AVPN

For the inaugural Global Dialogue on AI Governance to be deemed a success, it must transcend mere discussion and establish a credible foundation for future international cooperation. Success should be measured by three concrete outcomes. First, a shared diagnostic of the landscape. The Dialogue must produce a consensus-based mapping of the current AI governance ecosystem. This includes identifying existing governance gaps, acknowledging the diverse risk perceptions across different regions, and recognizing the specific vulnerabilities of developing nations. This shared understanding is the prerequisite for any subsequent collective action. Second, success requires the emergence of tangible pathways for capacity building. The Dialogue must move beyond high-level principles to generate actionable commitments. This means establishing pilot programs or partnerships focused on resource-sharing, such as creating open-access AI safety toolkits, funding computational access for researchers in under-resourced regions, and launching technical assistance programs for policy-makers. A clear, time-bound plan for how knowledge and tools will be shared is essential. Third, the Dialogue must prove its value as a convening body by fostering unprecedented inclusivity. Success means amplifying voices from the Global South, civil society, technical communities, and the private sector, ensuring they are not just observers but active contributors. The final outcome document should reflect this diversity, demonstrating that the UN is the unique forum capable of bridging geopolitical divides. Ultimately, success is planting the seeds for a durable, flexible, and inclusive global framework. If the Dialogue concludes with a clear roadmap for future work, concrete partnership pledges, and a demonstrated commitment to listening to all stakeholders, it will have set a gold standard for global AI governance.

• AI capacity-building
• Transparency, accountability, and human oversight
• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI

First, Safe, secure and trustworthy AI is the foundational prerequisite for all else. Without a bedrock of safety and security, no AI system can be sustainably integrated into society, nor can it engender the public trust necessary for adoption. This includes technical robustness against adversarial attacks and the alignment of systems with human intent. Second, these safety goals cannot be achieved globally without massive investment in AI capacity-building. The concentration of AI expertise and infrastructure creates a two-speed world, where the majority of nations are consumers of AI technologies designed elsewhere, with little ability to shape them according to their own needs or values. Urgent action here is about democratizing access to tools, skills, and compute power. Third, we must look beyond the technology itself to its social, economic, ethical, cultural, and linguistic implications. AI is not neutral; it reflects the data it is trained on. To prevent cultural homogenization and digital neo-colonialism, we must urgently develop frameworks that protect and promote linguistic diversity and ensure AI systems are compatible with diverse ethical traditions and local economic realities. Finally, all of this must be underpinned by transparency, accountability, and human oversight. These are the core governance mechanisms that make safety and trustworthiness operational. Without clear lines of accountability and the ability for humans to review and override automated decisions, we risk creating autonomous systems that erode democratic principles and individual autonomy. These four priorities are not separate tracks, but mutually reinforcing pillars of a just and effective AI future.

Yes, while the listed themes provide a comprehensive foundation, a critical cross-cutting issue that demands explicit attention is the environmental impact of the AI lifecycle. From the extraction of rare earth minerals for hardware, to the immense energy and water consumption required to train and run large-scale models, AI has a significant and rapidly growing ecological footprint. This dimension is conspicuously absent from the current thematic areas. It intersects directly with several of them: Social and Economic Implications: The environmental costs of AI disproportionately affect communities in the Global South, where resources are often extracted and where the physical impacts of climate change are most acutely felt. Capacity-building: For many developing nations, the environmental cost of adopting or building advanced AI infrastructure (e.g., data centers) may be prohibitive, creating a new barrier to entry. Safety and Trustworthiness: A system that is technically safe for a user but environmentally destructive for the planet cannot be considered truly "trustworthy" in a holistic sense. As we pursue the immense benefits of AI, we must urgently develop metrics and standards for "Green AI." This involves promoting research into more energy-efficient algorithms, requiring transparency reports on a model's carbon and water footprint, and ensuring that AI governance frameworks align with global climate commitments like the Paris Agreement. Ignoring this dimension risks creating a form of governance that is socially conscious but ecologically blind. Addressing the environmental impact of AI is not an alternative to the existing themes; it is a prerequisite for ensuring that the technology's development is genuinely sustainable and equitable for both people and the planet.

Building on the priorities previously outlined, the governance gaps and current developments in these thematic areas are creating a clear set of challenges and opportunities for our entity and the global landscape we operate within. Safe, secure and trustworthy AI: The Challenge of Fragmentation: The most significant governance gap here is the lack of universally accepted technical standards and certification mechanisms. This creates a fragmented landscape where an AI system deemed "safe" in one jurisdiction may not meet the bar in another. For our sector, this increases compliance costs and stifles cross-border innovation. However, it also presents an opportunity to lead the push for international, interoperable standards that ensure a baseline of safety without stifling progress. AI Capacity-building: The Widening Divide: The rapid advance of AI is outpacing the ability of most nations to build the necessary human and infrastructural capacity. This is creating a two-tiered system: a small group of "AI-makers" and a large group of "AI-takers." For our region, the primary challenge is the brain drain and the dependency on foreign-built models that may not reflect local languages or values. The opportunity lies in fostering south-south cooperation and leveraging open-source ecosystems to leapfrog development stages, building local expertise and solutions tailored to local needs. Social, Economic, and Cultural Implications: The Risk of Homogenization: Current AI models, trained predominantly on English-language, Western-centric data, risk erasing cultural and linguistic diversity. The governance gap is the absence of proactive measures to protect and promote this diversity. For our entity, the challenge is to ensure AI becomes a tool for cultural preservation, not erosion. The opportunity is to champion the development of foundational models that are inherently multilingual and multicultural from the ground up, requiring targeted investment in diverse data curation. Transparency, Accountability, and Human Oversight: The Black Box Problem: As AI systems become more complex, their decision-making processes become more opaque. The governance gap is the lag between this complexity and the legal and technical tools required to audit them. In our sector, the challenge is ensuring that critical decisions affecting individuals (e.g., in finance, healthcare, or justice) can be explained and contested. The opportunity lies in driving innovation in "explainable AI" (XAI) and establishing clear legal liability frameworks that hold developers and deployers accountable, thereby reinforcing public trust in the technology. In essence, these gaps highlight that the greatest risk is not AI itself, but our collective failure to govern it inclusively. The opportunity, therefore, is to build a governance ecosystem that is as dynamic and globally interconnected as the technology it seeks to guide.

The AI Dialogue can serve a unique and indispensable role that no other existing forum currently fills: it can act as the universal legitimizing and bridging mechanism for global AI governance. First, it can serve as the primary forum for norm-setting and legitimation. While technical standard-setting may happen in bodies like the ISO or IEEE, and detailed policy discussions occur in venues like the OECD or GPAI, these often lack universal membership. The UN, through this Dialogue, provides the only table where every member state has an equal seat. It can take the best technical work from these other bodies and translate it into broadly accepted normative frameworks, giving them political legitimacy and a global stamp of approval. Second, the Dialogue can function as an essential "connecting hub" or clearinghouse. The AI governance landscape is currently fragmented, with numerous initiatives operating in silos. The UN Dialogue can map these efforts, identify gaps and overlaps, and facilitate communication between them. It can ensure that a standards body is aware of a human rights initiative, and vice versa, fostering a more coherent global ecosystem. Third, it plays a critical role in inclusion and capacity-building. By design, it brings in the voices of the Global South, civil society, and other stakeholders often marginalized in more technocratic or exclusive forums. This ensures that global norms are not just dictated by a few powerful nations but are co-created with the world. It can also serve as the primary platform for launching and coordinating global capacity-building efforts, ensuring that all nations have the ability to participate in and shape the AI future. In essence, the Dialogue is not meant to replace existing initiatives but to unite them under a common umbrella of shared principles and inclusive deliberation, ensuring that the pursuit of AI governance is as global and representative as the technology itself.

To be effective, the AI Dialogue must not operate in a vacuum. It should strategically build upon and connect with the rich ecosystem of existing initiatives. Key initiatives to connect with include: Multilateral and Inter-governmental Bodies: The OECD's AI Policy Observatory offers a wealth of data and policy analysis. The Council of Europe's Framework Convention on AI provides a foundational treaty focused on human rights, democracy, and the rule of law. The Global Partnership on AI (GPAI) brings together leading experts to advance responsible AI. Technical Standard-Setting Organizations: Bodies like the ISO/IEC (through its subcommittee on AI) and IEEE are where the technical standards that operationalize "safe and trustworthy" AI are developed. The Dialogue must be connected to these efforts to ensure they align with broader societal values. Multi-stakeholder and Civil Society Initiatives: The Partnership on AI includes companies and civil society. Initiatives like the Montreal Declaration for Responsible AI offer principles developed through public deliberation. These bring grassroots and ethical perspectives. UN Specialized Agencies: UNESCO's Recommendation on the Ethics of AI is a critical normative instrument. ITU works on AI for good and technical standards. OHCHR provides the indispensable human rights lens. The Added Value of the AI Dialogue: The Dialogue's unique contribution is its political universality and convening power. While the OECD brings together mostly wealthy nations, and GPAI is a limited group, the UN Dialogue includes every country. Its added value is threefold: Synthesis and Harmonization: It can synthesize the principles from UNESCO, the technical standards from ISO, and the policy expertise from the OECD into a coherent, globally owned framework. Gap Identification and Agenda-Setting: With its panoramic view, it can identify critical gaps—like the environmental impact of AI—that fall between the mandates of existing bodies and set a global agenda to address them. Legitimacy and Inclusivity: It can take the work of more exclusive expert bodies and give it political legitimacy by subjecting it to inclusive, intergovernmental deliberation, ensuring that the final outcomes have the buy-in of all nations. This makes the Dialogue the indispensable anchor for a truly global AI governance architecture.

Different stakeholders bring unique assets to the table, and the Dialogue's structure must be designed to harness this diversity effectively. Member States provide political legitimacy and are the primary actors for implementing governance frameworks. They should contribute by articulating their national priorities, regulatory needs, and the specific challenges faced by their constituencies, particularly from the Global South. Private Sector Entities possess technical expertise and are on the front lines of innovation. They can contribute by sharing practical insights on what is technologically feasible, demonstrating existing safety and transparency practices, and committing to voluntary principles that raise the bar for the entire industry. Civil Society Organizations serve as the essential ethical compass and advocates for public interest. They can hold other stakeholders accountable, amplify the voices of marginalized communities, and ensure that human rights and social justice remain at the core of all deliberations. Technical and Academic Communities bring data-driven analysis and long-term foresight. They can provide evidence-based assessments of AI's impacts, develop measurement standards for safety and fairness, and offer independent evaluations of governance proposals. International Organizations offer expertise and convening power within specific domains (e.g., UNESCO on ethics, ITU on technical standards). They can ensure coherence and prevent duplication of efforts. Recommendations for Format and Structure: The Dialogue should adopt a hybrid, multi-track structure to accommodate these diverse contributions. First, a High-Level Segment for political vision-setting by Heads of State and Ministers. Second, a series of Thematic Deep-Dive Tracks running in parallel, corresponding to the priority areas (e.g., a track on Capacity-Building, a track on Safety Standards). These tracks should be co-moderated by a diverse team (e.g., a representative from a Member State, a technical expert, and a civil society leader). Third, an Innovation and Solutions Expo where stakeholders can showcase tangible tools, partnerships, and commitments. Crucially, all proceedings must be accessible remotely, with real-time translation, to ensure global participation beyond those physically present.

Despite progress, several critical voices remain on the margins of global AI governance. Underrepresented Voices: The Global South, particularly from Africa, Latin America, and parts of Asia: These regions are often the recipients of AI technologies developed elsewhere, yet their policymakers, researchers, and civil society lack a proportional voice in shaping the rules. Informal Sector Workers and Labor Unions: The impact of AI on the future of work is profound, yet the perspectives of gig workers and those in vulnerable employment are rarely centered in high-level discussions. Indigenous Peoples and Local Communities: Their knowledge systems, languages, and cultural heritage are at risk of being appropriated or erased by AI, but they are seldom at the table when data governance and cultural implications are debated. Disability Advocates: AI presents both immense opportunities and serious risks for persons with disabilities, from accessibility tools to algorithmic bias in hiring or healthcare. Their lived expertise is often an afterthought. Youth: As the generation that will live the longest with AI's consequences, young people have a profound stake, yet intergenerational dialogue is largely absent. Strategies for Inclusion: Dedicated Fellowship and Sponsorship Programs: Provide funding and logistical support for delegates from underrepresented groups to participate fully in the Dialogue and its preparatory processes. Decentralized Consultations: Hold regional and national-level consultations in the lead-up to the global gathering, ensuring that inputs are gathered in local languages and contexts, then synthesized into the main proceedings. Partnerships with Representative Organizations: Work directly with organizations that already have deep trust within these communities, such as trade unions, indigenous rights networks, and youth-led movements, to co-design participation. Accessible and Diverse Formats: Ensure all materials are available in multiple formats (easy-read, sign language interpretation) and that sessions are scheduled to accommodate different time zones. Create dedicated spaces for these voices, such as a "Civil Society and Community Assembly," that reports directly into the main plenary.

To move beyond static presentations and foster genuine collaboration, the Dialogue should embrace a suite of innovative engagement formats. 1. The "Global AI Policy Sprint": A 24-48 hour, fast-paced, collaborative workshop held virtually across all time zones before the main event. Diverse, distributed teams would compete to draft policy briefs, regulatory prototypes, or technical standards on a given challenge. The winning solutions would be presented at the Dialogue, injecting fresh, co-created ideas into the formal proceedings. 2. Structured Deliberation and "World Cafés": Instead of one-way speeches, dedicate significant time to small-group, facilitated discussions using the "World Café" methodology. Participants rotate between tables, each focused on a specific question (e.g., "How do we fund global AI safety research?"). This fosters cross-pollination of ideas and ensures that every participant, not just the loudest voices, can contribute. 3. The "Futures Studio": A dedicated space where artists, designers, and science fiction writers collaborate with policymakers and technologists to create tangible "artifacts from the future." This could involve designing a public service announcement about AI risks from 2035, or prototyping a user interface for an accountable AI system. This format makes abstract consequences visceral and sparks creative thinking about desirable futures. 4. Interactive Scenario Exercises: Present participants with a realistic, high-stakes crisis scenario involving AI (e.g., a deepfake-driven election interference event, or a catastrophic AI failure in critical infrastructure). Divide into stakeholder groups and run a real-time simulation of how they would respond, govern, and coordinate. This reveals gaps in existing frameworks and builds trust through collaborative problem-solving under pressure. 5. The "Commitment Wall" and Progress Tracker: Create a dynamic, public-facing digital platform where stakeholders can make concrete, trackable commitments (e.g., "We will fund 100 AI safety scholarships," "We will open-source our model card template"). This shifts the focus from rhetoric to action and enables transparent accountability over time, turning the Dialogue from an event into a process.

The AVPN and Google.org AI Opportunity Fund: Asia-Pacific serves as an exemplary model of a multi-stakeholder approach that addresses several critical governance challenges simultaneously-particularly in the areas of capacity-building and the social and economic implications of AI . This initiative, launched in 2024 with a USD 15 million commitment from Google.org and expanded with an additional USD 10 million in Phase Two, represents a holistic platform designed to ensure that the benefits of AI are distributed equitably across one of the world's most diverse regions . Its approach offers several concrete lessons for effective AI governance. Key Features of the Approach 1. Multi-Stakeholder Collaboration: The Fund is built on a partnership between AVPN (the largest network of social investors in Asia), Google.org (philanthropic arm of a technology leader), and the Asian Development Bank (a multilateral development institution) . This structure brings together philanthropic capital, technical expertise, and regional development knowledge, demonstrating how different sectors can pool resources for a common public interest goal. 2. Localized and Contextualized Training: Rather than imposing a one-size-fits-all curriculum, the Fund operates through a "train-the-trainer" model. It selected 57 local training providers across Phase One and Two-organizations deeply embedded in their communities . Strategic partners like AI Singapore helped develop localized training materials based on Google's foundational courses, which were then adapted by local providers for their specific contexts . For example: Hapinoy in the Philippines trains women micro-entrepreneurs running "sari-sari" stores, combining "High-Tech with High-Touch" through face-to-face community training that makes AI accessible for daily business operations like inventory management and financial tracking . Tictag in Singapore develops accessible AI modules for migrant workers, teaching skills such as data management that can enhance employability and enable entrepreneurship both in Singapore and when workers return home . Ruang Kolaborasi Perempuan in Indonesia equips agricultural workers and fishers with AI tools for weather predictions, yield assessments, and production planning-applications directly relevant to their livelihoods . Centre for Social and Behaviour Change in India strengthens capabilities of self-help group women, integrating AI modules into government training systems through collaboration with the AI Mission of Government of Uttar Pradesh . 3. Targeting the Underserved: The Fund explicitly focuses on workers most impacted by AI-driven workforce transitions, including domestic workers, caregivers, persons with disabilities, farmers, unemployed individuals, and MSME employees . This addresses a critical governance gap identified in earlier discussions-ensuring that marginalized communities are not left behind. 4. Building a Sustainable Ecosystem: Beyond direct training, the initiative is constructing a lasting infrastructure for AI skilling. This includes: A publicly available AI training content hub ("AI Learning for the Future of Work") where anyone can discover and enroll in localized courses by market or language . An AI Skilling Policy Toolkit for governments and policymakers, translating design principles into actionable policy considerations and implementation pathways . The AIM ASEAN programme, a dedicated track for supporting MSMEs in Southeast Asia launched with the ASEAN Foundation . 5. Scale and Ambition: The Fund aims to train 720,000 workers across Asia-Pacific and 100,000 MSMEs in Southeast Asia by 2027 . As of December 2025, Phase One had already trained over 300,000 workers . This demonstrates that inclusive, localized AI capacity-building can operate at significant scale. why the AVPN and Google.org AI Opportunity Fund is a good example of AI governance in action: It solves a real problem. Instead of just talking about the "