University of Hull, UK

A dedicated global platform on AI governance can contribute to the development of broad principles and common commitments and eventually international laws or agreements on AI governance, as exist in other areas. However, to be effective, such a governance structure would of course need to be integrated into other kinds of governance and regulation (not just over AI development and use, its relation to broader digital economies, but also in relation to IP, product liability and so on) at the regional and national levels. Managing this interface with other kinds of regulation will be key to the success of the platform in the long term, since the impacts and ethical considerations differ very widely across sectors and applications of the technology. To maximise impact, explicit acknowledgement of the differences between international governance of a technology whose most powerful applications are owned by a small number of companies, and broader international governance platforms focused on problems such as climate change and areas of law such as intellectual property or trade should inform the debate on metrics for success. Monitoring and evaluation of key performance indicators such as accuracy or bias, for example, has to be independent of metrics developed and implemented by large AI companies operating under their own commercial imperatives.

• AI capacity-building
• Protection and promotion of human rights
• Social, economic, ethical, cultural, linguistic and technical implications of AI

In regard to AI capacity building, we need more critical digital literacy materials for people who have a real choice about data governance at the beginning of their digital journey. Most digital literacy materials focus on getting people online first, and then worrying about what to share. In regard to social, cultural and other impacts of AI, we believe that better monitoring is needed of both short and long term effects, not just on the economics of creative work (where some work is already being done), but also on the cultural impacts of biased and hallucinatory LLMs, whether by replacing human interactions in some cases and answering human queries about human culture. Human rights law (the right to participate in culture) can play a role both in regard to discussions on safeguarding cultural interactions between humans (and preventing harmful AI interactions) and in addressing the economic impact of AI development and deployment on artists.

Metaphors play a large role in regulatory discussions about complex, abstract ideas. Industry lobbyists have been working hard to set the stage for regulatory dialogue by shaping the metaphorical landscape in their favour. Analysis of the utility and biases present in these metaphors is essential if the global dialogue is to represent all stakeholders fairly. For example, our work has critiqued the metaphor of the 'balance' or trade-off between innovation and protection often invoked in AI policy. We argue that the benefits of innovation have accrued mainly to large technology companies while the costs of loss of rights protection have been borne by artists and cultural communities (where automated data scraping is not sanctioned by artists themselves). Where artists want to use AI tools to innovate, they should not have to face the choice of losing existing rights to benefit from their data. Policies supporting artists might be better represented as a chair, supported by four legs of empowerment, rights protection, innovation and access.

In regard to AI capacity building, our research team are engaged in a British Academy-funded research project developing capacity-building materials for cultural heritage artists in Burkina Faso. Creative uses of AI tools can help to inspire and benefit artists in Africa, as elsewhere, but access should not be at the price of loss of control over cultural data. While there have been efforts to train models on African data, per capita use of free AI platforms in developing countries seems to outstrip many other countries according to the ChatGPT statistics. Yet there is a gap in providing critical digital literacy materials to people who have a real choice about data governance at the beginning of their digital journey. Most digital literacy materials focus on getting online first, and then worrying about what to share. We are thus developing some capacity-building tools, working with local communities to calibrate them accordingly. In regard to social, cultural and other impacts of AI, we believe that better monitoring is needed of both short and long term effects, not just on the economics of creative work (where some work is already being done), but also on the cultural impacts of biased and hallucinatory LLMs, whether by replacing human interactions in some cases and answering human queries about human culture. Human rights law (the right to participate in culture) can play a role both in regard to discussions on safeguarding cultural interactions between humans (and preventing harmful AI interactions) and in addressing the economic impact of AI development and deployment on artists. The lack of appropriate regulation, enforcement and training to control and manage data scraping and AI-enabled uses of art and cultural information once data is uploaded online is causing many artists to stop sharing, lose income and attribution to their works. The failure of copyright law in fostering artist benefit is a longer-term problem simply exacerbated by the way in which AI models have been developed. Artists' rights and the rights to cultural participation should be more closely linked in regulatory discussions. African continental and national AI policies are dominated by discussions of healthcare, agriculture and other benefit, and generally pay little attention to culture; cultural policies pay little attention to AI. However, both sectors are intended to contribute to development, and there may be negative as well as positive interactions between them. When it is mentioned, AI policies tend to talk more about culture in terms of making AI tools relevant and usable, than about protecting cultural interaction and participation, creator access, benefit, and attribution without loss of income. Since AI tools will have a big impact on cultural industries as well as cultural practice, and their use will be informed by cultural factors, policy on AI and the culture sector is an important part of governance dialogue. This is fundamentally a different kind of governance discussion than in sectors such as healthcare and agriculture, but is perhaps most aligned to discussions in the education sector.

At present, it is very difficult for stakeholders to engage on a level platform, especially in an environment of regulatory capture and commercial and geo-political competition. Deeper analysis of these challenges (beyond the North-South divide) and changes in commercial contexts (for example the shift away from venture capital-funded initiatives based on 'free' data and free access) need to be inform strategies for creating a level playing field for discussions.

The AI Dialogue enables dialogue between a wider range of stakeholders which can be coupled with dialogues in sector-specific initiatives working on AI policy and African creative sector actors such as Electric South in South Africa; a diversity of sector-specific issues and impacts have to be considered in developing general guidance and approaches.

The opinions of some stakeholders (especially in regard to cultural impacts of AI) will not be collected through processes such as this one, which is aimed specifically at government, commercial and institutional stakeholders. Targeted pilot projects in advance of a dialogue on a specific issue with groups such as artists or cultural practitioners might be used alongside broader data collection activities (for example analyses of impacts and uses of AI tools) to help clarify specific issues relevant in policy discussions. Pilot projects might also be useful to target insights from stakeholders with low digital access.

It is important to ensure better participation from non-English speaking communities, especially in the Global South, including from people who are not yet frequent users of digital / AI systems. Early-access stakeholders, who in our experience are interested in these issues before they are fully connected, may be involved through local projects or focus groups through research projects such as ours. Linking dialogue to access to tools and critical digital literacy could make it more meaningful for participants.

Strategies such as the vTaiwan approach for public engagement could be used to foster engagement and dialogue with a wide range of stakeholders: the vTaiwan community used the pol.is tool to collect public opinions, in-person discussions for divisive topics, and Talk to The City to analyse transcriptions. Such strategies work well in some contexts, but would need adapting to contexts with low internet access, as discussed above.

In the culture sector, a number of relevant initiatives act as platforms for discussion on policy questions relating to culture and AI, including WIPO's AI Conversations, UNESCO's AI, culture and IP subgroup which proposes an online Repository and Resource Hub, UNESCO 2003 Convention's nascent AI initiative and many artist and indigenous organizations. Networks such as AI4LAM (AI for Libraries, Archives and Museums) and HAIL (Heritage, AI and Law) can act as conduits for discussion. There are some strategies for signalling appropriate AI uses tailored to cultural content including Creative Commons Signals, the Local Contexts project on Indigenous data governance. However, opt-out and signalling strategies depend on the compliance and ethics of the data scraping entities, which is not guaranteed. Enforceable measures that target commercial incentives, including platform and product liability, need to be used alongside ethics-based strategies.