The Hague University of Applied Science (THUAS)

A successful first Global Dialogue on AI Governance should not be measured by the breadth of discussion, but by the clarity of outcomes. Three concrete results would make this dialogue meaningful: 1. Explicit recognition of infrastructure-level governance The Dialogue should establish that the most consequential AI governance decisions occur upstream - in infrastructure, data layers, and system design - not only at the level of applications or use cases. This recognition should be reflected in a shared conceptual framework. 2. A requirement for articulating governance choices Participating actors (governments, institutions, and organizations) should be encouraged to explicitly state their position regarding different AI infrastructure models (commercial, public, and commons-based), including associated trade-offs in control, accountability, and dependency. 3. A practical pathway from principles to implementation The Dialogue should move beyond high-level principles by outlining how public values can be translated into concrete system requirements, procurement criteria, and governance mechanisms. Ultimately, success would mean shifting AI governance from implicit, fragmented decision-making toward explicit, accountable, and contestable design choices. Without such a shift, there is a risk that global AI systems will continue to evolve by default, rather than by deliberate and collectively shaped governance.

• Safe, secure and trustworthy AI
• AI capacity-building
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches

These priorities reflect a need to move from fragmented, implicit AI adoption toward coordinated and explicit governance. "Safe, secure and trustworthy AI" remains essential, but in practice this cannot be achieved without addressing how systems are designed and integrated. This is why interoperability of governance approaches is critical: without alignment across frameworks, standards, and institutional practices, governance remains inconsistent and difficult to operationalize. AI capacity-building is equally important, particularly in public sector and educational contexts, where institutions are expected to implement AI responsibly but often lack the structural support to do so. Capacity-building should therefore include not only skills, but also governance capabilities and institutional design. The broad category of social, economic, ethical, cultural, linguistic and technical implications reflects the reality that AI systems do not operate in isolation, but reshape institutional and societal structures. Taken together, these priorities emphasize a shift toward governance that is systemic, implementable, and aligned across contexts, rather than fragmented or purely principle-driven.

A key cross-cutting issue not sufficiently captured in existing themes is the governance of shared context and underlying infrastructure. As AI systems increasingly operate on shared data layers, models, and contextual frameworks, a significant portion of decision-making is effectively embedded upstream, before any specific application is deployed. These layers determine how information is interpreted, what is considered relevant, and which perspectives are prioritized. This introduces a systemic risk: the emergence of a dominant or standardized interpretation of reality, driven by technical design choices rather than explicit governance. Current governance approaches tend to focus on outcomes (e.g. safety, fairness, accountability), but do not sufficiently address who defines the underlying context on which AI systems operate, nor how plurality of perspectives is preserved. This issue cuts across all thematic areas, including safety, human rights, and interoperability. Addressing it requires: * governance mechanisms at the level of shared data and context * explicit consideration of plurality and contestability * recognition that infrastructure design is inherently a governance act Without this, AI governance risks remaining reactive, while the most consequential decisions are made implicitly and upstream.

In higher education and the broader public sector, governance gaps are increasingly visible as a mismatch between the pace of AI adoption and the capacity to govern it coherently. A primary challenge is that AI is often introduced through decentralized pilots, tools, and local initiatives, while governance remains fragmented. This leads to implicit decision-making regarding infrastructure, data use, and dependencies on external providers, without clear ownership or mandate. As a result: - institutions become structurally dependent on commercial AI systems - governance responsibilities are diffuse and difficult to operationalize - alignment between public values and actual system design remains weak Another key challenge is the gap between principles and implementation. While frameworks for trustworthy and ethical AI are widely available, institutions lack practical mechanisms to translate these into procurement criteria, architectural choices, and day-to-day practices. At the same time, there are clear opportunities. AI creates the possibility to redesign institutional processes, improve accessibility of services, and strengthen knowledge work. In education, it can support more adaptive learning, reduce administrative burden, and enhance research capabilities. However, realizing these opportunities requires a shift toward explicit, system-level governance. This includes developing institutional capacity not only in skills, but in governance design, and making foundational choices about infrastructure and dependencies transparent and accountable.

The AI Dialogue can play a critical role by shifting international cooperation from principle alignment toward operational alignment. Currently, many global AI governance efforts converge at the level of high-level principles, while diverging significantly in implementation. This creates fragmentation across regions, sectors, and institutions, making governance difficult to operationalize in practice. The Dialogue can address this gap in three ways: 1. Creating a shared layer of operational governance Beyond principles, the Dialogue should facilitate convergence on how governance is implemented in practice — including approaches to infrastructure, procurement, data governance, and system design. 2. Enabling interoperability between governance frameworks Rather than enforcing uniformity, the Dialogue can support compatibility between different regulatory and institutional approaches, allowing systems to interact without creating governance gaps or inconsistencies. 3. Making implicit choices visible at a global level The Dialogue provides a unique space to surface and compare foundational choices regarding AI infrastructure, dependencies, and models of control (commercial, public, commons-based). Making these choices explicit is a prerequisite for meaningful cooperation. Ultimately, the value of the AI Dialogue lies not in producing additional principles, but in enabling coordination at the level where governance actually takes place. This would allow international cooperation to move from alignment in intent to alignment in practice.

The AI Dialogue should build on existing global and regional initiatives, including the OECD AI Principles, UNESCO Recommendation on the Ethics of AI, the EU AI Act, and emerging sectoral and national governance frameworks. In addition, initiatives focused on public digital infrastructure and open ecosystems provide important reference points for alternative governance models. These efforts have established a strong foundation at the level of principles, risk frameworks, and regulatory approaches. However, they remain fragmented in their implementation and often operate in parallel rather than in coordination. The added value of the AI Dialogue lies in connecting these initiatives at the level where governance becomes operational. Specifically, the Dialogue can: * Bridge principle and practice by facilitating translation of existing frameworks into concrete implementation mechanisms (e.g. procurement standards, infrastructure choices, governance processes) * Enable interoperability between regulatory and institutional approaches, reducing fragmentation across regions and sectors * Surface and compare systemic design choices, particularly regarding AI infrastructure, dependencies, and models of control (commercial, public, commons-based) While existing initiatives define what responsible AI should achieve, the AI Dialogue can focus on how these objectives are realized in practice, across different contexts. In doing so, it can prevent duplication, strengthen coherence, and accelerate the transition from fragmented governance efforts toward coordinated, system-level implementation. This perspective is grounded in practical governance work in higher education, where the gap between AI adoption and institutional control is becoming structurally visible.

Inclusive participation requires more than broad representation; it requires structures that ensure contributions meaningfully shape outcomes. Different stakeholders contribute distinct forms of knowledge: governments provide regulatory authority, the private sector brings technical capacity, academia contributes critical analysis, and public institutions offer insights from implementation. The Dialogue should be designed to integrate these perspectives at the level of decision-making, not only consultation. To achieve this, the format and structure should include: 1. Structured pathways from input to outcome Contributions should be transparently linked to outputs, with clear mechanisms showing how inputs inform recommendations, frameworks, or follow-up actions. 2. Thematic working groups focused on implementation Rather than general discussion, smaller groups should focus on operational challenges (e.g. infrastructure governance, procurement, interoperability), allowing stakeholders to contribute concretely from their domain. 3. Iterative dialogue cycles Participation should not be limited to single events, but organized as ongoing cycles where stakeholders can refine and respond to evolving proposals. 4. Recognition of contribution and authorship To ensure sustained engagement, contributions should be acknowledged and traceable, preventing the loss of perspective that often occurs in aggregated processes. Inclusive participation is therefore not only about who is present, but about how contributions are structured, integrated, and recognized within the governance process.

Global discussions on AI governance often include governments, large technology companies, and academic experts. However, a critical set of perspectives remains underrepresented: those responsible for implementing AI within institutions. This includes professionals in public sector organizations, education, healthcare, and other domains where AI systems are adopted in practice. These actors operate at the intersection of policy, technology, and organizational reality, and are directly confronted with the challenges of translating principles into implementation. Their absence creates a structural gap. Governance frameworks are often designed at a high level, without sufficient grounding in how decisions around infrastructure, procurement, data use, and system integration are actually made within institutions. In addition, perspectives focused on institutional design and governance architecture - rather than on individual applications or ethical principles - remain limited in current discussions. To address this, inclusion should go beyond representation and focus on structural integration: * Incorporate implementation-level expertise through dedicated working groups and case-based contributions * Create feedback loops from practice to policy, ensuring that governance frameworks evolve based on real-world constraints and experiences * Recognize institutional actors as governance stakeholders, not only as adopters of externally defined systems Broadening participation in this way would strengthen the link between global governance efforts and their practical realization, and help ensure that AI governance is not only defined, but also implementable.

Meaningful engagement requires moving beyond traditional panel discussions toward formats that actively surface trade-offs, decisions, and implementation challenges. Three formats could significantly strengthen the AI Dialogue: 1. Decision labs focused on real governance dilemmas Participants work on concrete scenarios (e.g. infrastructure choices, procurement decisions, data governance trade-offs) and are required to make explicit decisions. This shifts the dialogue from abstract principles to operational choices. 2. Comparative governance mapping sessions Stakeholders present how similar challenges are addressed in different regions or sectors, making implicit assumptions and design choices visible. This enables learning not only from outcomes, but from underlying governance models. 3. Iterative co-design processes Instead of one-time exchanges, structured cycles allow participants to refine proposals over time, incorporating feedback from different stakeholder groups and real-world constraints. Across all formats, two principles are essential: 1) Traceability of contributions: participants should be able to see how their input influences outcomes 2) Focus on implementation: discussions should consistently connect to how governance is enacted in practice Innovative engagement is therefore not about increasing participation alone, but about designing formats that make decision-making explicit, comparable, and actionable.

Several existing frameworks and initiatives provide valuable foundations for effective AI governance. At the policy level, the OECD AI Principles, UNESCO Recommendation on the Ethics of AI, and the EU AI Act offer important structures for risk-based regulation, ethical alignment, and accountability. In practice, these are complemented by emerging institutional approaches such as AI impact assessments (e.g. DPIA/FRIA), governance frameworks within public sector organizations, and sector-specific implementations. In addition, initiatives around public digital infrastructure and open ecosystems illustrate alternative models that aim to reduce dependency on commercial providers and strengthen public control. However, a key challenge remains: many of these approaches operate effectively at the level of principles, compliance, or individual tools, but are less developed at the level of integrated system governance. Promising practices therefore include: * Linking governance to procurement and architecture: embedding requirements for transparency, accountability, and control directly into system selection and design processes * Institutional governance frameworks that define roles, responsibilities, and decision-making structures across the lifecycle of AI systems * Iterative assessment mechanisms that move beyond one-time compliance (e.g. DPIA) toward continuous monitoring and adaptation The most effective approaches are those that connect policy, infrastructure, and implementation. Strengthening AI governance therefore requires not only better frameworks, but better integration - ensuring that principles, technical design, and institutional practices are aligned and mutually reinforcing.