licensed and are promptly and transparently negotiated.
Remuneration: Fair Remuneration. Equitable and unwaivable remuneration for each use/exploitation of work, for example ‘bestseller clauses’ so that if a work does far better than expected, the creator shares in its success even if copyright was assigned.
Exploitation: An obligation of Exploitation for each mode of exploitation. Also known as the Use-it-or-Lose-it Clause.
Accounting: Fair, understandable and comprehensive Accounting clauses.
Terms: Reasonable contract Terms (for example time limited) with regular reviews where appropriate, to take into account new forms of exploitation; underpinned by a reversion right where appropriate.
Ownership: Creators should be appropriately credited (including metadata) for all uses of their work; the integrity of artistic works should be respected, and the Ownership of moral rights should be unwaivable.
Reasonable: All contracts should be subject to a general test of Reasonableness including a list of defined clauses which are automatically deemed to be void, and a general safeguarding provision that any provision contrary to the requirement of good faith, which causes a significant imbalance in the parties’ rights and obligations arising under the contract to the detriment of the author, shall be regarded as unfair
Any regulatory and non-regulatory tools should be developed in a manner which is fully consistent with the UK’s current copyright framework, and must align with the existing regime for determining copyright infringement and penalties for infringement, ensuring that the ability to obtain injunctions, and the availability of damages and other remedies remain for copyright holders.
AI systems which do not abide by the UK’s existing legislative system should be prohibited.
Any regulation (whether statutory or non-statutory) should include obligations for independent/third-party audits and certification to show principles are being met; technical tools and checklists to assess AI products against regulations and principles; and a clear classification of AI risk and suggested risk assessments.
Government should create statutory obligations on AI suppliers for transparency and labelling – obligations that should apply equally to users of AI systems or their outputs. AI system developers must provide labelling solutions to any AI outputs, as watermarks in the metadata, and in a way that is easily understandable to an untrained user.
We believe that additional, low-cost mechanisms for holding AI developers accountable for copyright infringement would help build consumer trust. This includes establishing clear lines of responsibility and accountability for AI decision-making, and establishing processes for addressing grievances and complaints related to copyright infringement by AI systems such as LLMs. This could be achieved for example through the creation of dedicated AI reporting systems, adjudicators or ombudsmen.
In addition to this, there is a need for more accessible, low-cost dispute resolution mechanisms. In particular, we do not believe that current routes to redress for AI-related harms are adequate. We see clear challenges in implementation which need to be addressed in order to avoid further eroding the UK’s copyright regime and causing irreparable harm to our creative industries.
CMSC suggests the ‘Government should establish a discrete AI regulation co-ordination unit within Whitehall to ensure coherent working and enable robust stakeholder engagement. This unit should publish regular reports to enable Parliament to fully consider the progress of the regime’s introduction and implementation.’
a) How would such options work in practice and what are the barriers to implementing them?
We believe we have set out clearly how this work would work in practice, however the main barriers to implementation are the Government’s reluctance and the lobbying power it allows AI developers.
b) At what stage of the AI life cycle will interventions be most effective?
Sadly, the Government has been very slow in recognising the need for intervention on the development and learning stage of the AI process. This is not for want of creator led groups warning of the dangers of text and data mining that has been happening unchallenged for some decades. The sooner intervention is made, and legal frameworks and reporting tools are in place to ensure developers are legal and compliant, the better. There must be a clear legal obligation to developers of LLMs and AI systems from the offset that the creation and development of their systems must always remain within the current IP law and legal obligations.
We are calling for existing protections to be acted upon and openly publicised. It is not for LLM and AI developers to give creators or Government a definition of how they would like legal intellectual property protections to be interpreted and administered for their benefit.
Before even the development stage is reached there should be a registry of intent for AI and LLM developers, setting out the type of system they intend to develop - and how, where and with what protections for rightsholders it intends to access input data - to ensure safeguards so that creators’ work is not ingested or accessed without prior explicit permission. Permission must be obtained before ingestion and not after or through blanket agreements. Individual creators must remain in control of their work; it must be assumed that permission is not given until it expressly is.
c) How can the risk of unintended consequences be addressed?
By implementing the recommendations outlined above, and by being clear from the very offset that creators have the unwavering right to maintain copyright in their creations and that these rights are not simply given away because someone else wants to use their creativity to provide fuel to their business, and profit for themselves. The Government must make clear in strong and in no uncertain terms, that it will uphold the rights of creators and that those who transgress will incur the greatest of sanctions.
The law must be the law and we must expect our policy makers to not only champion it, but to ensure that it is being followed, implemented, and upheld.
The huge danger to jobs and career prospects must be acknowledged as a legitimate concern which needs much thoughtful discussion. There should be an independent cross sector impact assessment of the impact of these new LLMs and AI systems on employability prospects, impact on pay and any reduction of access to careers, especially for those who already experience challenges and preconceptions when aspiring to careers in the creative, tech and wider economy. Recommendations from this impact assessment must be acted upon.
The Government must implement the Beijing Treaty, which is also a recommendation of the CMSC report into AI, to provide an improved moral rights framework. As part of its implementation, we would like the Government to make clear that performers’ (economic) right of reproduction and moral right of integrity (to object against the derogatory treatment of the performances) apply to synthetisation.
AI must be continually monitored, especially when it is used for making decisions that affect individuals, for it will not understand all the nuances that there are and will make unethical decisions based on bad data and misinterpretation. Individuals must have the right to challenge reviews or assessments fulfilled by AI systems. People must be confident that decisions have been made and assessed to human standards. AI systems have constantly been shown to exercise institutional bias - and because they are basically copying machines working on statistical probabilities, they leave no possibility to deal with the innovative, creative or exceptional, or indeed to compensate for gaps in their process.
The data protection authorities of several countries have raised serious concerns over the misuse of personal data by such systems in breach of privacy laws, including in Spain, Italy and the USA. The European Data Protection Board has launched a taskforce dedicated to monitoring the compliance of developers and their systems with data privacy laws. EU civil servants are prohibited from using AI systems in their work for the analysis of any data that is not in the public domain. The British Copyright Council has noted that there are also serious concerns about ‘data laundering’ in the datasets utilised by AI developers, especially in relation to open-source foundational models – and we refer you to their written evidence on the subject.
6. How does the UK’s approach compare with that of other jurisdictions, notably the EU, US and China?
The UK’s framework for UK copyright legislation has up to now often been heralded as a ‘gold standard’, being flexible enough to support innovation and drive the growth of incredible creative content to the extent that we are significant global net exporters. Investment and economic growth in the UK creative industries is arguably a direct result of this framework, which has a long history in recognising and supporting our creators. Even our framework for ‘fair dealing’ is a recognition of the skill, labour and judgement afforded to UK copyright works - AOP position paper on AI 2023
The UK’s international influence cannot be ignored. Systems should also work internationally for overseas creators and for overseas licensing. Economic and moral rights must be protected. However, the UK Government’s ‘pro innovative approach’ is in danger of backing the needs and wants of LLM and AI developers at the cost of others, which seems short sighted.
The EU has set out its intention to regulate to address the most potentially harmful aspects of AI. The European AI Strategy aims at making the EU a world-class hub for AI and ensuring that AI is ‘human-centric and trustworthy’. Such an objective translates into the European approach to excellence and trust through concrete rules and actions.
Their regulatory framework defines four levels of risk in AI with different rules depending on the risk a system might have on fundamental rights: Unacceptable risk; High risk; Limited risk; and Minimal or no risk.
This too is currently in danger of being diluted by the lobbying of AI developers. The recently passed draft AI Act should impose conditions on the use of facial recognition software, while requiring makers of AI systems like the ChatGPT chatbot to disclose more about the data used to create their programs.
The EU had set out a proposed requirement for providers of foundation models to compile and publish a summary of copyright-protected material used for their training data, and to train their models to safeguard them from generating law-breaking content.
The EU already has regulations that apply to AI. Its GDPR (General Data Protection Regulation) legislation has put restrictions on the collection of personally identifying data since 2018, for instance. And EU citizens already had the right, through the GDPR, to ‘meaningful information’ about the logic involved in automated decisions (sometimes referred to as the right to explanation) as well as a right to opt out. In practice, however, these rights are currently of limited use: only a few processes are fully automated.
President Biden signed an order forbidding algorithms from discriminating on grounds of race. Interestingly, when Biden made a recent speech, he was joined at the White House by Microsoft president Brad Smith, Google president Kent Walker, Anthropic CEO Dario Amodei, Inflection AI CEO Mustafa Suleyman, Meta president Nick Clegg, OpenAI president Greg Brockman and Amazon Web Services CEO Adam Selipsky, but none of those whose work and data is the foundation of these LMM/AI systems.
a) To what extent does wider strategic international competition affect the way large language models should be regulated?
Unknown
b) What is the likelihood of regulatory divergence? What would be its consequences?
Extremely high. It is almost impossible to regulate without international cooperation controlling the implementation and use of AI systems, and how they access copyright protected work for data and material. A worldwide AI implementation free for all will be devastating for economies and society. We must ensure that we stand by our moral and legal obligations and champion them as the ‘gold standard’ system for others to copy and implement.
It is impossible to regulate internationally in just one region, but we can at least get ours right.
August 2023
Association of Illustrators (AOI)
Association of Photographers (AOP)
Association of Authors’ Agents (AAA)
Authors' Licensing and Collecting Society (ALCS)
British Association of Picture Libraries and Agencies (BAPLA)
Chartered Institute of Journalists (CIJ)
DACS (Design and Artists Copyright Society)
Directors UK
Equity
Independent Society of Musicians (ISM)
Ivors Academy
Musicians Union (MU)
National Union of Journalists (NUJ)
Picture Industry Collecting Society for Effective Licensing (PICSEL)
Royal Photographic Society (RPS)
Society of Artists’ Agents (SAA)
Society of Authors (SOA)
The Royal Society of Literature (RSL)
Writers Guild of Great Britain (WGGB)
20
[1] https://www.gov.uk/government/statistics/dcms-economic-estimates-2019-gross-value-added/dcms-economic-estimates-2019-gross-value-added-technical-and-quality-assurance-report
[2] https://www.gov.uk/government/statistics/dcms-sector-economic-estimates-employment-oct-2019-sep-2020 approx. 707,000