The UK’s attempt to deal with generative AI, training data and copyright law has taken yet another turn. On 6 February 2024, in its response to the AI White Paper consultation, the UK government announced that it will drop its plans for a code of practice on copyright and AI – a work it has been carrying out for less than a year since its announcement in May 2023.
Background
The UK’s policy-making efforts in the field of AI and copyright date back to its 2021-2022 public consultation, with which it hoped to collect enough evidence to identify the best way forward to deal with the challenges AI poses, particularly with regards to (i) the copyright input issues, namely text and data mining as per section 29A of the Copyright Designs and Patents Act 1988; (ii) the copyright output issues, namely the computer-generated works as per section 9(3) of the Copyright Designs and Patents Act 1988; and (iii) patent inventorship issues (you can read detailed accounts of these challenges, as well as the precursor consultations and government initiatives here, here and here).
The UKIPO’s first bold attempt to deal with the input issues was its decision to reform its current text and data mining exception by opening it up to all commercial uses while maintaining its mandatory nature, namely leaving rightholders no possibility to contract out. Following a backlash from the influential stakeholders from the music and publishing industries, the UK took its first sharp turn by dropping the legislative reform approach and announcing that it will instead work on a code of practice developed in consultation with a diverse group of experts in the field (see more here).
The working group
In June 2023, the UKIPO started hosting meetings by a technical working group. The mandate of the group was to identify, develop and codify good practice on the use of copyright, performance and database material in relation to AI, including data mining. This is in line with the UK’s attempts to resolve complicated AI issues through industry self-regulation, rather than top-down legislative intervention. In the words of the UKIPO: “this is an issue that industry can, and should, seek to fix itself, although it does not rule out the possibility of legislative action in this area if industry does not take adequate steps to improve the situation.” The working group comprised a notable selection of different representatives from the various ‘stakeholders’ interested in this discussion – from OpenAI and StabilityAI, through to Getty Images, Microsoft, IBM, BBC, all the way to UK Music, Copyright Licensing Agency, the British Copyright Council and many others. For a full list, see here. The selection of these members was done by the UKIPO in consultation with industry bodies and others, with the aim of providing “balanced representation and expertise”. The UKIPO chaired the meetings, while representatives from the Department for Culture, Media and Sport, the Office for Artificial Intelligence and the Competition and Markets Authority attended as observers. The aims and objectives of the working group were ambitious:
- identifying any creator concerns relating to the use of copyright works, performances and databases by AI systems and users;
- outlining ways in which any concerns can be addressed;
- identifying any barriers to the access to copyright works, performances and databases by AI systems and users, including for the purposes of text and data mining;
- outlining options to address any barriers;
- setting out commitments and expectations in relation to AI firms’ use of protected material and the right holders who own protected material
On paper,the UKIPO’s working group approach sounds ideal – bringing together different stakeholders to devise a working solution to this very complex issue. However, there were several flaws in the way that the UKIPO constituted and operated the working group.
Transparency and inclusivity
Parties at these meetings (generally) do not get on well. Copyright law has always been an emotional topic. Fuelled by AI concerns, it becomes borderline explosive. These representatives come from opposing camps by default, so a lot of groundwork needs to be done to ensure this is not merely another lobbying arena. That said, I am not opposed to the idea of bringing ‘enemies’ together – the initiative as such is perhaps the most sensible way of treating this complex area.
The bigger problem I see in this policy-making exercise is transparency and inclusivity of the process itself. Such issues emerge in two respects.
First, when selecting the members of the group. While there was a public call for responses to the consultation run by the UKIPO in 2021-2022, nowhere was there a call for participation in the working group. Indeed, Stability AI and Getty did not even reply to the 2021 consultation, but were invited to sit on the working group. Of course, their ongoing litigation makes them valid and attractive candidates to join these talks. However, being engaged in litigation cannot, by itself, be sufficient selection criteria for participation in such an important policy-making task. It is understandable that the group working on this project has to be manageable both in terms of numbers and expertise. As the Terms of Reference (ToR) point out, the group membership was determined by the UKIPO in consultation with industry bodies and others, with the aim of providing “balanced representation and expertise”. That said, to an outsider, this is far from inclusive. Some important stakeholders may have easily been forgotten. For example, it is not clear whether any of the members was also a representative of or an advocate for specific types of licensing modes, such as Creative Commons for example. In addition, while UK Research and Innovation was present and is said to represent the views of academics in the field, little is known about how these views were drawn together.
A second transparency-related flaw is the absence of public materials documenting the process. As the ToR mentions, the UKIPO intends to publish “some materials relating to the working group”. The only available documentation around the workings of the group is the ToR and the list of the members. Furthermore, the ToR mentions that the discussions were ‘Chatham House’ style, i.e. the identity of persons making contributions and offering comments was not to be revealed to others outside the meeting. Indeed, aside from the ToR and the member list, the general public got access to nothing else – none of the meetings’ minutes, agendas or discussed questions were made available anywhere. However, records may be subject to Freedom of Information requests.
Comment
This was a commendable exercise in the sense of bringing very different perspectives to the same room, but what it lacked at all stages of development (from its formation, through its meeting(s?), to the dissolution) was transparency and inclusivity. Yes, we can all imagine the general lines along which the arguments of various parties would have gone. For example, it would not have been surprising to hear rightholders pushing for compulsory licensing schemes with opt-outs, while the AI industry seeking low and potentially flat rates for training sets, or even arguing that training AI with copyright material is not considered reproduction at all. These are the standard narratives played out in various such occasions – see the public consultation responses.
However, in times when the policy makers globally are actively trying to find an effective solution to please all parties by tailoring the approach to the complex genAI technologies (text-to-text, text-to-image, text-to-music, text-to-code, text-to-video, etc), transparency in sharing whether there is any middle ground between all these parties would have been surely appreciated widely, and not just by academics. This is more the case considering that the ToR stresses that since “this is a technical working group, participants are not being asked to endorse outputs personally, or on behalf of the businesses or organisations they represent or are affiliated to”.
Wide industry and public consultations of this kind have also been carried out by the US Copyright Office since the Spring of 2023. It is fair to say that the US is a much bigger market, with bigger players, so the pressure on being transparent and the numbers of participants and involved parties is, by default, higher. Yet, not only did the US Copyright Office hold four listening sessions with different industry representatives (literary works, visual arts, audiovisual works, music and sound recordings), but these sessions were livestreamed (followed live by nearly 4000 viewers) and recorded. In the Spring of 2023, the Office also held webinars on registration guidance and international perspectives. It is, without doubt, a very daunting and burdensome task as everyone wants to have a say in these discussions (the US public consultation received more than 10,000 replies). Processing this amount of data is a heavy task for policy-makers, but it is equally good news. The volume of replies is evidence that, when asked about copyright law, not only do the experts in the field have an opinion on the matter, but so do the wider public.
I believe that a code of practice is not merely wishful thinking. Though, introducing and insisting on transparency and inclusivity safeguards is, without a doubt, time-consuming and very expensive. However, the rewards from such a process justify and outweigh these costs. To repeat, copyright law is an emotional topic. So, listening to diverse views is not only therapeutic, it also leads to well-rounded and balanced legislation (hard or soft).
The author would like to thank João Pedro Quintais, Thomas Margoni and Dheemanth Vangimalla for comments on earlier drafts of this post.
________________________
To make sure you do not miss out on regular updates from the Kluwer Copyright Blog, please subscribe here.
