Harvard University

The first Global Dialogue on AI Governance represents a historic opportunity — and its success, I believe, should be understood in both immediate and longer-term terms. In the near term, success would mean creating genuine space for multistakeholder dialogue: one where governments, industry, civil society, academia, and affected communities do not merely speak past one another, but begin to build a shared vocabulary around the values and risks at stake in AI development. From my research at Harvard and my advisory work with institutions such as the United Nations and the European Commission, I have observed that the most durable governance frameworks emerge not from top-down mandates, but from processes in which diverse perspectives are heard and meaningfully integrated. Concretely, I would hope the Dialogue produces three tangible outcomes. First, the identification of a foundational set of shared principles — around transparency, accountability, and human oversight — that can serve as a common reference point across different legal and cultural contexts. Second, the launch of structured working groups with clear mandates, timelines, and cross-sectoral representation, ensuring that momentum from the Dialogue translates into ongoing collaboration rather than dissipating after the closing session. Third, a commitment to mapping existing governance initiatives globally, so that future efforts build on what already works rather than reinventing the wheel. Equally important is the question of inclusion. AI governance that does not reflect the realities of the Global South, of marginalized communities, or of workers navigating rapid technological change will struggle to earn legitimacy — and legitimacy is the foundation of effective governance. Ultimately, I would consider this Dialogue a success if it leaves participants with greater mutual understanding, a clearer map of the work ahead, and concrete next steps that hold us collectively accountable for ensuring AI serves human flourishing broadly, not narrowly.

• Transparency, accountability, and human oversight
• Social, economic, ethical, cultural, linguistic and technical implications of AI

My selections reflect both the empirical realities I observe in my research and advisory work, and a deep conviction - rooted in my legal training - that effective AI governance must be grounded in the rule of law, not left to voluntary commitments alone. I prioritize social, economic, ethical, cultural, linguistic and technical implications of AI because these are the lived consequences that governance must ultimately address. My work on behavioral science and human capital has shown that AI systems do not operate in a vacuum - they reshape labor markets, amplify or mitigate bias, and interact with cultural contexts in ways that are often underestimated by technologists and policymakers alike. Safe, secure and trustworthy AI is inseparable from this. Trust is not a soft metric; it is the precondition for adoption, legitimacy, and long-term societal benefit. Without robust safety standards, AI risks eroding the public confidence that governance institutions depend upon. Transparency and accountability speak directly to my legal background. Accountability without enforceable mechanisms is aspiration, not governance. Legal frameworks have long grappled with how to assign responsibility in complex systems - AI governance must draw on those traditions rather than treating itself as entirely novel territory. Human oversight is not a technical feature; it is a legal and ethical imperative. Finally, interoperability of governance approaches recognizes a practical reality: we will not achieve a single global regulatory framework, nor should we necessarily aim for one. What we need are governance architectures that can speak to one another across jurisdictions - coherent without being uniform, coordinated without being imposed. Together, these four areas form the foundation on which legitimate, durable, and effective AI governance can be built.

Yes. While the thematic areas identified in Resolution 79/325 provide a strong foundation, several cross-cutting issues merit explicit attention if the Dialogue is to remain fit for purpose as AI capabilities continue to evolve rapidly. The first is algorithmic accountability in institutional decision-making. As AI systems are increasingly deployed by governments, courts, employers, and financial institutions, the question of how existing legal doctrines - due process, equal protection, administrative accountability - apply to automated decisions remains critically underexplored. This is not merely a technical or ethical question; it is a legal one, and governance frameworks must treat it as such. The second is the concentration of AI power. The current landscape is characterized by a small number of actors - predominantly private, predominantly located in a handful of countries - controlling the foundational infrastructure of AI development. This concentration poses structural risks to pluralism, sovereignty, and democratic governance that cut across virtually every thematic area identified in the resolution, yet none addresses it directly. Third, I would highlight the governance of AI in the workplace. The future of work is being reshaped at a pace that existing labor law frameworks were not designed to accommodate. Questions around algorithmic management, automated performance evaluation, and workforce displacement require dedicated attention - not as a subset of broader economic implications, but as a domain with its own urgency and specificity. Finally, intergenerational equity deserves recognition as a cross-cutting principle. Decisions made today about AI development and governance will constrain or enable the choices available to future generations. Governance frameworks that do not explicitly account for long-term consequences risk optimizing for present interests at the expense of those who will inherit the systems we build

Operating across academia, law, and applied consulting on multiple continents, I have the privilege of observing both the remarkable opportunities that AI presents and the very real governance challenges that accompany its rapid advancement — challenges that are simultaneously local in their impact and global in their structure. In the legal and governance sector, one of the most significant dynamics is jurisdictional diversity. The EU AI Act, evolving frameworks in the United States, and emerging approaches across the Global South each reflect distinct and legitimate philosophical traditions regarding the relationship between the state, the market, and the individual. This diversity is not inherently problematic — it reflects the richness of different legal cultures and democratic choices. However, without greater efforts toward interoperability, there is a risk that gaps between frameworks are exploited in ways that undermine the protective intent of any individual regime. Strengthening dialogue across legal traditions, while respecting sovereignty, represents one of the most consequential opportunities before us. In academia and research, the central challenge is one of pace and resource. The speed of AI development creates genuine pressure on research institutions to simultaneously train future practitioners, conduct independent critical inquiry, and meaningfully inform policy — often with institutional frameworks that were not designed for this degree of interdisciplinary urgency. Closing this gap requires sustained investment and new models of collaboration between universities, governments, and the private sector. In organizational and workplace contexts, AI is already reshaping how decisions are made about people — in hiring, performance evaluation, and resource allocation. This creates both risks, particularly around fairness and accountability, and genuine opportunities for organizations willing to lead responsibly. Across all three domains, the common thread is that the window to shape norms rather than simply react to them remains open — but it will not remain so indefinitely.

The AI Dialogue arrives at a genuinely consequential moment. International cooperation on AI governance has, until recently, proceeded largely through parallel and often siloed processes — regional regulatory initiatives, bilateral technical agreements, and sector-specific standards — each valuable in its own right, but collectively insufficient to address the systemic, cross-border nature of AI's most significant implications. The Dialogue can play a distinctive and complementary role precisely because of its universal character. Convened under the auspices of the United Nations General Assembly, it carries a legitimacy that no regional framework or industry-led initiative can fully replicate. This legitimacy is not merely symbolic; in international affairs, the authority to convene diverse actors around a shared agenda is itself a form of governance capacity that should not be underestimated. Concretely, I see three areas where the Dialogue can add meaningful value. First, it can serve as a bridge between existing frameworks — mapping areas of convergence across the EU AI Act, national strategies, and emerging Global South initiatives, and identifying where harmonization is both feasible and desirable. Second, it can function as a forum for norm development in areas where no framework yet exists, particularly around the governance of frontier AI systems and the rights of individuals affected by automated decision-making. Third, and perhaps most importantly, it can elevate voices that have been underrepresented in AI governance conversations — smaller nations, civil society organizations, and communities bearing a disproportionate share of AI's risks — ensuring that the resulting norms reflect a genuinely global consensus rather than the preferences of the most powerful actors. In this sense, the Dialogue's greatest contribution may be less about producing a single governance document, and more about building the relationships and trust that durable international cooperation requires

A credible and effective AI Dialogue must begin from an honest recognition of what already exists. The international community has, over recent years, developed a rich — if sometimes fragmented — ecosystem of initiatives, and the Dialogue's added value lies not in replacing these efforts, but in weaving them into a more coherent whole. Several initiatives deserve particular attention. The OECD AI Principles and the work of the Global Partnership on AI have established important normative foundations and technical expertise that should inform the Dialogue's substantive agenda. The UNESCO Recommendation on the Ethics of AI represents a genuinely inclusive multilateral achievement, with near-universal endorsement, and provides a values framework that the Dialogue can build upon rather than duplicate. Within the UN system, the work of the ITU's AI for Good platform — with which I have had the privilege of engaging directly through the Steering Committee on AI and Virtual Worlds — offers both technical grounding and a model for multistakeholder collaboration that the Dialogue would do well to emulate. Regional frameworks, and bilateral initiatives between major AI powers also represent important reference points, not as models to be universally adopted, but as laboratories of governance experimentation from which broader lessons can be drawn. The added value the Dialogue can bring is threefold. First, universality — convening actors that existing initiatives have not fully reached, particularly from the Global South. Second, integration — creating structured linkages between parallel processes that currently operate with insufficient coordination. Third, accountability — establishing review mechanisms that allow the international community to assess progress against shared commitments over time. In this way, the Dialogue can transform a landscape of valuable but disconnected initiatives into something approaching a coherent international governance architecture.

The legitimacy and effectiveness of the AI Dialogue will depend, in no small measure, on whether its structure genuinely reflects the diversity of stakeholders who have a rightful interest in AI governance — and whether those stakeholders are engaged as substantive contributors rather than consultative afterthoughts. Governments will necessarily play a central role, as the primary bearers of regulatory authority and international legal obligations. However, the Dialogue should be deliberately designed to prevent intergovernmental dynamics from crowding out other essential voices. The history of technology governance offers cautionary lessons here: processes that begin as inclusive often become progressively narrower as political sensitivities intensify. To guard against this, I would recommend a tiered but integrated structure. Formal intergovernmental sessions should be complemented by structured multistakeholder forums — with dedicated tracks for civil society, academia, the private sector, and technical experts — whose outputs are formally transmitted to and considered within the intergovernmental process, not merely noted alongside it. This distinction matters: inclusion without influence is not meaningful participation. The private sector, particularly frontier AI developers, bears special responsibilities and should be engaged with both openness and appropriate scrutiny. Their technical knowledge is indispensable; their commercial interests, however legitimate, must be transparently disclosed and carefully managed within the deliberative process. Academia and independent research institutions can play a particularly valuable bridging role — translating technical complexity for policymakers, providing evidence-based assessments of governance proposals, and maintaining the intellectual independence that political and commercial actors cannot always preserve. Finally, the Dialogue should invest seriously in accessible formats — language accessibility, geographic representation, and participation modalities that do not systematically exclude actors from lower-resource contexts. A governance process whose structure contradicts its stated values of inclusion will struggle to produce outcomes that the world can genuinely own.

Ensuring that global AI governance reflects a genuinely diverse range of perspectives is both a matter of fundamental fairness and a prerequisite for producing frameworks that are legitimate, durable, and effective. From my research and advisory work across multiple continents, I have observed several communities whose valuable contributions to this conversation have not yet been fully realized. The first is geographic diversity. The AI governance discourse has naturally evolved in contexts where AI research and development is most concentrated — North America, Europe, and parts of East Asia. Yet the Global South is home to the majority of the world's population and to some of the most innovative applications of AI in areas such as financial inclusion, healthcare delivery, and agricultural development. Creating meaningful pathways for these perspectives to shape the governance agenda — not merely respond to it — would significantly enrich the quality and relevance of outcomes. Workers and labor communities also bring an indispensable perspective. Those navigating the day-to-day realities of AI-mediated workplaces — algorithmic scheduling, automated performance evaluation, changing skill demands — possess experiential knowledge that cannot be fully captured by employers, technologists, or policymakers alone. Structured mechanisms to incorporate their insights would strengthen governance frameworks considerably. Linguistic and cultural communities beyond the dominant international discourse represent another important opportunity. AI systems inevitably reflect the cultural assumptions embedded in their design, and governance frameworks benefit from engaging the full breadth of human experience and value systems. Finally, younger generations — who will live longest with the consequences of today's governance choices — deserve thoughtful and structured inclusion in deliberative processes. Practically, realizing this broader participation will require dedicated investment in capacity-building, multilingual engagement, and deliberative formats designed from the outset with inclusion as a core architectural principle rather than a supplementary consideration.

The design of deliberative processes is itself a governance choice, and one that deserves as much careful attention as the substantive agenda. From my work in behavioral science and organizational decision-making, I have observed that the format of a dialogue profoundly shapes its outcomes — determining not only who participates, but whose perspectives are genuinely heard and integrated. Several innovative engagement formats could meaningfully enhance the AI Dialogue's effectiveness. Structured deliberative sessions that move beyond traditional panel formats — incorporating facilitated small-group dialogue, scenario-based discussion, and collaborative problem-solving exercises — have been shown in behavioral research to produce richer exchanges and more durable shared understanding than large plenary debates alone. When participants work through concrete cases together rather than exchanging prepared positions, the quality of mutual learning increases substantially. Pre-Dialogue consultative processes — regionally organized and linguistically accessible — could ensure that the perspectives of communities unable to attend in person are substantively integrated into the agenda, rather than acknowledged only in principle. Digital participation platforms, thoughtfully designed to complement rather than replace in-person engagement, can extend reach without sacrificing depth. Cross-sectoral working simulations, in which participants from government, industry, civil society, and academia collaborate on shared governance challenges in real time, can build the relational foundations that formal negotiations alone rarely produce. Trust, in my experience, is built through shared problem-solving, not through the exchange of position papers. Dedicated reflection and synthesis sessions — built deliberately into the programme rather than appended at the end — would allow participants to identify emerging areas of convergence and to articulate collectively where further work is needed, creating a living record of progress rather than a static communiqué.

Effective AI governance is not an abstract aspiration - it is already being practiced, in varying degrees and forms, across jurisdictions and sectors. Drawing on my research and advisory experience, several examples stand out as particularly instructive. The EU AI Act represents the most comprehensive attempt to date to establish a risk-based regulatory framework for AI, with enforceable obligations calibrated to the potential harms of different applications. While its implementation will require careful monitoring and adjustment, its methodology - distinguishing between prohibited, high-risk, and lower-risk uses - offers a principled architecture that other jurisdictions can learn from, even where they choose different calibrations. The OECD AI Principles and associated monitoring framework demonstrate the value of combining normative guidance with empirical tracking. By systematically documenting how member and partner countries are implementing shared principles, the OECD has created a comparative learning infrastructure that supports policy coherence without mandating uniformity. Within organizations, I have observed that algorithmic impact assessments - modeled on established environmental and human rights due diligence traditions - represent a practical and scalable governance tool. When rigorously implemented, they create meaningful accountability checkpoints before AI systems are deployed in consequential contexts, and they generate documentation that supports external oversight. The ITU's AI for Good platform offers a valuable model for multistakeholder technical collaboration, bringing together governments, industry, and civil society around concrete problem-solving rather than abstract standard-setting. Its emphasis on practical application in sustainable development contexts is particularly relevant for ensuring that governance frameworks remain grounded in real-world needs. Finally, sector-specific guidance - such as that developed by the US Equal Employment Opportunity Commission on AI in hiring developed by Chair Burrow in 2020- illustrates how existing legal frameworks can be thoughtfully extended to address AI-specific challenges without waiting for comprehensive legislation, providing meaningful near-term protection while broader frameworks mature.