The George Washington University Regulatory Studies Center

Success means the Dialogue establishes mechanisms that translate good intentions into actual practice. Three outcomes matter most. First, procedural clarity on how underrepresented perspectives actually reach the agenda. "Inclusivity" is a stated goal, but without structured processes, louder voices dominate. The first session should establish clear pathways—for instance, by guaranteeing discussion time for themes that haven't yet appeared in most national strategies in the region. This isn't just fairness; it's good governance. My preliminary analysis comparing AI governance literature to national strategies suggests that themes prominent in academic discourse are often absent from government plans. If the Dialogue doesn't catch these gaps, we're building global governance on an incomplete foundation. Second, accountability mechanisms linking gap identification to implementation. Many international forums identify problems but lack follow-through. Success means states making documented voluntary commitments and reporting back. The IPCC model works because it links scientific synthesis to policy deliberation with the transparent disposition of findings. The Dialogue needs similar connective tissue. Third, managing vocabulary convergence without suppressing innovation. My research shows EU AI Act terminology becoming a de facto standard in some non-EU jurisdictions—initial findings suggest countries like Zimbabwe, Kenya, and India show alignment despite no binding obligation, while others like China and Australia show different patterns. This isn't inherently bad, but it needs to be deliberate, not adoption-by-default. Success means the Dialogue tracks this convergence and ensures states understand they're choosing frameworks, not inheriting them. The July session won't solve everything, but it should establish whether the Dialogue is a talk shop or a governance mechanism. Procedural decisions made now determine whether this becomes the institution AI governance needs.

• Interoperability of governance approaches
• Transparency, accountability, and human oversight
• Protection and promotion of human rights
• Social, economic, ethical, cultural, linguistic and technical implications of AI

I selected these priorities because they're where the science-policy gap is widest and where institutional design matters most. Interoperability is urgent because we're watching regulatory fragmentation in real time. My preliminary research comparing national AI strategies shows vocabulary convergence without coordination-some countries align primarily with EU AI Act terminology despite no legal obligation, while others show minimal alignment. This creates implementation chaos: companies face conflicting compliance regimes, researchers can't compare results across jurisdictions, and smaller countries adopt frameworks they don't fully understand. We need systematic monitoring of how governance approaches interact, not just descriptions of what each country is doing in isolation. Transparency and accountability determine whether the Dialogue produces theater or governance. Some research shows how algorithmic systems in federal agencies avoid mandatory oversight through vague impact assessments and the omission of disparity testing. International governance faces the same risk on a larger scale. Without transparent processes for how decisions are made and documented, and for follow-through on commitments, the Dialogue becomes another forum where problems are named but not solved. Human rights protection isn't abstract-it's about who gets heard. My analysis comparing academic AI governance literature with national strategies suggests that governance themes prominent in research don't consistently appear in government policy. Whose concerns are making it from research to policy? Whose aren't? Rights protection requires mechanisms that ensure marginalized perspectives reach decision-makers, not just statements that value inclusion. Social and technical implications intersect constantly in my work on algorithmic governance. You can't address bias in name-matching algorithms without understanding both the technical constraints of phonetic matching and the social context of differential error rates across demographic groups. Governance that treats these as separate domains fails. These priorities share a thread: institutional design determines who participates, what gets considered, and whether outcomes are enforceable.

Yes-the science-policy translation infrastructure is missing, and it's cross-cutting because it affects how all the listed themes get operationalized. The themes identify priorities, but they don't address how the Dialogue learns what's working and what isn't. Through policy analysis work and coursework examining federal AI deployments, I've noticed a pattern: systems get deployed without documented disparity testing, agencies claim they're doing "AI risk management," and when problems emerge, there's no institutional memory of what went wrong or systematic correction. The Dialogue risks repeating this at a global scale-lots of principles, weak feedback loops. Here's what's absent: mechanisms that detect when important governance concerns aren't reaching national policy agendas, then trigger structured responses. Not ad hoc consultations when someone notices a problem, but a systematic diagnostic infrastructure that runs continuously. Think of it like public health surveillance-we don't wait for doctors to individually notice disease clusters; we have systems that flag patterns and trigger coordinated responses. This matters because governance priorities change faster than formal treaty processes. Two years ago, "AI alignment" meant model behavior matching human values. Now it increasingly means geopolitical and regulatory alignment between jurisdictions. Foundation models went from research curiosities to infrastructure in 18 months. The Dialogue needs sensing mechanisms that catch these shifts before they're crises. Commentary work on AI governance suggests what happens without this: the same stakeholders dominate every consultation, the same concerns get raised repeatedly without resolution, and novel issues get ignored until they become scandals. International governance isn't immune to these patterns-it amplifies them because participation barriers are higher and accountability is even lower. The Dialogue should establish learning infrastructure alongside thematic discussions. Otherwise, we'll reconvene in 2027 having made the same mistakes, just on a larger scale.

The governance gaps I identified hit differently depending on where you sit in the global AI ecosystem, and I've seen this from multiple angles. Zimbabwe's recently launched National AI Strategy (2026-2030) illustrates this clearly. Challenge: Capacity constraints in framework development. Zimbabwe's strategy emphasizes building "domestic capacities" and warns that "international cooperation... cannot replace the development of domestic capacities." The document seeks partnerships with "AI Sprinter" nations for joint research but acknowledges Zimbabwe needs to develop indigenous AI governance—not just adopt external frameworks wholesale. Yet most African countries lack the technical teams and funding to build comprehensive frameworks from scratch, creating pressure to reference existing detailed models even when context doesn't align. Challenge: Language and knowledge access barriers. Zimbabwe's strategy was developed through collaborative processes anchored in UNESCO's AI Ethics Recommendation—requiring translation of international governance concepts into the local context. But most detailed AI governance documentation exists primarily in English, and even within English, regulatory terminology developed in Brussels or Silicon Valley doesn't map cleanly to Harare's priorities. The strategy emphasizes Ubuntu/Unhu philosophy, but there's no systematic way to identify when African ethical frameworks are absent from global governance discussions. Opportunity: Regional coordination infrastructure. The strategy positions Zimbabwe as seeking to be "a respected partner within the global AI ecosystem... advocating for an equitable AI order that serves the interests of the Global South." The African Union has started coordinating member states' AI policies, but lacks systematic mechanisms to track which governance approaches work where, or to identify themes emerging in one region that haven't reached others. Opportunity: Focused sectoral adoption. Zimbabwe prioritizes AI in agriculture, mining, health, and education—sectors where context-specific governance matters more than general frameworks. This creates space for innovative approaches if there's infrastructure to document and share what works. The core gap isn't technical knowledge—it's the diagnostic infrastructure that ensures governance develops from evidence of actual needs, not just the availability of frameworks.

The Dialogue's unique role is serving as the one forum where governments, civil society, and technical communities meet on equal footing—but that structural advantage only matters if there's infrastructure to ensure that participation translates into governance outcomes. Three roles the Dialogue should play: First, systematic gap identification. International cooperation fails when countries think they're coordinating on shared priorities, but they're actually working from different problem definitions. The Dialogue should establish mechanisms that continuously compare what's being discussed in technical literature and civil society submissions with what appears in national strategies. Not one-time assessments, but ongoing monitoring that flags when governance themes prominent in one community aren't reaching another. This catches divergence before it becomes a crisis. Second, structured consultation triggers. Identifying gaps isn't enough—there needs to be a process for addressing them. When monitoring reveals that a theme appears in fewer than a specified threshold of national strategies within a regional group, the Dialogue should convene targeted consultations with affected regions before the next session. Not ad hoc, not when someone notices a problem, but systematically triggered by evidence. Third, accountability without enforcement. The UN can't force states to adopt governance frameworks, but it can create transparency about what states commit to and whether they follow through. The Dialogue should document the voluntary commitments that states make during sessions and track their implementation. This isn't naming and shaming—it's institutional memory that prevents the same gaps from recurring. The IPCC model shows this works: synthesize scientific literature, identify policy-relevant gaps, convene structured consultations with policymakers, document findings transparently. The Dialogue should adapt this for AI governance—not waiting for governments to ask what's missing, but proactively identifying gaps and ensuring they receive structured consideration. International cooperation advances when institutions convert good intentions into replicable processes. The Dialogue's role is to build that institutional infrastructure.

The Dialogue should build on three existing mechanisms while filling a critical gap that none of them address. Build on the OECD AI Policy Observatory: Already tracks 1,000+ AI policy initiatives across 70+ countries through country- and sector-specific dashboards. The Observatory demonstrates that large-scale policy monitoring is operationally feasible. What it doesn't do: systematically compare governance themes in academic/civil society discourse to what's being adopted in national strategies. It tracks what countries are doing, not what they're not doing. Build on the UNESCO AI Ethics Recommendation: Provides a normative framework emphasizing human rights, inclusion, and diversity. UNESCO's AI Readiness Assessment Methodology helps countries evaluate capacity. What it doesn't provide: ongoing mechanisms that identify when themes from the Ethics Recommendation aren't appearing in national implementation, or trigger consultations when gaps emerge. Build on IPCC assessment model: Synthesizes peer-reviewed literature, identifies policy-relevant findings, engages policymakers through structured consultations. This model proves science-policy translation infrastructure works at scale. What makes AI governance different: the governance landscape changes faster than climate science, requiring more frequent gap monitoring. The Dialogue's added value: Establish a Governance Gap Monitor (GGM) that combines these approaches—OECD's technical infrastructure for policy tracking, UNESCO's normative framework for inclusion, IPCC's structured consultation model—but adds what's missing: systematic comparative analysis identifying governance themes prominent in literature/civil society that aren't reaching national strategies, with automatic triggers for regional consultations when adoption falls below meaningful thresholds. Existing initiatives excel at describing the current state. None systematically identify and address gaps. The Dialogue can be the forum that doesn't just document what countries are doing, but ensures that what they're not doing gets structured attention before it becomes a crisis. That's the institutional infrastructure international AI governance currently lacks.

Stakeholder contributions need to move beyond open comment periods, where whoever shows up gets heard. The Dialogue should establish roles tied to evidence-based gap identification. Academic/research community: Submit findings to a continuously updated repository analyzed using text mining to identify governance themes. Not selective literature reviews where conveners choose which research matters, but systematic analysis of peer-reviewed publications. Researchers contribute by publishing; the Dialogue's infrastructure identifies patterns. Civil society organizations: Submit governance concerns through a structured template that tags themes by sector, region, and affected populations. These submissions feed into the same text analysis, identifying whether concerns raised by civil society are appearing in national strategies. Civil society contributes advocacy; the infrastructure tracks whether it's being heard. National governments: Submit AI strategies, policy documents, and regulatory frameworks to a shared repository. Not voluntary disclosure of selected highlights, but systematic collection enabling comparative analysis. Governments contribute policy; the infrastructure identifies where approaches converge or diverge. Private sector: Participate in consultations triggered when gap analysis identifies themes affecting their operations. Tech companies often complain they're excluded from policy discussions until decisions are made. Gap-triggered consultations bring them in when their expertise matters, not just when governments invite them. Format recommendation: The Dialogue should operate on two tracks. Track 1: Annual plenary sessions following current format—statements, panels, negotiations. Track 2: Continuous between-session work analyzing submissions, identifying gaps, convening targeted consultations with affected stakeholders when gaps are detected. Track 2 feeds Track 1, ensuring plenary discussions address evidence-identified priorities rather than whoever secured speaking slots. This requires the Dialogue to establish what I've been calling a Governance Gap Monitor (GGM)—infrastructure that doesn't replace existing stakeholder engagement but makes it systematic rather than ad hoc.

Three categories of underrepresentation, each requiring different solutions: Linguistically marginalized communities. AI governance discussions happen primarily in English. French and Spanish get some representation; Arabic, Mandarin, Swahili, and Hindi barely register. The problem isn't translation—UNESCO can translate documents. The problem is that governance concepts developed in one language don't map cleanly to others. "Algorithmic accountability" has a specific meaning in Anglo-American legal tradition; forcing that terminology into Arabic or Swahili governance discussions predetermines the framing. Solution: Regional consultations should operate in regional languages, with synthesis happening afterward rather than requiring everyone to engage through English first. Capacity-constrained governments. Small states and developing countries can't staff multilateral processes as effectively as G7 nations can. When the Dialogue publishes a discussion paper, US/EU/China delegations have policy teams to analyze and respond. Zimbabwe might have one person tracking AI governance part-time while handling ten other portfolios. By the time they read the paper, the discussion has moved on. Solution: When gap monitoring identifies themes absent from a region's strategies, convene consultations that bring expertise TO governments rather than expecting governments to come TO Geneva prepared. Flip the burden. Affected populations without formal representation. Gig workers experiencing algorithmic management, content moderators facing AI surveillance, farmers adopting AI-driven agriculture—they're governed by AI but rarely at governance tables. Civil society organizations claim to represent them, but often don't. Solution: Gap-triggered consultations should include requirements for participation by the affected population, not just by civil society intermediaries. If analysis shows that "agricultural AI governance" is absent from African strategies, consultations must include actual farmers, not just NGOs speaking on their behalf. The common thread: make inclusion systematic through infrastructure, not aspirational through statements.

Three format innovations that would actually change participation dynamics: Pre-plenary gap reports with guaranteed response time. Four months before each Dialogue session, publish an analysis identifying governance themes that appear in academic literature or civil society submissions but are absent from national strategies, disaggregated by UN regional group. Guarantee 45 minutes of plenary discussion for any theme below the minimum regional adoption threshold. This isn't innovation for innovation's sake—it's a structural guarantee that gaps trigger consideration rather than hoping someone raises them. Regional pre-consultations with documentation requirements. When gap reports identify underrepresented themes, convene month-long regional consultations before the plenary. Not just virtual roundtables where whoever speaks loudest wins. Structured consultations producing draft governance language, with documentation showing who participated and what positions were taken. Co-Chairs must report to plenary on the disposition of draft language—adopted, modified, deferred—with explanations. Creates an accountability loop missing from typical UN processes. Working sessions instead of panel performances. UN plenaries default to statements and panels where people talk AT each other. Allocate plenary time to working sessions where delegations actually negotiate language on gap-identified themes. The IGF model shows this works—when you give people real work to do instead of listening to speeches, participation becomes meaningful. Example: If a gap analysis shows that most African strategies lack AI labor governance frameworks, convene a working session in which African delegations draft model language with input from the ILO/academic/, and civil society. Public dashboard of commitments and implementation. States make voluntary commitments during Dialogue. Establish a public dashboard tracking what was committed and what's implemented. Not naming and shaming, but transparency about follow-through. Creates reputational incentives for meaningful participation. The innovation isn't formats—it's making formats serve evidence-based priorities rather than procedural inertia.

I recommend establishing a Governance Gap Monitor (GGM) within the AI Dialogue-a systematic infrastructure that operationalizes the inclusivity and accountability principles I've outlined throughout this submission. What it does: The GGM would perform three functions continuously between Dialogue sessions: First, comparative text analysis using natural language processing (NLP) to identify governance themes appearing in academic literature and civil society submissions but absent from national AI strategies. Analysis disaggregated by UN regional group, published quarterly as public dashboards showing where governance discourse and policy implementation diverge. Second, gap-triggered consultations with automatic activation rules. When a theme appears in fewer than 20% of national strategies within a regional group (with room for sensitivity analysis), the mechanism convenes structured four-month consultations involving affected states, technical experts, and civil society to produce draft governance language. Outputs receive guaranteed plenary discussion time at subsequent Dialogue sessions. Third, vocabulary diversity tracking measuring whether regulatory terminology convergence (like EU AI Act language appearing in non-EU jurisdictions) reflects deliberate alignment or adoption-by-default due to limited alternatives in accessible languages. Implementation models it builds on: OECD AI Policy Observatory demonstrates large-scale policy tracking is feasible (1,000+ initiatives, 70+ countries). IPCC assessment cycles prove that systematic literature synthesis can inform policy deliberations. UNESCO's AI Readiness Assessment shows structured evaluations work. The GGM combines these approaches while adding what's missing: systematic gap identification with procedural guarantees that gaps trigger responses. I have developed a detailed policy recommendation with full implementation specifications (budget estimates, timeline, success metrics, institutional placement) and would be pleased to provide it to the Co-Chairs upon request. This isn't aspirational-it's operational infrastructure ensuring the Dialogue translates stated commitments to inclusivity into measurable governance outcomes.