Empowering teams with clarity, consistency, and lasting impact across sectors.

‹ Back

AI is learning faster than we are: What Digital Cooperation Day taught me about governance, capacity and continuous learning

Reflections from UN Digital Cooperation Day 2026 — New York City, NY

On September 21, I joined government representatives, international organizations, academia, industry and civil society in New York for the United Nations Digital Cooperation Day, organized by the UN Office for Digital and Emerging Technologies (ODET). The theme was “Shaping a Global AI Future Through Science, Policy, and Capacity”, and the idea I kept coming back to throughout the day was simpler: The true test of AI policy is not the quality of the idea. It is whether we have the capacity to implement it and keep learning fast enough to govern what comes next.
       That distinction matters as the Global Digital Compact already gives the international community an important foundation. It calls for responsible, accountable, transparent and human centric AI. It requires interoperable approaches to AI governance, capacity building and stronger human oversight. It also explicitly recognizes that the digital world is evolving at pace and that cooperation must be capable of anticipating, assessing, monitoring and adapting to emerging technologies.
       What I heard in New York was a powerful reminder that implementation capacity may now be the critical variable, and one of the most useful formulations I heard was: IDEA → MANDATE → STRATEGY. It sounds straightforward, but in practice, each transition represents a significant governance challenge as a compelling vision does not create implementation capacity. As Vilas Dhar emphasized, vision without a budget, meaningful capacity to deliver, and structures of accountability to execute remains just that: a vision. That observation connects directly with the work I have been developing around AI governance capacity, which we often talk about in terms of compute, infrastructure, data, technical talent and access to models. Those things do matter, but what about the capacity to decide:

• Where should AI be used?
• Where should it not be used?
• Who is accountable?
• Whose values and rights are represented?
• Who controls the data?
• What risks are acceptable?
• How do we know whether an AI system is delivering the intended outcome?

Add to that the question of how quickly governance can adapt when the technology changes, and these questions are clearly not primarily technology questions, but governance questions.
 

The Global Digital Compact is significant because it moves beyond a conversation about technological adoption toward international cooperation on the conditions under which digital technologies and AI should develop and be used. Among its commitments are building capacity, particularly in developing countries to access, develop, use and govern AI, supporting interoperability between governance approaches, and promoting transparency, accountability and robust human oversight as outlined by the UN. 
       That language matters as it suggests that governance is itself a form of capacity. My own work builds on this premise as I have been exploring what happens in the space between responsible AI principles and the point where an organization must decide. That is where I see the AI Governance–Innovation Gap; AI capabilities can move from experimentation to operational deployment extraordinarily quickly. Institutional policies, procurement processes, workforce capabilities, risk frameworks, legislation and accountability structures generally do not move at the same speed making for a dangerous mismatch:

                                            AI capability ↑          Governance capacity →

The question is not whether governance should stop innovation, but whether governance capacity can develop quickly enough to accompany it. AI-RESPECT™ (AI-R.™ for short) guides leaders from principles to decisions through this operational AI governance framework built around seven dimensions:

1. Responsibility
2. Ethics
3. Sovereignty
4. Protection
5. Explainability
6. Compliance
7. Transparency

I did not develop it as another global standard or as a replacement for regulation, conversely, it is a practical mechanism for helping organizations operationalize the principles, standards and obligations that already exist, contributing in the Digital Cooperation Portal. At its core are seven questions:

1. Who is accountable for the AI system?
2. Whose values and rights does the AI system reflect?
3. Who owns the data and under what authority?
4. Who is protected and who may be at risk of being harmed?
5. Can clients, collaborators or the public understand how it works?
6. Does the AI system meet current and future compliance standards?
7. Are we being transparent about AI use?

I use these questions as a pressure test; if an organization cannot answer them clearly, the problem may not be technological. It may be a governance capacity problem.

The discussion of capacity in New York became one of the strongest themes of Digital Cooperation Day. Across very different countries and contexts, the conversation repeatedly returned to foundational capacity. For example, Lesotho's experience implementing the UN Blueprint for AI and Data Strategies highlighted an all of government approach, connecting ministries with academia and the private sector. In discussions about capacity building, the ingredients were remarkably consistent:

• human capital;
• people and skills;
• infrastructure;
• digital innovation;
• institutional knowledge;
• policy and governance;
• interoperability;
• partnerships.

Senegal's experience highlighted the importance of building a regional ecosystem around AI, including innovation, policy and governance, interoperability and priority use cases. Other discussions emphasized the connection between universities, governments and the private sector. The lesson for me was clear: AI capacity is not a technology stack, it is an ecosystem and within that ecosystem, people are infrastructure. Another idea that stayed with me came from Tanzania's Minister of Communications and Information Technology, Mhe. Angellah Jasmine Mbelwa Kairuki (Mb), that readiness is a sequence, not a single step, which in itself, may be one of the most important concepts for AI governance.

Organizations often talk about being “AI ready” as though readiness is a status that can be achieved. I increasingly think about it differently, in that AI readiness is a continuous organizational capability and requires organizations to repeatedly assess:

• What has changed?
• What do we now know?
• What has the technology made possible?
• What new risks have emerged?
• What skills have become obsolete?
• What decisions should remain human?
• What governance controls need to change?

That is why I believe continuous learning must become part of AI governance itself. If AI is learning, are we, and if so, can we keep up? Perhaps the most provocative theme of the day was the speed of change. Several speakers discussed the accelerating pace of AI capability, the changing shelf life of skills and the need to continually reskill and reshape education.
       One observation shared during the conference was that AI capabilities are evolving so rapidly that skills and curricula cannot be treated as static assets. Whether we measure that acceleration in months, model generations or changing workplace tasks, the underlying problem is the same, our learning systems were designed for a world in which knowledge depreciated relatively slowly. AI is changing that assumption, as Sangheon Lee of the International Labour Organization (ILO) spoke about AI potentially compounding existing skills gaps. The OECD's Andreas Schleicher brought the conversation back to education and critical thinking. A particularly striking observation was that education systems using AI less intensively are, in some contexts, achieving stronger learning outcomes. The lesson is not necessarily “use less AI”, it is more nuanced. We need AI applications that strengthen human thinking rather than simply automate it. That means education must increasingly teach people not only how to use AI, but when to use it, when not to use it, how to challenge it, and when human judgement must prevail. That is what I would call socio-technical literacy, where the human remains the critical control point, and which connects directly to my work in AI governance. Across healthcare, education and national defence, the same fundamental question can be asked: What should remain human?
       In healthcare, that might mean accountability for an AI supported clinical decision; in education, it might mean determining when a student should receive AI assistance and when struggling through a problem is itself part of learning and in national defence, it might mean determining how AI supported surveillance or environmental intelligence is used when sovereignty, Indigenous rights and national security intersect (in the Arctic for example). At the heart of all three examples, the technology changes, yet the governance question remains. This is why I increasingly describe the human as the control point in AI governance, and not because humans are always better at every task, nut because accountability, values, rights and legitimate authority cannot simply be delegated to a system because the system is capable of performing the task.

Another phrase from the conference captured the broader architecture; INFORMATION → DIALOGUE → CAPACITY BUILDING. I think there is an important fourth step, which is IMPLEMENTATION. And then, because AI does not stand still, LEARNING. So perhaps the cycle looks like this: Information → Dialogue → Capacity → Implementation → Learning → Adaptation. Then the cycle begins again, through a continuous improvement/ continuous learning loop. That is what makes AI governance fundamentally different from writing a policy document and putting it on a shelf because a governance framework that cannot learn is likely to become obsolete.
 

The Global Digital Compact itself recognizes this challenge, calling for digital cooperation that is forward looking and capable of identifying, anticipating, assessing, monitoring and adapting to emerging technologies. That principle strongly resonates with the direction of my work, and my contribution of connecting governance to implementation. This is not to suggest that we need another universal AI rulebook as there are already important international principles, frameworks and governance processes. While the United Nations has established an Independent International Scientific Panel on AI and a Global Dialogue on AI Governance, the Global Digital Compact provides a framework for capacity building, interoperability, transparency, accountability and human oversight.
       The challenge I am interested in is what happens one level down, specifically, how organizations turn those principles into repeatable decisions. That query has shaped my work across several very different environments, with a central question: How do we build the governance capacity required to translate rapidly advancing AI capability into responsible, equitable and trusted outcomes? The next AI divide may be a capacity divide as the Global Digital Compact explicitly recognizes the need to build AI capacities and bridge AI divides. It also calls for education and training, interoperable competency frameworks and continuous adaptation to rapid technological change, which leads me to a broader proposition; the next AI divide will not simply be between those who have access to AI and those who do not. It may increasingly be between organizations and societies that have the capacity to govern, learn and adapt with AI and those that do not. This is where economic sovereignty, digital sovereignty and institutional capacity begin to converge. Access to infrastructure, compute, data and talent matters, but so does the ability to make good decisions about all of them.
       One question raised during the conference was particularly important: What role should the UN play? The answer was:

• Architecture
• Standards
• Policies

But also, there is something more fundamental: capacity. The UN's emerging AI governance architecture is already moving in this direction through the Global Digital Compact and the Scientific Panel and the Global Dialogue on AI Governance. The Dialogue's themes explicitly include AI capacity building, interoperability of governance approaches, transparency, accountability and human oversight. The opportunity now is to connect those international mechanisms to implementation on the ground. That means supporting countries and institutions not only to adopt principles, but to develop the people, structures, processes, measurement systems and learning mechanisms required to implement them.
       I left New York with a question that has become even more central to my work: If we measure AI capability, why don't we measure AI governance capacity with the same seriousness? We could begin measuring things such as:

• governance readiness;
• percentage of AI systems with clear accountability;
• AI risk assessment and mitigation;
• workforce governance literacy;
• data and sovereignty controls;
• explainability and recourse;
• compliance readiness;
• transparency;
• time required to adapt governance when technology changes;
• an organization's ability to learn from AI incidents and outcomes.

This would move AI governance from being primarily a principle to being a measurable organizational capability. That is the space where I see AI-RESPECT™ contributing as a guiding framework.

My biggest takeaway from Digital Cooperation Day was not that we need more ambition, there was plenty of ambition in the room. It was that ambition must be connected to implementation; vision needs a mandate, a mandate needs a strategy, a strategy needs capacity, capacity needs people and people need continuous learning, and that learning needs to be embedded into governance.
       AI is changing too quickly for governance to be something we design once. The organizations that are prepared for the next phase of AI will not necessarily be those that predict the future correctly, they will be those capable of learning, governing and adapting continuously as the future arrives. That is, ultimately, what I took away from New York. The question is no longer simply: “Are we ready for AI?” it is; “Can we build institutions capable of staying ready as AI continues to learn?” 
 
Perhaps that is the next frontier of digital cooperation: not simply building AI capacity, but building the human and institutional capacity to govern continuous technological change. I will lean on the United Nations Under-Secretary-General and Special Envoy for Digital and Emerging Technologies, Mr. Amandeep Singh Gill, for further collaborative direction.

CT