University of Europe for Applied Sciences

Three outcomes would signal real success: 1) From principles to enforceable commitments: we already have no shortage of ethical guidelines. Success would mean agreeing on minimum viable obligations, such as baseline transparency standards, documentation requirements (e.g., model cards), and clear accountability for AI-assisted decisions. Not a perfect consensus, but a shared floor that no actor falls below. 2) Expanding the definition of "AI stakeholders": the dialogue must move beyond governments and tech companies to explicitly include those often invisible in the system: data workers, annotators, affected communities, and public-sector implementers. Governance that excludes the "human supply chain" of AI is incomplete. Success would be recognizing and embedding their rights, protections, and voice into governance structures. 3) Operationalizing oversight, not just recommending it: we need clarity on who audits, how, and with what authority. A meaningful outcome would include pathways for independent auditing, cross-border regulatory cooperation, and mechanisms to challenge or contest AI decisions, especially in high-stakes domains like healthcare, hiring, and public services.

• AI capacity-building
• Transparency, accountability, and human oversight
• Protection and promotion of human rights
• Social, economic, ethical, cultural, linguistic and technical implications of AI

The transparency, accountability, and human oversight theme comes first because without them, everything else is aspirational. We are already seeing decisions justified by "the model recommended it." That is not governance. We need clarity on how decisions are made, who owns them, and how they can be challenged, especially in high-stakes domains. This directly connects to the protection and promotion of human rights. AI systems are not neutral; they shape access to opportunities, services, and justice. Governance must ensure that rights are not indirectly eroded through automated systems, particularly for vulnerable or underrepresented groups. I also prioritize the social, economic, ethical, cultural, linguistic, and technical implications of AI because most failures are not purely technical-they emerge at the intersection of systems and society. Bias, exclusion, and misalignment often reflect deeper structural issues. If governance does not address this complexity, it will remain surface-level. Finally, AI capacity-building is critical. We cannot govern what we do not understand. This is not only about technical skills, but about building institutional and leadership capacity to question, interpret, and make informed decisions about AI systems. Without this, oversight becomes symbolic rather than effective. Across all four, the underlying concern is the same: moving from abstract principles to decisions that can be explained, challenged, and owned.

1) Cognitive drift and over-reliance: As AI becomes embedded in decision workflows, the risk is not just error, but it is the gradual erosion of human judgment. When systems consistently suggest, summarize, and decide, humans stop interrogating. Governance must address not only what AI does, but what it does to us over time. 2) The hidden labor behind AI: AI systems are often presented as automated, yet they rely on vast human input: data annotators, content moderators, and outsourced labor. Their working conditions, rights, and well-being are largely absent from governance discussions. If we govern outputs but ignore the human supply chain, we are governing incompletely.

In the UAE, and more broadly across fast-moving sectors like higher education, public services, and digital transformation, the gap is no longer about adoption. It is about governing at the same speed we are deploying. The most significant challenge is the illusion of control. We are integrating AI into decision-making, student evaluation, hiring, and service delivery faster than we are building the mechanisms to question those decisions. "Human oversight" often exists in policy, but in practice, decisions are accepted as efficient rather than being interrogated. This creates a quiet shift from decision support to decision substitution. A second challenge is capacity asymmetry. Leadership is expected to govern AI strategically, while operational teams interact with it daily; yet both often lack the depth needed to critically evaluate outputs. This turns governance into a compliance exercise rather than a capability. There is also a growing risk around context misalignment. Many AI systems are trained and optimized outside the region, yet applied locally in multilingual, culturally specific, and policy-sensitive environments. Without adaptation, this creates blind spots in fairness, interpretation, and relevance. But the opportunity is equally significant. The UAE is uniquely positioned to design governance while building systems, not after. With strong national AI agendas and centralized coordination, there is an opportunity to move beyond frameworks into live governance models that embed transparency, auditability, and accountability directly into platforms and workflows. This means rethinking not only how AI is used but how humans are prepared to use it: from users of AI → to informed decision-makers in AI-augmented systems. Finally, and often missing, is the designer/developer layer. The assumptions, data choices, and optimization goals built into systems are rarely visible once deployed. Yet they shape every output. Governance tends to focus on use, not design, leaving a critical blind spot where bias, limitations, and trade-offs are embedded upstream but only discovered downstream. At the same time, there is a real opportunity. Moving early from frameworks to engineered governance, where transparency, traceability, and accountability are built across the full lifecycle, not added after deployment. The priority here is not just to keep humans "in the loop," but to ensure that design, deployment, and decision-making are all accountable to one another.

The AI Dialogue can matter, but only if it moves beyond coordination into shared responsibility with consequences. Today, international cooperation risks becoming a comparison exercise: frameworks, principles, positioning. Useful, but not enough. The Dialogue should instead function as a bridge between systems, not just a forum for alignment. Three roles are critical: 1) Setting a global floor, not a global ceiling: we don't need perfect agreement—we need minimum non-negotiables. Baseline expectations for transparency, accountability, and human rights that apply across contexts, while still allowing regional adaptation. Without this, cooperation fragments into incompatible standards. 2) Connecting accountability across borders: AI systems are built, trained, and deployed across jurisdictions. When harm occurs, international responsibility becomes blurred. The Dialogue can push for interoperable accountability—shared approaches to auditing, documentation, and redress mechanisms that don't stop at national boundaries. 3) Rebalancing capacity, not just influence: there is a real asymmetry: a few actors shape AI, many are expected to govern or absorb it. Cooperation must include capacity-building as a core mechanism, not an afterthought; so countries and institutions can meaningfully evaluate, adapt, and challenge AI systems, not just adopt them. But beyond structures, the Dialogue has a deeper role: shifting the mindset from competition to stewardship. AI governance cannot be treated as a race for advantage. These systems already operate across borders, shaping outcomes beyond any single jurisdiction, so responsibility cannot remain fragmented. If the Dialogue moves participants from simply aligning positions to sharing ownership of risks and decisions, it will achieve something meaningful: connecting countries and anchoring accountability at a global level.

It should anchor itself in established frameworks such as the OECD AI Principles and the UNESCO Recommendation on the Ethics of Artificial Intelligence, while linking to implementation-driven efforts such as the Global Partnership on AI and regulatory frameworks such as the EU AI Act. Technical standards bodies like ISO/IEC JTC 1/SC 42 should also be part of the same conversation. The Dialogue should build on what already exists (the issue is not a lack of frameworks, but a lack of connection). This is where the AI Dialogue can add value: 1) Translation across layers: connect high-level principles to technical standards and operational practices. Today, ethics, regulation, and engineering often evolve in parallel—not together. 2) Interoperability in practice: not just aligning language, but enabling systems, audits, and documentation to work across jurisdictions. This is critical for AI that is developed in one place and deployed in another. 3) Linking accountability across ecosystems: existing initiatives tend to focus on specific actors—governments, companies, or researchers. The Dialogue can bridge these, ensuring accountability flows across the full lifecycle: design, deployment, and decision. 4) Moving from guidance to evidence: encourage shared pilots, benchmarks, and reporting mechanisms that test governance in real settings, not just in policy documents.

Different stakeholders must contribute from their respective realities: governments highlighting concrete governance gaps, industry sharing system limitations and accountability challenges, academia translating complexity into usable frameworks, civil society surfacing real-world impacts, and technical experts making systems understandable and auditable. Structurally, the Dialogue should be grounded in real use cases, span the full lifecycle, and avoid siloed discussions by integrating diverse perspectives. Most importantly, it should produce testable outputs with clear ownership and operate as a continuous cycle—convene, test, report, refine—so governance is not just discussed, but practiced.

Global AI governance still underrepresents those closest to its real impact: the hidden data workforce (annotators and moderators), frontline decision-makers (teachers, clinicians, caseworkers), affected communities, and actors from the Global South who often adopt but do not shape AI systems. The issue is not their absence, but their distance from decision-making. Inclusion must therefore be designed, not assumed—through formal representation of data workers, structured input from practitioners based on real deployments, participatory mechanisms for affected communities, and capacity-building that enables meaningful engagement in standard-setting. Until governance reflects those who build, apply, and live with AI, it will remain coherent in principle—but incomplete in practice.

For me, meaningful engagement comes from designing the Dialogue as a working space, not a speaking space. I would emphasize formats that bring people into real decision contexts—case-based sessions, cross-role interactions, and moments where assumptions are challenged, not just shared.

Practical AI governance means embedding transparency, auditability, and clear decision ownership directly into systems-from design to deployment-not adding them afterward. Happy to exchange more in the space!