Wansati Lab

The first Global Dialogue on AI Governance would be a success if we leave with clarity, trust, and action. We need clarity on shared principles. We don't have to agree on every rule, but if we can align on basics like safety, transparency, fairness, and human control, that gives everyone a common language. Right now AI is discussed in very technical terms, and that keeps many people out of the conversation. We also need trust between groups that don't usually sit at the same table. Governments, researchers, companies, and civil society often work in silos. If this dialogue brings in voices from the Global South and local communities, not just big players, we'll start to build real relationships. Trust is what makes future cooperation possible. Finally, we need concrete next steps, not just speeches. Success would be 2-3 practical commitments we can track. For example, a pilot for cross-border AI incident reporting, or a simple toolkit for assessing AI risks in education and health. Small steps show people this process is serious.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Open-source software, open data and open AI models
• AI capacity-building
• Transparency, accountability, and human oversight

From my work with Wansati Lab in Mozambique, these 4 areas are where AI can make or break inclusion. AI capacity-building - is urgent because communities in the Global South are still excluded from AI design and use. Without local skills, training, and infrastructure, we become consumers of AI instead of co-creators. My association focuses on digital education for women and youth, so building capacity is our core work. Social, economic, ethical, cultural, linguistic and technical implications - matter because AI built in one context often fails in another. In Mozambique we have multiple languages, low connectivity, and informal economies. If AI systems don't reflect these realities, they deepen inequality. We need active engagement to ensure local impact is studied and addressed. Transparency, accountability, and human oversight - are critical for trust. People need to understand how AI decisions affect their jobs, education, and access to services. For our association, this means pushing for systems that communities can question and influence, not black boxes. Open-source software, open data and open AI models - reduce cost and dependency. In low-resource settings, proprietary models are not sustainable. Open tools let us adapt solutions for local languages and needs, and let communities audit how systems work.

Yes, two cross-cutting issues could be included: 1. AI and environmental impact AI training and deployment use large amounts of energy, water, and hardware. This affects climate goals and local communities, especially in countries already facing energy shortages. The current themes focus on social and technical risks, but they don't address sustainability. For global AI governance to be fair, we should also discuss how to measure and reduce AI's environmental footprint, and how to share green computing resources across regions. 2. AI access in low-connectivity, low-resource contexts Most themes assume stable internet, electricity, and devices. In many parts of Africa and rural areas, that is not the reality. If governance only focuses on "open models" or "capacity-building" without addressing offline use, local data collection, and lightweight AI systems, we will still exclude millions of people. An emerging priority should be designing AI governance for edge computing, offline-first tools, and local data sovereignty. These issues are cross-cutting because they touch every other theme. You can have transparent, ethical AI, but if it only runs in cloud data centers or in English, it will not serve most of the world.

For Wansati Lab, the governance gaps in these four areas directly affect how we operate and serve communities. AI capacity-building Our team has strong local knowledge but limited technical AI skills. We cannot yet build or adapt models for local languages or for offline use. Training and mentorship are scarce and expensive. Opportunity: We are trusted by women and youth groups across Mozambique. If governance supports capacity-building partnerships, we can train community facilitators to use and explain AI tools, making us a bridge between technology and local needs. Social, economic, ethical, cultural, linguistic and technical implications Most AI tools available to us do not reflect Mozambican languages, culture, or informal economies. When we pilot digital education programs, we see low adoption because the content feels foreign. Opportunity: Our association works directly with communities. This gives us insight to co-design AI solutions that are culturally relevant. With the right framework, we can test and share models that work in real local contexts. Transparency, accountability, and human oversight We use AI tools from partners but do not always know how decisions are made. This creates risk for our beneficiaries and for our reputation. Opportunity: We can set an example by adopting simple transparency practices: explain how tools are used, record community feedback, and involve beneficiaries in oversight. This builds trust and shows donors we are responsible. Open-source software, open data and open AI models Proprietary platforms are costly and limit customization. We also lack access to local datasets to train models. Opportunity: Open-source tools let us adapt AI for local languages and low-connectivity settings. If governance promotes open data sharing, we can build solutions that are affordable and scalable for our work.

It can bring governments, civil society, academia, and Global South communities into AI governance talks. This prevents rules from being shaped only by big economies and tech firms, and makes sure local realities are reflected. The Dialogue can turn shared values like safety and transparency into joint projects. Examples include cross-border AI incident reporting or common toolkits for education and health. Practical steps build trust and show progress. With many countries creating different AI rules, the Dialogue can align approaches on interoperability, open data, and oversight. This lowers costs and helps smaller organizations comply without duplication.

UNESCO Recommendation on AI Ethics (2021): Leverages the established, 193-country adopted framework to ensure ethical alignment. African Union AI Strategy (2025-2030): Amplifies regional priorities focusing on data sovereignty, capacity building, and local, development oriented AI solutions.

To ensure the AI Dialogue moves beyond high-level discussion and achieves measurable local impact, it must leverage the specific strengths of diverse actors within a structured, inclusive format. I. Stakeholder Roles: A Collaborative EcosystemThe Dialogue functions as a partnership where each group provides a critical component of the AI lifecycle: Governments (The Enablers): Set policy direction and provide the legal frameworks and funding necessary for cross-border pilots and regional capacity-building. Civil Society & Associations (The Conscience): Provide "ground-truth" evidence of AI's social and cultural impact. They act as essential testers for tools, ensuring they serve community needs. Academia & Researchers (The Validators): Offer independent, evidence-based analysis of risks and benefits. They are responsible for developing open datasets and objective evaluation standards. Private Sector (The Engine): Contribute technical expertise, practical governance models, and open-source tools, while committing to transparency and local partnership. Youth & Local Communities (The North Star): Ground the Dialogue in reality by sharing lived experiences regarding education, jobs, and daily life, ensuring policy remains human-centric. II. Recommendations for Format and Structure To transform these contributions into action, the Dialogue should be organized around the following four pillars: Multi-Track Design: A dual-track system should combine High-Level Plenaries (for political alignment) with Thematic Working Groups. This allows technical experts and community leaders to tackle specific issues such as open data or agricultural AI, without waiting for high-level consensus. Regional-First Consultations: Global agendas are often reactive. By holding Regional Dialogues prior to global summits, the priorities of the Global South and local hubs will shape the international agenda from the outset. Action-Oriented Outputs: Each cycle must culminate in Concrete Commitments, such as shared open-source toolkits, localized pilots, or joint reporting standards. These outputs should include clear ownership, timelines, and accountability mechanisms. Inclusion & Accessibility: To remove barriers for small NGOs and grassroots groups, the Dialogue must utilize hybrid formats, provide materials in local languages, and offer financial support for participation

Effective AI governance requires shifting from a top-down approach to actively including voices from the Global South, grassroots civil society, Indigenous communities, and workers, who are currently marginalized in international policy discussions. Meaningful inclusion can be achieved through decentralized regional dialogues, language justice initiatives, dedicated funding, and formal reporting channels for local impact, transforming tokenism into equitable partnership.

The AI Dialogue can accelerate impact by adopting proven engagement models, including UNESCO-style decentralized regional hubs, GPAI-inspired policy sprints, and OECD-aligned live policy labs. Integrating community-led "Fishbowl" formats and open-source innovation sprints ensures inclusive participation and the development of actionable, tangible tools.

1. Comprehensive Policy Frameworks UNESCO Recommendation on AI Ethics: Adopted by 193 countries, it acts as a global ethical baseline, enabling nations like Brazil and Senegal to build culturally relevant national strategies. EU AI Act: A pioneering, risk-based legal framework that establishes clear compliance criteria for transparency and human oversight, providing a blueprint for enforceable AI legislation. 2. Accountability Practices AI Impact Assessments (AIIA): Utilized in Canada and Singapore, these assessments mandate the evaluation of bias and privacy risks prior to public-sector deployment, embedding accountability into the design phase. Regulatory Sandboxes: Controlled environments (e.g., UK and Singapore) where startups can pilot innovations under regulatory supervision, reducing market entry barriers while ensuring safety. 3. Collaborative Platforms & Infrastructure OECD AI Policy Observatory: A centralized hub that aggregates global data and case studies, helping policymakers avoid duplication and adopt evidence-based approaches. Open-Source Hubs (e.g., Hugging Face): These platforms democratize access by providing auditable, adaptable models for low-resource languages, preventing "vendor lock-in" for the Global South. 4. Innovative Delivery Approaches AI Policy Hackathons: Interdisciplinary sprints where developers and lawyers co-create functional tools, such as bias-detection kits for local dialects. Community AI Labs: Grassroots initiatives across some countries in Africa that empower youth to co-design AI solutions for local agricultural or health needs.