Tech Policy Conversation #10: The AI and intellectual property debate
Taking stock of the thorny intersection of AI and intellectual property, the policy-action industry is calling for, and what’s likely to happen.
Around the world, policymakers are trying to balance the interests of rights holders and AI developers in pursuit of a thriving AI ecosystem, while also supporting fair remuneration for creative content. This is difficult, not least because of the volume of high-quality data needed to train large language models, and the challenge in tracking the provenance of protected content in the process.
Meanwhile, key players in the tech and media industries have started to take matters into their own hands. Some have struck licensing deals (e.g. Financial Times and OpenAI), whilst others are taking AI developers to court (e.g. New York Times versus OpenAI).
In this month’s edition we take stock of the current policy debate, and look ahead at what’s likely to happen.
In this issue:
The Milltown View: AI and intellectual property: what’s up and what’s next?
Key Developments: September 2024
Théâtre D'opéra Spatial - an image created by Jason Michael Allen with the generative AI tool Midjourney. The image won the 2022 Colorado State Fair's annual fine art competition in the digital art category, one of the first images made using artificial intelligence (AI) to win such a prize.
THE MILLTOWN VIEW: AI AND INTELLECTUAL PROPERTY - WHAT’S UP AND WHAT’S NEXT?
Milltown Partners has asked three organisations for their view on the current state-of-play: a rights holder representative organisation, an AI developer, and a company operating in between the two. We quizzed them on what is working well, what isn’t working so well, and what (if any) policy action is needed.
What’s working well
“Our copyright framework in the UK allows those with copyright in their work to license it for a fee. It has worked well in a multitude of uses, from publishing to broadcasting and beyond. There’s no reason why licensing can’t work in an AI context”, says Reema Selhi, Head of Policy at The Design and Artists Copyright Society (DACS). “Licensing is a way of bringing fair remuneration and value for content. Recent deals struck by news publishers show that this is possible”. Reema also sees encouraging signs of artists being early adopters of the technology. “Work like the upcoming exhibition The Call, by Holly Herndon and Mat Dryhurst, showcases the output of AI trained on a variety of choral voices, whilst putting forward a data trust governance structure to allow those involved in the project to permit ongoing uses of their voice.” Adobe, a leader in software and AI, bridges the AI and copyright conundrum as follows: “Firefly, Adobe’s family of generative AI models, is trained only on openly licensed and public domain content available in Adobe Stock, or where copyright has expired. For creatives, AI can advance creative expression, supercharge exploration and productivity, and lower the barrier to work, allowing them to focus their attention on new ideas and opportunities”, says Stefanie Valdés-Scott, Head of Policy and Government Relations EMEA at Adobe.
What isn’t working so well
Despite these promising industry developments, challenges remain. AI developers are scrambling to get their hands on sufficient high-quality data to train models, and rights holders are struggling to be fairly remunerated for the use of their content in AI training. “Limiting access to data and protected content risks leading to poorly performing or biased models and, ultimately, less benefit for users”, a representative of a key AI developer said. “It is not desirable to train today’s leading AI models without using copyrighted materials”. Training models using copyrighted material is one thing, doing that with the creator’s consent is another. “Earlier this year a list of 16,000 artists whose works were used in AI training was leaked to the press. Even though many artists can tell their works were used for training simply by generating an output that’s almost identical to their work, there has been no real transparency over what material is used and how extensively”, says DACS. Adobe shares this concern: “One of the biggest concerns we have heard from our creative community has been around the misappropriation of image, likeness, voice and artistic style. Rights of publicity laws exist but they don’t help empower creators”.
What policy action is needed
So where does that leave us in policy terms? The AI developer says: “If you want to develop large language models, you want to operate in a jurisdiction with flexible copyright laws. In the UK, expanding the current text and data mining exception would have been a good step to signal that the UK is the place to build generative AI tools.” Adobe believes that there should be a remedy (called an ‘Anti-Impersonation Right’) for artists to tackle the misuse of AI when people generate content with AI tools that result in deceptive practices in the marketplace and economic harm. “Moreover, as well as providing transparency about the artefacts produced using generative AI, Content Credentials, that act like a nutrition label for digital content, allow creators to attach a ‘Do Not Train’ tag to the metadata of their work to signal they do not want their work to be used for AI training. Adobe would like to see the UK Government introduce legislation mandating that social media platforms ‘do not strip’ content credentials on their platform.” DACS welcomes steps taken in the EU’s AI Act. “The Act requires providers of certain AI models to keep an up-to-date document that includes information about input data, as well as transparency in terms of outputs, for example labelling of deep fakes. The UK should implement similar measures, which should also have the effect of enabling licensing”. Finally, creators need the necessary capabilities to experiment with AI, and to set the terms and conditions for their content’s use in AI training. “From those surveyed by DACS, a third saw a lack of training in AI as a barrier to using it in their work, and 9 in 10 artists had not received any training or education in AI tools. The skills gap is particularly felt by those over 10 years’ into their practice, suggesting a need for further education and training programmes that can close the gap.”
The Milltown View
As industry finds ways to tackle the current AI and copyright debate, policymakers are at pains to find solutions that serve the needs of both content creators and AI developers. In the UK, the Department for Science, Innovation and Technology and the Intellectual Property Office (IPO) are expected to pick up the policy from where they left off following the stranded IPO Code of Practice on AI (with likely legislative changes as a result). In the EU, much remains to be clarified on the EU AI Act’s transparency obligations’ implementation, and any wider plans to revisit the Copyright Directive with generative AI in mind. Is it possible to have a thriving AI ecosystem and a flourishing creative industries sector? At Milltown, we do not think this has to be a binary choice. Choosing the right policy direction is inherently cultural, and much can be achieved by policymakers acting as a bridge between AI developers’ and rights holders’ interests - recognising that both camps are not mutually exclusive: many creators operate across the two. We will be tracking developments closely, and would love to hear your take on this debate.
KEY DEVELOPMENTS
EU: Mario Draghi publishes long-awaited report on EU competitiveness (link)
Former Italian Prime Minister, and Director of the European Central Bank (ECB) Mario Draghi released a long-awaited report on the state of European competitiveness on 9 September 2024.
This report draws harsh conclusions on the faltering state of the EU economy. The report calls for heavy investments in AI, research, and cloud, a cautious approach to new legislation, and further help for SMEs and smaller businesses to develop and grow. Draghi is also critical of the AI Act and the GDPR, which he calls inconsistent and fragmented across Member States. To address these challenges, Draghi calls for a hefty €800 billion investment in research and development, more telecoms mergers, and a reform of the EU institutions to make the legislative process faster. Draghi argues that the EU’s weakened productivity is mainly due to fragmentation, regulatory burden, and a failure to adapt to new technologies. The report was commissioned by current European Commission President Ursula Von der Leyen, and is expected to shape her upcoming policies.
EU: Von der Leyen II announced (link)
On 17 September, European Commission President Ursula von der Leyen presented the list of the 27 Commissioners and their respective portfolios. This College of Commissioners includes six Executive Vice-Presidents, who are expected to wield greater authority compared to Vice-Presidents from previous mandates. Candidate Commissioners will present their commitments during European Parliament hearings, and MEPs will ask for clarifications ahead of a vote of confidence (21 October is the expected date for the first hearings). The European Commission's new College of Commissioners is planning to officially start on 1 December 2024, and their mandate will last until 30 October 2029.
After the dramatic resignation by French Commissioner Thierry Breton, who had been one of the key figures in EU tech policy, he was swiftly and de facto replaced by Finnish Henna Virkkunen in the new Commission. While Breton was the commissioner for the internal market, Virkkunen will work with the same team but operate under the upgraded title of Executive Vice President for tech sovereignty, security and democracy, meaning she reports directly to von der Leyen. Margrethe Vestager, one of the world’s most powerful antitrust figures after 10 years in the post, has been replaced by Spanish Teresa Ribera - also an Executive Vice president. Get in touch if you would like to receive Milltown’s more detailed summary.
UK: Election of Select Committee Chairs (link)
On 11 September, the Speaker of the UK House of Commons announced the results of the election for select committee chairs. Key appointments include:
Business and Trade Committee: Liam Byrne (Labour), re-elected to the role
Science, Innovation & Technology Committee: Chi Onwurah (Labour)
Culture, Media & Sport Committee: Dame Caroline Dinenage (Conservative), re-elected unopposed
Treasury Committee: Dame Meg Hillier (Labour)
With committees responsible for scrutinising government departments and holding ministers accountable, these appointments give us a good indication of kinds of scrutiny specific legislation may receive in Parliament. For example, Chi Onwurah has been a key advocate for strengthened data rights for workers in the Data Protection and Digital Information Bill, and is likely to continue to do so when it is reintroduced later this year. Get in touch for Milltown’s more detailed analysis on key appointments.
UK: Party conference season in full swing
With Parliament having entered Conference recess on 12 September, party conference season is officially on. The Lib Dems, having secured their best election performance since the party was established in 1988, enjoyed a jubilant conference in Brighton setting out its ambition to become the ‘main opposition party’ in the UK. Labour’s Party Conference in Liverpool saw a programme filled with panels on the potential of AI in public services, whilst the overall mood was perceived to be understated considering their recent election victory (not least because of the pouring rain on day one). The Conservative Party conference starts on 29 September in Birmingham.
As the main temperature check of a party’s overall mood and strategy, party conferences remain key events in the political calendar. Milltown has attended the Labour Conference, and will be attending Conservative Party Conference - get in touch if you’d like to find out more.