National ICT Company of Trinidad and Tobago (iGovTT)

A successful first Global Dialogue on AI Governance shouldn't feel like another conference where everyone nods, takes photos, and nothing changes. For me, success would look like three simple things: 1. People actually leave with something usable. Not vague principles, but practical agreements—like "this is how we ensure AI systems stay accurate," or "this is how governments should handle data updates." We already know that if the data is wrong or outdated, the AI fails and people lose trust quickly. That's the real problem to solve. 2. Smaller countries aren't just in the room—they're shaping the rules. Too often, governance gets defined by big tech or big countries. But places like ours are actually deploying AI in public services right now. If the dialogue doesn't reflect those realities, it misses the point. 3. It stays focused on people, not hype. AI governance isn't about the tech—it's about whether a citizen gets the right answer, at the right time, without confusion. If systems are inaccurate or inconsistent, trust drops fast. So the conversation has to stay grounded in real outcomes. And honestly, the biggest signal of success? If six months later, you can point to actual pilots, shared frameworks, or policies that came out of it. Not a PDF. Not a declaration. Real things being used. If that happens, then the dialogue mattered.

• Safe, secure and trustworthy AI
• AI capacity-building
• Interoperability of governance approaches
• Transparency, accountability, and human oversight

Safe, secure and trustworthy AI is the foundation. If users don't trust the system, adoption fails. In public services especially, even small inaccuracies can quickly erode confidence. Ensuring safety, reliability, and protection from misuse is essential to delivering real value. AI capacity-building is equally critical. Many governments lack the technical and institutional expertise to deploy, manage, or govern AI effectively. Without building this capacity, countries risk becoming passive consumers rather than informed implementers. Strengthening skills and understanding ensures better decision-making and long-term sustainability. Interoperability of governance approaches addresses a growing global challenge. AI systems and providers operate across borders, but regulations often do not. If governance frameworks are too fragmented, it creates inefficiencies and barriers to collaboration. Interoperability helps align standards, making it easier to adopt solutions, share knowledge, and scale responsibly-especially for smaller states. Transparency, accountability, and human oversight ensure that AI remains aligned with human values. Users need to understand how decisions are made, who is responsible, and when a human can intervene. This is essential for maintaining trust and preventing AI from becoming an unaccountable "black box."

a lot of AI governance discussion focuses on principles at the design stage-fairness, transparency, safety, rights, accountability. Those matter. But the harder question is what happens after deployment, when the system is live and interacting with real people in messy, changing conditions. "Operational reality" is about the day-to-day life of an AI system. It includes questions like: Who updates the knowledge base when policies change? How quickly are errors corrected? How are wrong or incomplete answers logged, reviewed, and fixed? Who decides when the AI should answer directly and when it should hand over to a human? How is performance measured over time-not just accuracy in testing, but user trust, consistency, accessibility, and usefulness in real situations? In real service environments, information changes constantly, edge cases appear, and public expectations are high. A system may perform well at launch but become unreliable if content is not refreshed, if no one owns quality control, or if there is no feedback loop from frontline staff and users. In that sense, governance is not only about setting rules for AI; it is about creating the institutional habits that keep the system dependable. So the "missing layer" is the practical machinery around AI: content governance, escalation paths, audit routines, human oversight in practice, staff training, user feedback channels, and continuous improvement. Without that layer, even a well-designed AI system can drift, confuse users, or lose trust over time. Put simply: responsible AI is not just about how you build it. It is also about how you run it.

The governance gaps are affecting the public sector in very practical, day-to-day ways. First, on safe, secure and trustworthy AI, the main impact is on public trust. Even when systems are fast and easy to use, gaps in data validation and update processes can lead to incomplete or outdated responses. In a public service context, this quickly undermines confidence, especially for high-stakes services where accuracy matters most. Second, on AI capacity-building, there is a clear gap between ambition and readiness. While there is strong interest in adopting AI, many institutions do not yet have the internal skills, processes, or governance structures needed to manage it effectively. This affects everything from content quality to oversight and long-term maintenance. Third, interoperability of governance approaches remains a challenge. Public services are delivered across multiple ministries and agencies, but citizens expect a single, consistent experience. Without aligned standards and coordinated governance, fragmentation persists, leading to inconsistent information and user frustration. Fourth, on transparency, accountability, and human oversight, the gaps become visible when things go wrong. There is often limited clarity on who is responsible for correcting errors, how quickly issues are resolved, and when human intervention should occur. This makes it harder to maintain trust and ensure accountability. Overall, these governance gaps translate into three real impacts: reduced public trust, inconsistent service delivery, and slower, more cautious adoption of AI across the public sector.

The AI Dialogue can actually make a difference if it focuses on getting countries aligned in a practical way—not just agreeing on principles, but agreeing on what those principles look like in action. It's a chance for countries to compare notes honestly: what's working, what's failing, and where the real risks are showing up when AI is used in everyday services. It also creates space for real collaboration, especially for countries that don't have the same level of resources or expertise. Sharing tools, approaches, and lessons learned can level the playing field and prevent everyone from reinventing the wheel. If done right, the Dialogue becomes less about talk and more about helping countries move forward together in a way that actually improves how AI works for people.

There are already important initiatives emerging from the Global South that the AI Dialogue should build on. For example, countries like India have shown how digital public infrastructure can support AI-enabled services at national scale, while Brazil has advanced open data and participatory digital governance models. Across Africa, initiatives like Smart Africa and national AI strategies in Rwanda and Kenya are actively deploying AI in sectors like health and agriculture under real constraints. In the Caribbean, systems like Anansi provide another strong reference point—particularly through its AI + Human Intelligence (AI + HI) approach, where AI handles routine queries but seamlessly hands off to human agents when needed. This hybrid model has already demonstrated high user satisfaction and practical viability in a public service context . The common thread across these examples is that they are operational systems, not just policy frameworks. They deal with real issues like data gaps, evolving services, and user trust—making them highly relevant to global governance discussions. The AI Dialogue can add value by connecting and elevating these lived experiences. It can strengthen South–South collaboration, enabling countries to share what is actually working—whether it's hybrid service models, mobile-first delivery, or scalable digital infrastructure. It can also help translate these practical approaches into adaptable governance patterns. Most importantly, the Dialogue can ensure that global AI governance is informed not just by theory, but by real-world implementation from the Global South, where innovation is often driven by necessity and impact.

For the AI Dialogue to be effective, it needs to be structured in a way that allows different stakeholders to contribute based on what they actually do, not just what they represent. Governments should bring real implementation experiences—what has worked, what has failed, and where governance gaps are showing up in practice. Private sector and developers can contribute technical insights, especially around system design, safety, and scaling. Academia and civil society should play a critical role in testing assumptions, highlighting risks, and ensuring human rights and inclusion remain central. Frontline service providers (often overlooked) can share how AI performs in real interactions with users—this is where many issues first appear. In terms of structure, the Dialogue should move away from purely plenary discussions and adopt a more working model: Thematic working groups focused on practical areas (e.g. data governance, trust, AI + human collaboration), producing short, actionable outputs—not long reports. Case-based sessions, where countries and organizations present real deployments, including challenges and lessons learned—not just success stories. Peer exchange forums, especially for Global South collaboration, allowing countries with similar constraints to share approaches directly. Ongoing engagement, not a one-off event—through follow-up sessions, shared pilots, and progress tracking. Most importantly, the Dialogue should prioritise implementation over theory. Its value will come from creating a space where stakeholders don't just discuss AI governance, but actively shape how it works in real-world settings.

Several important voices are still underrepresented in global AI governance—and their absence creates a gap between policy and reality. First, practitioners from the Global South who are actively deploying AI are often missing. While many discussions are led by larger economies, countries in the Global South are already implementing AI in public services under real constraints. Their experience—especially around hybrid models, data limitations, and citizen trust—is highly relevant and should be more visible. Second, frontline public servants and service providers are rarely included. These are the people who see how AI performs in real interactions with citizens—where confusion happens, where systems fail, and where human intervention is needed. Their insights are critical to designing systems that actually work. Third, low digital literacy and vulnerable populations are often spoken about, but not directly engaged. This includes the elderly, rural communities, and persons with limited access to technology. Their experiences should inform design and governance, not just be considered as an afterthought. Fourth, small states and regional blocs are underrepresented. They face unique challenges—limited resources, smaller datasets, and dependence on external technologies—but also tend to innovate in practical, efficient ways. To include these perspectives, the Dialogue should move beyond traditional representation and adopt deliberate inclusion mechanisms: Dedicated Global South and small-state forums Case-based sessions led by implementers, not just policymakers Structured input from frontline staff and service agencies Community engagement channels that capture user experiences directly In short, governance will only be meaningful if it reflects the voices of those actually building, running, and using AI systems—not just those regulating them.

To make the AI Dialogue truly engaging, the format needs to shift from speeches to interaction and problem-solving. One effective approach is case-based sessions. Instead of abstract discussions, countries or organizations present real AI deployments—including what went wrong. This creates more honest, practical conversations and allows others to learn from lived experience. Another strong format is live problem labs. Small, mixed groups (policy, technical, frontline) work together on a specific challenge—e.g. how to ensure data accuracy or when AI should escalate to a human. These sessions can produce tangible outputs within the Dialogue itself. Simulation exercises could also be powerful. Participants are given realistic scenarios—such as an AI system giving incorrect guidance in a public service—and must decide how governance, accountability, and response mechanisms should work. This helps stress-test policies in a practical way. To amplify Global South voices, the Dialogue could include peer exchange circles—smaller, informal sessions where countries with similar constraints share tools, lessons, and approaches directly, without the pressure of formal presentations. Another option is "open floor" practitioner sessions, where frontline staff, developers, or operators can share short, unfiltered insights from real use—what users ask, where systems fail, and what fixes are needed. Finally, the Dialogue should include continuity mechanisms—such as ongoing working groups or shared pilot initiatives—so engagement doesn't end when the event does.

Several practical approaches offer useful models for effective AI governance. One example is the use of digital public infrastructure (DPI) in countries like India, where platforms such as Aadhaar and UPI provide a trusted, interoperable foundation for delivering AI-enabled services at scale. These systems show how governance can be embedded into infrastructure-through standards, identity verification, and secure data exchange. Another is the adoption of hybrid AI + human models in public service delivery. In the Caribbean, for instance, AI systems are designed to handle routine queries while seamlessly escalating complex or sensitive cases to human agents. This approach ensures efficiency without sacrificing accountability or user trust, and provides a clear pathway for oversight. In Africa, countries like Rwanda and Kenya are advancing national AI strategies that focus on capacity-building and sector-specific deployment (e.g. health, agriculture). These strategies prioritize practical use cases, local data ecosystems, and partnerships, rather than purely theoretical frameworks. Open data and participatory governance approaches in countries like Brazil also provide a strong model. By making datasets accessible and involving citizens in oversight, these systems improve transparency and allow for broader scrutiny of AI-driven decisions. At an operational level, continuous audit and feedback mechanisms are critical. This includes regularly testing AI systems against real-world scenarios, tracking errors, and updating content and models based on user feedback and policy changes. Governance, in this sense, becomes an ongoing process rather than a one-time certification. Together, these examples highlight a shift toward practical, implementation-driven governance-where trust is built through systems that are transparent, adaptive, and grounded in real-world use.