Brunel University of London

• The Dialogue would be a success if it ensures inclusive and diverse participation
• addresses environmental impacts of AI (both AI for Green and Green AI perspectives)
• is grounded in human rights
• and produces a detailed roadmap for the Dialogue beyond July 2026, adopting clear terminological and technological definitions, scope, goals, actionable steps, and measurable indicators for progress between 2026 and 2027. AI should not be reduced only to Generative AI, but interpreted broadly to also include responsibly trained models used for non-profit heritage management and science. These uses of AI include heritage conservation, preservation, reconstruction, and analysis of large and fragmented datasets, allowing researchers to uncover patterns and connect information. Narrow definitions of AI therefore risk excluding critical perspectives which require specific governance approaches (for example as regards copyright law). Discussions should be organised into thematic sub-clusters, grounded in practical case studies, and anchored in human rights. This allows the required depth and diversity of perspectives, so that both risks and opportunities can be effectively discussed. Attribution of responsibilities should be clearly defined to ensure the roadmap is followed and goals are achieved. The UN Independent International Scientific Panel on AI could act as stewards of progress, ensuring evidence is recorded and translated into actionable policy. Diversity of participation must be treated as a matter of priority, with special attention should be given to underrepresented stakeholders and perspectives, including Indigenous and marginalised communities, and environmental concerns. Hybrid, asynchronous and multilingual participation should be allowed through a dedicated online platform (see recommendation below in answer to question 17). Strategies and plans for Dialogue continuation and roadmap execution should be agreed by the end of the first Global Dialogue event in July 2026.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

I believe all the thematic areas above are essential and interdependent, but I selected the options above as I believe they are foundational for the development of the other themes. AI capacity building, interoperability and openness are also fundamental, but they need to be anchored by safety, security, trustworthiness, socio-cultural impacts, human rights, transparency and human oversight. For example, I advocate for responsible openness, which in my view means that open-source and open-data discussions must be grounded on human rights and transparency requirements. I also believe that meaningfully inclusive and diverse stakeholder participation in policy and decision-making is essential for this Dialogue. This is mandated by human rights frameworks such as cultural rights, where the right to participate in cultural life (art 15 of the UN Covenant on Economic Social and Cultural Rights 1966) has been interpreted in UN General Comment no. 21 to include "the right to take part in the development of the community to which a person belongs, and in the definition, elaboration and implementation of policies and decisions that have an impact on the exercise of a person's cultural rights.". Therefore, anchoring the Dialogue in human rights is foundationally critical to ensure the relevant perspectives, implications and impacts are fully addressed. Furthermore, under the theme "social, economic... implications of AI" should fall a fundamental interrogation explored by Joana Pedroso's in the 2026 report "The AI Curse: The Hidden Public Costs of the "AI Magic"": who benefits from, and who ultimately pays the AI bill? Pedroso explains that, while AI's productive capacity is undeniable, AI unsettles traditional trade-offs: "jobs created may be limited, whereas the space, the resources, and demand on public infrastructures can be immense."

I believe environmental sustainability, which must be treated as a core cross-cutting dimension, is not being sufficiently captured. It could in my opinion be explicitly added to one of the themes i.e. "Social, economic, ethical, cultural, linguistic, environmental and technical implications of AI". Environmental discussions should encompass both "AI for Green" and "Green AI" perspectives. "AI for Green" includes the potential for AI to be used to research and support the environment, climate and sustainability objectives - for example, AI is being used in climate change mitigation, wildlife conservation, plant identification, environmental monitoring (such as air quality and natural disasters), and management of resources (such as water) and waste. "Green AI" relates to the environmental costs of AI systems - including energy use, carbon emissions, water consumption, land use and unsustainable mining of critical minerals, and the related impact in local communities, including Indigenous groups. Adopting a decolonial approach to AI governance is a critical cross-cutting theme that should also underpin the UN AI dialogue. Lastly, while I am interpreting Joana Pedroso's point above on the hidden public costs of AI (under answer to question 10) under the theme "social, economic... implications of AI", I believe this is a cross-cutting theme that should be more explicitly included in the Dialogue.

The use of Internet content for commercial AI training is concerning open access/open culture movements. Wired reported (April 2026) that the Internet Archive's Wayback Machine is threatened as publishers are blocking its bots, fearing that the archived data is being used by AI firms to train competing AI models. GLAM-E Lab reported (June 2025) heritage sector's concerns over their collections and servers straining/breaking due to bots scraping collections data to build AI models training datasets, overwhelming the systems designed to keep those collections online, and creating "an environment of unsustainably escalating costs for providing online access to collections". This governance gap regarding commercial AI training on Internet content must be bridged urgently, to protect our cultural commons. Opt-out models (e.g. art 4, EU Digital Single Market Directive) are inefficient, inappropriate and hard to implement. Responsible openness approaches must be adopted, preserving heritage organisations' open ethos, while mindful of not disclosing confidential, sensitive or inappropriate/unethical data (sacred imagery, geolocations etc.). Another gap pertains AI for non-profit research and heritage management. The UK Text and Data Mining (TDM) exception for non-commercial research (s 29A, CDPA) is not fit for purpose for non-profit AI research in digital humanities/heritage science (e.g. not addressing public-private partnerships or data transfers between project partners). Exceptions for heritage management do not sufficiently accommodate AI (e.g. s 42 CDPA allows making a copy of an institution's permanent collection item to preserve/replace it, leaving out e.g. preservation of heritage sites, and AI-based preservation/reconstruction work requiring other materials). Another governance gap concerns environmental sustainability; the rapid expansion of AI/data centres impacts e.g. energy, water, carbon emissions and critical minerals. Lastly, Indigenous data governance/traditional knowledge are rarely included in IP/AI governance discussions. All these gaps have human rights implications and must be urgently addressed.

• A key contribution of the AI Dialogue would be to strengthen alignment across existing international repositories and governance efforts and regional initiatives. Rather than duplicating existing work, the Dialogue can build on and connect existing initiatives. It can also have an important agenda setting role, for example advancing environmental impacts of AI as a crucial topic in the international policy/cooperation agenda. The AI Dialogue could have a meaningful role in connecting and building on current international initiatives, such as the World Intellectual Property Organisation (WIPO) on exceptions and limitations for cultural heritage and the WIPO AI Conversation
• the UN Special Rapporteur in the field of Cultural Rights work on AI and creativity and heritage digitalisation
• the Expert Mechanism on the Right to Development work on AI and cultural rights
• UNESCO's Global AI Ethics and Governance Observatory, including the AI, culture and IP subgroup
• and the UN Environment Assembly UNEA-7 (2025), which focused on the AI/environment intersection. These initiatives are crucial but fragmented, and in need of an impactful convergence space, which the UN Dialogue could offer. Cooperation with initiatives such as Creative Commons, Communia and Local Contexts (who work with Indigenous communities and local organizations) would also be beneficial. A further important role is the expansion of meaningful global participation in AI governance. This includes improving the inclusion of stakeholders who are currently underrepresented in international debates, particularly from the Global South, and affected communities, thereby strengthening the legitimacy, comprehensiveness and balance of global decision-making. This is particularly important where unequal access to skills and resources, such as compute and infrastructure, limits meaningful interoperability of technologies and governance approaches. Therefore, international cooperation should also look at such regional challenges and discrepancies to build capacity to empower broader participation, and to avoid unrealistic (and even harmful) one size fits all approaches.

The AI Dialogue should build upon and connect existing international, regional, and research initiatives to strengthen global AI governance while avoiding work duplication and initiatives fragmentation. Multilateral initiatives include UNESCO's Global AI Ethics and Governance Observatory's "AI, culture and IP" sub-group, proposing an important resources repository. The AI Dialogue should also connect with the UN Special Rapporteur in the field of cultural rights (whose work include two important recent reports, one on AI and Creativity (report A/80/278 - July 2025) and one on Digitalization of Cultural Heritage (report A/HRC/58/60 - February 2025), and with the UN Expert Mechanism on the Right to Development (who produced a recent report on AI, Cultural Rights and the Right to Development A/HRC/EMRTD/13/CRP.1 – April 2026). These connections would be valuable to ensure that the Dialogue builds on existing work and that AI governance is framed within broader questions of human rights, equity, inclusion, and sustainable development. It is crucial that data quality is addressed as an integral part of AI governance. As AI is only good as the data it is trained on or deployed on, considerations must be made on biased, harmful, incorrect and incomplete data and information. At the research and policy interface, UKRI/AHRC-funded programmes such as Bridging Responsible AI Divides (BRAID) and Towards a National Collection (TaNC) provide important empirical and interdisciplinary insights into the technical, societal and cultural implications of data and AI. Data governance initiatives and frameworks such as FAIR and CARE principles for Indigenous data governance, Local Contexts' Traditional Knowledge and Biocultural labels (developed with indigenous communities and local organisations), the Maori AI Governance Framework and Creative Commons Signals, are also important initiatives to connect with and build on. Lastly, the AI Dialogue should connect with the UN Environment Assembly UNEA-7 work (2025), which focused on the AI/environment intersection.

Attribution of responsibilities should be clearly defined to track progress and ensure Dialogue goals are being met. The UN Independent International Scientific Panel on AI could function as stewards of goal achievement for the Dialogue, ensuring surfacing evidence is recorded and translated into actionable policy. Its role should include monitoring progress, identifying implementation gaps, and ensuring that commitments are translated into measurable policy outcomes rather than remaining at consultation level alone. Diversity of participation should be a priority, with special attention to underrepresented stakeholders and perspectives, including Indigenous and marginalised communities and environmental concerns. Discussions should be organised into thematic sub-clusters and grounded on practical case studies, allowing greater depth, evidence-based discussion, and clearer identification of both risks and opportunities across sectors. Stakeholder participation should include those traditionally underrepresented in global governance processes, such as Indigenous groups and local communities, and stakeholders from the heritage sector in its full breadth and diversity. The first Global Dialogue event in July 2026 is only a starting point, and the UN should engage in a continuous dialogue with stakeholders, including with regional expansion, particularly in Global South countries, to fully engage local communities and regional perspectives (see under question 17 Indigenous methods of engagement with Indigenous groups, under Mariaelena Huambachano's 'Khipu Model'). Hybrid multilingual participation formats, including regional town halls/focus groups and asynchronous digital engagement, should be used to ensure accessibility across geographies, languages, cultures and resource settings, to ensure inclusive and meaningful global participation. Groundwork should be done on AI and digital literacy, to ensure stakeholders are made fully aware of and educated on AI risks and opportunities so that their Dialogue engagement is informed. Grassroots engagement should be actively supported through partnerships with local civil society organisations and community actors, ensuring participation is meaningful and capable of influencing final outcomes.

Underrepresented voices include Indigenous and local communities, particularly in the Global South. Their inclusion is essential for equitable and globally legitimate governance. Communities located near AI infrastructure, including data centres, are underrepresented despite experiencing significant environmental impacts. Also traditionally underrepresented are stakeholders from the heritage sector in its full breadth and diversity (organizations of all sizes, resources, geographical location and types, including libraries, archives, museums, galleries, national parks, natural heritage organisations, heritage science and research organisations etc.). Suggestions for meaningful inclusion of these voices include those outlined in the answer to question 15 above. I also suggest a more intentional calling for views of these underrepresented stakeholders, with the language of calls for comment and processes of consultations needing to be adjusted to ensure these stakeholders will be engaged. For example, I am noticing in AI and IP public consultations (for example in the UK) an emphasis, in the text of the call for views, on the need for views of AI firms on one side and the creative industries on the other. I believe this is a reductive binary that leaves out a range of stakeholders including Indigenous groups and the heritage sector, who should be expressly called for views. However, perhaps more effectively than collating their views through written consultations with short turnarounds, town halls/focus group/working group/roundtable approaches, in environments where these stakeholders could feel safe to share their views and concerns, could be more effective. Partnering up with researchers, organisations (such as Local Contexts) and communities of practice working with these groups could be a good way of connecting with them and building trust. Disseminating consultations and policy events through heritage sector listservs, such as AI4LAM, Museum Computer Group, Heritage AI and Law (HAIL) Network and Europeana, could also be a good inclusion strategy.

Innovative engagement formats for the AI Dialogue should prioritise interactive, practice-based, and inclusive approaches that move beyond traditional consultation models and enable more meaningful deliberation on complex governance challenges. Hybrid, asynchronous, multilingual and accessible digital platforms should be used to enable sustained engagement beyond live sessions, allowing participation across time zones and reducing travel costs, thus broadening accessibility and global inclusion. Examples of best practice in this regard, which the Dialogue could build on, include the Brazilian consultations between 2009/2010 for the Internet Bill of Rights (Marco Civil da Internet) and for the proposed Copyright Law Reform. These were made through asynchronous platforms, where submissions could be viewed by all if comments were made public by the respondents. Now, this is a standard approach for Brazilian public consultations, with a collaborative platform hosted by the Brazilian Government on the website Participa Mais Brasil: https://www.gov.br/participamaisbrasil/consultas-publicas. It includes accessibility translations, including with a sign language avatar widget. I would recommend connecting with Brazilian researchers at the Institute of Technology and Society (ITS-Rio) and InternetLab in Sao Paulo, to further explore these public consultation platforms and how can the AI Dialogue build something similar for Global AI Governance. Lastly, Indigenous methods for knowledge production and sovereignty, which meaningfully consider Indigenous voices, should be adopted. As Mariaelena Huambachano (2018) explains: "The Khipu Model may appear to have elements that are similar to the Western approach, for example, Western "focus groups" may be similar to Quechua "talking circles." However, talking circles in the Khipu Model distinguish the rituals and protocols underpinning Indigenous' ceremonial performance." (https://doi.org/10.1080/21683565.2018.1468380). Although Huambachano was discussing this approach in the context of food sovereignty discussions, I believe the Khipu Model talking circles method could also be applied as a method of engagement with Indigenous communities in AI governance discussions and consultations.

In AI's fast-paced and ever-changing governance landscape, relying on static documentation creates visibility gaps, whereas "living" trackers of cases, regulation and governance act as real-time resources that evolve with the work, ensuring decisions are based on current reality. In this respect, I believe good practices include the UNESCO subgroup on AI, culture and IP proposal for a civil-society led and co-designed online Repository and Resource Hub on AI, Intellectual Property, and Culture. Another example is the Brazilian model of public consultation platforms, identified in my answer to question 17 above. The Swedish "Remiss" system, with consultation outcomes published in Regeringen.se, should also be investigated as a possible best practice example in governance approaches. Due to the importance of good quality data and provenance identification, initiatives such as Local Contexts labels and Creative Commons licenses offer good models for the data. It should be investigated how can these be applied into AI domains. Other governance frameworks include indigenous data governance principles FAIR and CARE, and the Maori AI Governance Framework. The Creative Commons Signals proposal is also worth investigating, as it was designed to sustain the commons in the age of AI, offering a flexible, values-driven framework for content stewards to communicate expectations around AI use of content or data. The development of sector-specific AI governance guidance resources and toolkits should also be flagged as an important approach to support organisations planning and implementing responsible AI adoption.