GlobalSouth.ai

A successful first Global Dialogue on AI Governance should deliver three concrete outcomes. First, a shared evidence baseline. The Dialogue should establish a structured, open, and continuously updated repository of real-world AI deployments across regions—especially in developing countries. This should capture use cases, context, benefits, risks, and governance gaps. Without a common evidence base, discussions on safety, fairness, and human rights remain abstract and difficult to operationalize. Second, a practical interoperability layer for governance. Rather than attempting to harmonize all regulations, the Dialogue should define a minimal set of interoperable elements—such as documentation standards, risk classification principles, and audit expectations—that can travel across jurisdictions. This would allow countries with different capacities to adopt governance approaches that are compatible, without imposing one-size-fits-all frameworks. Third, a clear pathway for capacity-building tied to real needs. Capacity-building efforts should be grounded in observed gaps from real deployments—such as data quality, linguistic inclusion, institutional readiness, and technical expertise. The Dialogue should map these gaps and align existing UN and multilateral mechanisms to address them, particularly for countries with limited infrastructure and regulatory capacity. Across all three outcomes, success depends on one shift: treating Global South participation not only as representation, but as a source of evidence and knowledge production. If the Dialogue can move from principles to evidence, from fragmentation to interoperability, and from inclusion to capability, it will create a foundation that future governance efforts can actually build on.

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

My selected priorities reflect a common underlying gap: the need to ground AI governance in real-world deployment contexts, particularly in developing countries. Safe, secure and trustworthy AI cannot be achieved through principles alone. In many Global South settings, risks emerge not only from model design, but from data quality, institutional capacity, and deployment environments. Understanding these requires field-based evidence. AI capacity-building is similarly context-specific. Capacity gaps are not limited to technical skills, but include data infrastructure, linguistic representation, institutional readiness, and governance capabilities. Without identifying these gaps through real-world use cases, capacity-building efforts risk being misaligned with actual needs. The social, economic, cultural, and linguistic implications of AI are especially pronounced in diverse and resource-constrained environments. AI systems deployed in multilingual and informal settings can produce unintended exclusions or inequities that are often not captured in existing evaluation frameworks. Finally, the protection and promotion of human rights in AI requires operationalization. Principles such as fairness, accountability, and non-discrimination must be translated into measurable and enforceable practices across different institutional contexts, including those with limited regulatory capacity. Across all four areas, a common requirement is the development of a structured evidence base that captures how AI systems function in practice across sectors such as healthcare, finance, education, and public services.

Yes. While the listed themes are comprehensive, a key cross-cutting gap is the absence of a structured focus on real-world deployment evidence and feedback loops. Most current governance approaches are designed around principles, risk categories, or model-level assessments. However, in practice, especially in developing countries, AI systems interact with complex institutional environments, including uneven infrastructure, multilingual populations, informal systems, and varying levels of regulatory capacity. Many risks and unintended consequences only become visible after deployment. This creates a need for a dedicated focus on evidence generation from real-world use cases, including both opportunities and harms, across sectors such as healthcare, finance, education, and public services. Without this, it is difficult to assess whether AI is contributing to development outcomes, how digital divides are evolving, or what types of safeguards are effective in different contexts. A second emerging issue is the need to treat context sensitivity as a core design principle for governance, rather than as an afterthought. Governance approaches that work in high-capacity environments may not translate effectively to resource-constrained settings, raising questions about portability and interoperability. Finally, there is a growing need for continuous governance mechanisms, where systems are monitored and evaluated throughout their lifecycle, rather than assessed only at the point of deployment. This is particularly important in dynamic environments where data, usage patterns, and risks evolve. Addressing these cross-cutting issues would strengthen all existing themes by ensuring that governance frameworks remain grounded, adaptive, and responsive to real-world conditions.

Governance gaps in the selected thematic areas are already shaping how AI is being deployed across sectors in India and other Global South contexts, with distinct implications across financial services, healthcare, and public systems. In the BFSI sector, AI is increasingly used for credit assessment, fraud detection, and customer risk profiling. The primary challenge is data quality and explainability in heterogeneous data environments, where informal financial behavior is not fully captured in structured datasets. This creates risks of exclusion and misclassification, particularly for underserved populations. At the same time, there is an opportunity to build more inclusive and context-aware financial models, supported by stronger model validation, documentation, and fairness assessment practices. In healthcare, AI systems are being introduced in diagnostics, triaging, and telemedicine platforms. A key governance gap lies in data representativeness and clinical validation across diverse populations. Models trained on limited or urban-centric datasets may not generalize well across regions, leading to potential disparities in care. However, this also creates an opportunity to develop locally validated, population-aware AI systems, and to strengthen governance around data quality, evaluation standards, and human oversight in clinical workflows. In public systems, including welfare delivery and digital public infrastructure, AI is used for targeting, monitoring, and decision support. The main challenge is the interaction between AI systems and institutional capacity constraints, including documentation gaps, limited auditability, and weak grievance redress mechanisms. This raises concerns around transparency, accountability, and rights protection at scale. At the same time, there is an opportunity to design public-interest AI governance frameworks that embed auditability, explainability, and inclusion from the outset. Across all three sectors, a common pattern emerges: governance gaps are not only technical but institutional. Addressing them presents an opportunity for the Global South to lead in developing context-sensitive, deployment-driven AI governance approaches that are both practical and scalable.

The AI Dialogue can play a meaningful role in advancing international cooperation by shifting the focus from principle-setting to practice-sharing and implementation support. First, it can serve as a platform for building a shared global evidence base on AI deployment. By systematically documenting real-world use cases—across sectors such as healthcare, finance, education, and public services—the Dialogue can help countries understand what works, what fails, and under what conditions. This is particularly important for developing countries, where institutional capacity, data quality, and infrastructure constraints shape how AI systems behave in practice. Second, the Dialogue can facilitate interoperability without uniformity. Rather than harmonizing regulations, it can help define a minimal set of common elements—such as documentation standards, risk classification approaches, and evaluation practices—that allow different governance frameworks to remain compatible across jurisdictions. Third, it can strengthen capacity-building through cooperation. By mapping real-world governance gaps—technical, institutional, and linguistic—the Dialogue can help align existing UN and multilateral initiatives with the actual needs of countries, especially those with limited resources. Finally, the Dialogue can elevate the role of the Global South from participation to co-production of knowledge. Enabling researchers, practitioners, and institutions from developing countries to contribute evidence and insights will make global governance more representative and more effective. If structured in this way, the AI Dialogue can move beyond coordination toward collective learning, which is essential for governing a rapidly evolving technology across diverse contexts.

The AI Dialogue should build on existing global frameworks and initiatives, including the OECD AI Principles, UNESCO Recommendation on the Ethics of Artificial Intelligence, Global Partnership on AI, and the NIST AI Risk Management Framework. It should also connect with regional regulatory efforts such as the EU AI Act, as well as emerging national approaches in countries like India. These initiatives have established important foundations—common principles, risk management approaches, and governance taxonomies. However, they remain fragmented and unevenly implemented, particularly in contexts with limited institutional capacity. The added value of the AI Dialogue lies in three areas. First, it can act as a convergence layer, not by harmonizing regulations, but by identifying interoperable elements across frameworks—such as documentation standards, risk classification approaches, and evaluation practices—that can travel across jurisdictions. Second, it can serve as a platform for linking principles to practice by integrating real-world deployment evidence into governance discussions. This includes documenting how frameworks perform across sectors and contexts, especially in developing countries. Third, it can strengthen capacity-aligned cooperation by mapping governance gaps and aligning existing multilateral initiatives toward practical support—such as technical assistance, data infrastructure, and institutional capability building. In doing so, the Dialogue can move from a landscape of parallel initiatives to a more coordinated and implementation-focused ecosystem, where global frameworks are informed by, and adaptable to, diverse real-world conditions.

To make these contributions actionable, the AI Dialogue should adopt a more structured format. First, it should organize discussions around use-case tracks (e.g., healthcare, finance, public services), rather than only thematic principles. This anchors conversations in real-world contexts. Second, each track should require standardized inputs from participants, including: use case description, deployment context, benefits, risks, governance gaps, and capacity constraints. This enables comparability across countries and sectors. Third, the Dialogue should incorporate evidence synthesis outputs, where insights from discussions are translated into practical guidance—such as governance patterns, common failure modes, and transferable practices.

Several voices remain underrepresented in global AI governance discussions. First, practitioners working within public systems in developing countries—including administrators, frontline implementers, and technical teams—who experience how AI systems interact with institutional constraints such as data gaps, legacy infrastructure, and limited oversight capacity. Second, communities affected by AI in informal and low-resource settings, where impacts on livelihoods, access to services, and inclusion are significant but often undocumented. Third, researchers and institutions from the Global South, whose work is frequently underrepresented in global evidence bases and policy frameworks, despite operating in some of the most complex deployment environments. Fourth, domain-specific experts in sectors such as healthcare, financial inclusion, and welfare delivery, who can provide grounded insights into how AI affects high-stakes decision-making in real-world contexts. These gaps persist not only due to limited participation, but due to the way contributions are structured.