AI Policy Lab, Umea University

Successful Global Dialogue on AI Governance would establish a clear shift from technology-led to purpose-led governance. AI should be developed and deployed only where it demonstrably addresses clearly defined societal needs, not as a default solution. This requires a shared global commitment to asking whether and under what conditions AI is appropriate in a given context before considering how to implement it. Such an approach reflects the importance of "Question Zero" ("Under what conditions should an AI system be adopted, if at all?") as a foundational step in responsible AI governance, ensuring that decisions are justified, proportionate and aligned with public interest, and supported by structured ex ante assessment processes. The Dialogue should reaffirm a human-centric and rights-based approach to AI. Governance frameworks must prioritise societal value, democratic principles and long-term sustainability over short-term economic competitiveness. Current policy trends that emphasise speed risk overlooking fundamental questions of purpose, benefit and harm. A successful outcome would therefore reorient global discussions towards inclusive, principled and evidence-based governance grounded in interdisciplinary research and real-world impacts. Inclusivity must be a core outcome. The Dialogue should ensure meaningful participation from diverse stakeholders, including academia, civil society and underrepresented communities, particularly those most affected by AI systems but least equipped to influence their development. This requires recognising global asymmetries in capacity, resources and decision-making power. In parallel, the Dialogue should promote practical tools and frameworks that support organisations in assessing motivations, stakeholders, risks and alternatives prior to AI adoptions. Finally, the Dialogue should strengthen accountability and coordination in AI governance. This requires clear allocation of responsibilities, alignment between decision-making power and risk ownership and enforceable regulatory mechanisms. At the global level, success also requires moving beyond narratives of an "AI race" towards a cooperative model grounded in shared direction, mutual learning and long-term public value.

• 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

These priorities reflect the need to re-centre AI governance on purpose, societal value and responsibility. Safety and trustworthiness cannot be reduced to technical robustness alone. They must include assessment of whether AI systems should be developed and deployed in the first place, under which conditions and for whose benefit. This requires moving beyond techno-solutionist approaches and ensuring that AI adoption is justified, proportionate and aligned with public interest. The broad societal implications of AI must be addressed in an integrated manner. AI systems reshape labour, institutions, knowledge practices and social relations. These impacts are unevenly distributed and often affect vulnerable groups most strongly. For indigenous communities, like those in the Arctic and others, whose knowledge is relational and place-based, the demand for explainable outputs is a rights issue. It requires translating knowledge into forms that models can process and it is in this translation that cultural and epistemic loss occurs. Transparency, accountability and human oversight are essential to make these principles actionable. However, they must be meaningful and context-aware, not reduced to formal compliance. Responsibility should be clearly allocated, aligned with decision-making power, and supported by practical tools and governance structures that enable informed and responsible use of AI in practice. Existing human rights treaties provide a robust and already established foundation for AI governance. Rather than creating new global mechanisms, efforts should focus on consistent and context-sensitive implementation. A key challenge is translating these norms into operational requirements for design, procurement and deployment. Without deliberate governance, AI risks reinforcing existing inequalities and creating new forms of harm.

We would like to highlight several additional cross-cutting and emerging issues that require greater attention. First, the distribution of benefits from AI remains insufficiently addressed. Current developments risk concentrating value, infrastructure and decision-making power in a small number of actors and regions. A key priority should be ensuring that AI does not exacerbate existing inequalities. A first step in this direction could be the concentration in compute, data and talent. Open-source software, open data and open models can play an important role in this regard, if supported by appropriate governance, capacity-building and safeguards. They can contribute to a more inclusive innovation ecosystem and reduce dependency on dominant providers. Second, there is a need to address governance gaps in areas beyond clear national jurisdiction. AI infrastructures and impacts increasingly operate across borders, including domains such as the high seas, outer space and global digital commons. These spaces lack comprehensive governance frameworks, yet are central to data flows, environmental impacts and resource use linked to AI systems. This raises important questions of accountability, stewardship and equitable access. Third, more attention should be given to the question of purpose in AI adoption. Governance discussions often assume that AI deployment is desirable and inevitable. A cross-cutting issue is the need to systematically assess whether AI is appropriate in a given context, and to consider alternatives. This requires embedding ex ante reflection into governance frameworks and organisational practices. Together, these issues highlight the importance of moving beyond narrow technical or sectoral approaches, towards a more holistic and globally coordinated understanding of AI governance.

Current governance gaps can be understood as a lack of structured approaches to assessing the appropriateness of AI deployment in specific contexts. Existing frameworks tend to prioritise compliance and risk mitigation after deployment, rather than evaluating whether the use of AI is justified in the first instance. This highlights the need for decision-oriented governance tools incorporating criteria such as necessity, proportionality and societal impact. Governance gaps in the selected thematic areas are already shaping developments, for example, within the European context, particularly in relation to the implementation of the EU AI Act. One key challenge is the assumption that regulatory instruments such as the AI Act can be readily aligned with existing human rights frameworks at international, regional and national levels. While this alignment is both necessary and desirable, in practice it is complex. Differences in interpretation, enforcement and institutional capacity create risks of fragmentation and inconsistent protection. At the same time, this process offers an important opportunity to operationalise human rights in concrete AI governance mechanisms and to strengthen coherence across legal regimes. A second challenge concerns behavioural and socio-technical aspects that are not fully addressed in current regulatory approaches. For instance, the EU AI Act focuses primarily on risk classification and system characteristics, but pays less attention to how AI systems shape behaviour, influence decision-making and enable practices such as nudging. These dynamics can have significant implications for autonomy, consent and democratic processes, yet remain underexplored in governance frameworks. More broadly, there is a persistent gap between formal governance requirements and practical implementation. Organisations are often expected to manage complex legal, ethical and technical responsibilities, without sufficient guidance or capacity. This creates risks of superficial compliance and unclear accountability structures.

The AI Dialogue can play an important role in advancing international cooperation by fostering transparency, inclusive deliberation and shared priority-setting. It should function as a space where assumptions about AI development and deployment are openly examined, including whether and where AI is appropriate. A key contribution of the Dialogue is to prioritise democratic deliberation across borders, including states, academia, civil society, affected communities and industry. At present, many global AI governance discussions risk underrepresenting actors from the Global Majority and indigenous communities. The AI Dialogue should include interests of underrepresented communities in international AI policy and governance conversations. Such participation is essential to ensure that governance reflects diverse perspectives and lived experiences, rather than narrow technical or economic interests. The Dialogue should actively address this imbalance by ensuring equitable participation, agenda-setting power and access to knowledge and resources. Without this, international cooperation risks reinforcing existing global inequalities in both AI development and governance. The Dialogue should also support coordination across existing governance frameworks. Rather than attempting to replace them, it can facilitate interoperability between international human rights instruments, regional regulations and sector-specific standards. This can help reduce fragmentation while respecting contextual differences and legal traditions. Finally, the Dialogue can promote the sharing of evidence, best practices and practical tools for responsible AI adoption. By grounding discussions in real-world challenges and interdisciplinary research, it can strengthen trust, mutual learning and long-term cooperation in AI governance. Effective international cooperation requires convergence not only on shared principles, but also on decision-making processes and governance capabilities. In this regard, the AI Dialogue could play a critical role in fostering common approaches to assessing risks, responsibilities and acceptable uses of AI across jurisdictions. By focusing on how decisions are made, the Dialogue can support more coherent, adaptive and resilient forms of international cooperation.

The AI Dialogue should build on existing initiatives that already provide substantive foundations for responsible and evidence-based AI governance. These include international frameworks such as the UNESCO Recommendation on AI Ethics and the OECD AI principles, as well as emerging research and practice-oriented initiatives such as AI policy research roadmaps, organisational self-assessment tools and AI policy literacy programmes. Greater emphasis should be placed on linking these initiatives to implementation practices. At the research and coordination level, several initiatives offer relevant approaches. For example, the AIPL and MILA Roadmap for AI Policy Research and similar roadmaps, highlight the importance of interdisciplinary, evidence-based approaches that connect technical development with societal impact, human rights and sustainability. They also emphasise the need for transboundary governance, stakeholder collaboration and capacity-building. The Dialogue can build on this by providing a global platform to coordinate such efforts, align research agendas and support shared evidence base for policymaking. At the operational level, structured self-assessment tools, such as Question Zero (Q0), offer practical mechanisms to operationalise responsible AI governance at organisational level, guiding decision-makers through reflection on purpose, alternatives, stakeholders and risks. The Dialogue could promote such tools as part of a broader toolkit for implementation, bridging the gap between high-level principles and everyday practice. Lastly, diverse AI Policy Literacy initiatives address a critical gap in knowledge and engagement. They equip diverse stakeholders with the knowledge and skills to navigate complex governance landscapes, understand policy processes and participate meaningfully in decision-making. Strengthening such initiatives is essential for inclusive and effective global governance. The added value of the AI Dialogue lies in connecting these efforts. It can foster coherence across fragmented initiatives, support interoperability between frameworks and promote the exchange of tools, knowledge and best practices. In doing so, it can move beyond principle-setting towards coordinated, practical and globally inclusive AI governance.

Stakeholder contributions should be framed in terms of clearly defined responsibilities and capacities within the AI lifecycle. Rather than treating stakeholders as homogeneous groups, the Dialogue should differentiate between roles such as developers, deployers, regulators and affected communities, and tailor participation mechanisms accordingly. Ensuring inclusive participation requires the AI Dialogue both structural support and carefully designed formats that enable meaningful contributions from diverse stakeholders. Participation should not be limited to states and large industry actors, but must actively include academia, civil society, public sector organisations and underrepresented regions. Regional and multilateral initiatives should be meaningfully integrated into the Dialogue. For example, regional bodies such as the African Union and initiatives like Smart Africa Alliance play an important role in including context-specific priorities and governance approaches. Their inclusion can help ensure that the Dialogue reflects diverse socio-economic realities and does not impose a one-size-fits-all approach. Further, it is crucial to provide sustained funding for independent expertise. This includes supporting long-term, independent research positions across regions to monitor AI developments and inform governance debates. Such roles should be protected from conflicts of interest and linked to transparent accountability mechanisms. Beyond formal participation, independent investigative journalism should be supported as a critical component of democratic oversight. Without these safeguards, AI governance risks being dominated by industry-driven narratives and interests. Finally, capacity-building and AI literacy should be embedded into the structure of the Dialogue. Enabling stakeholders to understand, engage with and shape AI governance is essential for meaningful participation and long-term positive impact. In terms of format, the Dialogue should combine global plenaries with smaller, thematic and regional working groups. This allows for both high-level coordination and context-sensitive discussion. Mechanisms for continuous engagement are also important, including iterative consultations, open calls for input and feedback loops that ensure contributions are reflected in outcomes.

Marginalised and vulnerable communities are often excluded from governance processes, despite being disproportionately affected by AI systems. This includes groups defined by socio-economic status, gender, disability, language and geographic location. Their perspectives are rarely reflected in system design, policy development or impact assessment. Despite the existence of global frameworks, less attention is given to context-specific approaches and knowledge developed at regional or national levels. This limits the diversity of governance models and risks imposing solutions that do not align with local realities. For instance, actors from the Global Majority, including African and Latin American countries, continue to face structural barriers to meaningful participation. Both regions contribute only a small share of global AI talent and have limited access to computing capacity, which constrains their ability to shape AI development and governance on their own terms. Also, Arctic and sub-Arctic communities' voices are underrepresented in AI governance. Their knowledge is structured in ways that face barriers and obstacles when it comes to participation and conversation formats. To address these gaps, inclusion must go beyond formal representation. It requires sustained investment in capacity-building, infrastructure and AI literacy, enabling stakeholders to participate on equal footing. By applying a human-centered approach, underrepresentation should be defined in terms of functional roles within socio-technical systems. Governance processes often overlook actors who are central to implementation, including public sector intermediaries, system operators and oversight bodies. Their systematic inclusion is essential to ensure that governance frameworks are both normatively robust and operationally feasible. This requires moving beyond ad hoc consultation towards structured, continuous engagement across the full lifecycle of AI systems. The AI Dialogue should therefore ensure equitable agenda-setting power, provide targeted support for underrepresented actors and create mechanisms for continuous and meaningful engagement. This is essential for legitimate, effective and globally relevant AI governance.

Effective engagement formats should explicitly reflect the socio-technical nature of AI systems by integrating technical, institutional and societal dimensions within a single deliberative space. AI should be addressed not merely as a technical artefact, but as a system embedded in complex social and governance contexts. Innovative engagement formats are essential to ensure that the AI Dialogue moves beyond symbolic participation towards meaningful and actionable outcomes. Traditional panel-based discussions should be complemented with structured, problem-driven and participatory formats. One promising approach is the use of facilitated "decision labs". These sessions would bring together diverse stakeholders to work on concrete use cases, guiding them through structured reflection on purpose, stakeholders, risks and alternatives. Such formats, inspired by Question Zero (Q0) approach, encourage participants to assess whether AI is appropriate rather than assuming its usage. Complementarily, "policy stress tests" could be introduced. These would simulate high-risk, cross-border or rapidly evolving scenarios in which existing governance frameworks are tested under pressure. This format allows participants to identify gaps, overlaps and inconsistencies in current approaches and to explore how different governance systems interact in practice. For inclusivity, engagement formats must actively reach beyond English-speaking and well-resourced communities. This includes multilingual formats, regional dialogues and locally facilitated sessions that integrate into global discussions. Hybrid models can open participation for stakeholders who cannot engage in centralised, in-person processes. Finally, continuous and iterative engagement should be prioritised. Mechanisms such as open consultations, feedback loops and collaborative drafting processes can ensure that contributions are reflected in outcomes and revisited over time. Together, these formats can support a more dynamic, inclusive and practice-oriented AI Dialogue, grounded in real-world challenges and diverse perspectives.

We would like to share several existing initiatives and approaches that provide concrete and transferable approaches for effective AI governance. First, roadmaps for AI Policy Research, such as the one developed by the AI Policy Lab and MILA, and many others, offer comprehensive frameworks for evidence-based and interdisciplinary governance. They emphasize human and planetary welfare, accountability, transparency and inclusivity, as well as promote transboundary cooperation and stakeholder collaboration. They also propose practical mechanisms such as communities of practice, policy fellowships and AI literacy initiatives to bridge gaps between research, policy and implementation. Second, self-assessment tools, such as the Question Zero (Q0), provide a practical approach to responsible AI adoption at organisational level. By guiding potential adopters through structured reflection on why AI is needed, who is affected and what risks and alternatives exist, it operationalises core governance principles in real-world decision-making. This type of tools help move beyond compliance towards meaningful accountability and informed adoption. Third, the AI Policy exchange forums, such as AIPEX, are open, online platforms designed to foster timely and accessible academic and policy discussions on AI policy and governance. They are positioned as between traditional academic journals and blogs, offering contributions that are reviewed for relevance, quality and appropriateness with a rapid publication process to support ongoing global debates. Finally, the AI Policy Literacy initiatives address a key gap in governance capacity. By equipping stakeholders with the knowledge to understand policy processes, compare global frameworks and engage with decision-making, they strengthen inclusive participation and informed decision-making. Together, these examples show that effective AI governance requires not only high-level principles, but also practical tools, capacity-building efforts and purpose-driven approaches that connect policy with on-the-ground implementation.