Soochow University, Taiwan

A successful first Global Dialogue should produce three concrete outcomes. First, it should establish a shared understanding that AI governance requires not only high-level principles, but also practical and auditable mechanisms for transparency and accountability. Second, it should identify a small set of interoperable governance priorities that can be adapted across jurisdictions while remaining sensitive to different legal, linguistic, cultural, and developmental contexts. Third, it should launch a roadmap for inclusive multistakeholder cooperation, including researchers, governments, industry, and civil society, to develop measurable governance tools for real-world AI systems. In particular, success would mean recognizing that many AI systems now shape public knowledge through answer-first interfaces, where visibility, sourcing, and representation are actively mediated by the system. In such settings, governance must include methods for external auditing of how sources are selected, cited, and prioritized. The Dialogue would be especially valuable if it encourages practical metrics and evaluation frameworks that help detect opacity, source concentration, and representational imbalance across languages and regions.

• Safe, secure and trustworthy AI
• Transparency, accountability, and human oversight
• Protection and promotion of human rights
• Interoperability of governance approaches

These priorities are closely connected. Safe, secure and trustworthy AI cannot be achieved without transparency and accountability. When AI systems generate answers, summaries, or recommendations, users need meaningful ways to examine how those outputs were constructed and which sources were made visible. Human oversight is therefore not only about intervention after harm occurs, but also about enabling external review of how systems allocate attention, credibility, and representation. Protection and promotion of human rights is equally important because opaque source selection can affect access to information, equality, non-discrimination, and fair representation, especially across underrepresented languages, regions, and communities. Governance should therefore consider not only model safety, but also whether AI systems systematically privilege some sources while marginalizing others. Interoperability of governance approaches matters because AI systems operate across borders, while governance remains fragmented. Shared audit principles, common transparency expectations, and comparable evaluation tools would help different jurisdictions coordinate without requiring a single uniform regulatory model. This would support practical international cooperation while allowing contextual flexibility.

One important cross-cutting issue is the governance of answer-first AI interfaces. Many current discussions focus on model development, training data, or general safety, but less attention is given to how deployed AI systems reorganize visibility at the interface level. When AI systems provide direct answers with selected citations, they do not simply retrieve information neutrally; they actively shape which sources become visible, trusted, and influential. This creates a need for auditable metrics that examine source allocation, citation diversity, ranking divergence, and representational balance across languages, regions, and domains. Such tools are important because the presence of citations alone does not guarantee accountability, fairness, or pluralism. A system may appear transparent while still concentrating visibility in ways that are difficult for users to detect. For this reason, the Dialogue should consider external auditability of source selection and citation practices as an emerging governance issue. This would connect transparency, human rights, trustworthiness, and interoperability in a practical way.

For Taiwan's digital information and research sectors, the main development is a shift from broad AI promotion to early-stage institutionalization of AI governance. Taiwan's Artificial Intelligence Fundamental Act was passed in December 2025, and the government's Ten AI Initiatives Promotion Plan was approved in January 2026. Together, these moves create important opportunities: stronger public investment, clearer governance principles, support for AI adoption across industries, and a stronger foundation for trustworthy and human-centered AI development. The most significant challenge, however, is the gap between framework-level principles and operational governance. Taiwan now has high-level commitments on transparency, explainability, fairness, accountability, digital equality, and safety, but many sector-specific rules, risk-classification practices, audit procedures, and enforcement arrangements are still being developed. Taiwan's National Human Rights Commission has also highlighted unresolved issues in oversight, remedies, compensation, and coordination between AI governance and data-protection institutions. For my sector, this matters especially in answer-first AI systems such as generative search, where governance gaps are not only about model development, but also about how systems allocate visibility, citations, and representational authority. The challenge is that users may see citations and assume accountability, even when source selection remains opaque. The opportunity is that Taiwan can help advance practical governance tools, including external audit metrics, multilingual evaluation methods, and interoperable transparency standards that are relevant both domestically and internationally. This would allow Taiwan not only to adopt AI, but also to contribute meaningfully to global AI governance.

The AI Dialogue can play a unique role by providing an inclusive UN platform where governments, researchers, companies, and civil society can discuss AI governance problems that no country can solve alone. The UN describes the Global Dialogue as a universal and inclusive forum, and it is also designed to receive input from the Independent International Scientific Panel on AI, which can help ground political discussion in scientific evidence. Its most important contribution would be to move international cooperation beyond broad principles toward practical coordination. This includes identifying a common vocabulary, shared risk priorities, and interoperable governance approaches that different jurisdictions can adapt without requiring a single global regulatory model. The Dialogue can also help connect fragmented national and regional initiatives, reducing duplication and making governance more comparable across borders. For rapidly deployed AI systems, especially answer-first and generative interfaces, the Dialogue can be valuable by encouraging common expectations for transparency, accountability, and external auditability. International cooperation is especially needed where AI systems affect information access, public trust, and representation across languages and regions. In this sense, the Dialogue should not only be a discussion space, but also a coordination mechanism that supports shared evaluation methods, evidence-based policymaking, and capacity building for countries with fewer technical and regulatory resources.

The AI Dialogue should build on, and connect, existing efforts rather than duplicate them. At the UN level, it should be anchored in the Global Digital Compact and work closely with the Independent International Scientific Panel on AI. Beyond the UN, it should connect with UNESCO's Recommendation on the Ethics of AI, which applies to all 194 UNESCO Member States; the OECD AI Principles and the OECD.AI/GPAI ecosystem, which already provide policy guidance, data, and tools; the Internet Governance Forum, especially its Policy Network on AI, for bottom-up multistakeholder input; and the Council of Europe's AI Framework Convention, the first legally binding international treaty in this field. The added value of the AI Dialogue is its universal UN legitimacy and convening power. Other initiatives already contribute important principles, evidence, standards, and regional legal innovations, but they remain institutionally fragmented. The Dialogue can serve as the place where these streams are translated into a more coherent global conversation, especially by identifying areas of convergence, clarifying governance gaps, and elevating perspectives from countries and communities that are often underrepresented in technical or regulatory forums. The UN describes the Dialogue as complementary to existing forums, particularly the IGF, and as linked to the Scientific Panel. Its practical contribution should be to connect principles to implementation: interoperable transparency expectations, comparable audit methods, capacity-building support, and shared governance vocabulary across jurisdictions. In this sense, the Dialogue should function as a coordination layer that helps turn scattered initiatives into more inclusive and actionable international cooperation.

Different stakeholders should contribute in distinct but connected ways. Governments can share regulatory experiences, institutional constraints, and public-interest priorities. Industry can provide technical implementation knowledge, deployment challenges, and information about existing safeguards. Researchers can contribute independent evidence, evaluation methods, and comparative analysis. Civil society, journalists, educators, and affected communities can identify real-world harms, gaps in access, and representation concerns that may be overlooked in high-level policy discussions. To make these contributions meaningful, the AI Dialogue should be structured in layers. A plenary format is useful for identifying shared priorities, but it should be complemented by smaller thematic working sessions where stakeholders can engage in more practical exchanges. Each session should combine principle-level discussion with evidence-based case examples and implementation questions. The Dialogue would also benefit from written submissions, regional consultations, and virtual participation channels before and after the in-person meeting, so participation is not limited to those with travel capacity. A useful format would include: a high-level opening on shared governance priorities; thematic tracks on issues such as transparency, human rights, safety, and interoperability; multistakeholder roundtables focused on concrete governance gaps; and a closing synthesis that identifies common recommendations, unresolved tensions, and next-step workstreams. To strengthen continuity, the Dialogue should also create a mechanism for follow-up, such as annual reporting, thematic task groups, or a public repository of emerging practices and governance tools.

Global AI governance discussions still underrepresent many of the communities most affected by AI systems. These include voices from the Global South, small and medium-sized economies, low-resource language communities, Indigenous communities, disability advocates, educators, journalists, public-interest researchers, and workers whose labor is being reorganized by AI systems. Underrepresentation also affects sectors outside major technology hubs, including local media, public services, and smaller academic communities that may lack institutional access to global policy forums. One especially important gap concerns communities whose knowledge is visible to AI systems only in limited or distorted ways. When governance discussions are dominated by actors from major platforms, large economies, and high-resource languages, questions of representation, source visibility, and epistemic inequality may receive less attention than they deserve. Inclusion should therefore be designed proactively rather than assumed. The AI Dialogue should support multilingual participation, travel support, hybrid access, and regional pre-consultations. It should also reserve speaking opportunities for underrepresented stakeholders, not only invite open submissions. Civil society organizations, independent researchers, and community-based groups should be able to submit evidence in accessible formats, including short written inputs, case studies, and regional testimonies. Partnerships with universities, media organizations, and local digital rights groups could also help surface perspectives that are often absent from global forums.

The AI Dialogue would benefit from engagement formats that move beyond formal statements and allow stakeholders to work through real governance problems together. One useful approach would be scenario-based workshops, where participants examine concrete cases involving generative AI, public-sector deployment, cross-border data issues, or transparency failures. This would help connect abstract principles to operational challenges. Another effective format would be evidence clinics or audit labs, where researchers, civil society groups, and technical experts present practical evaluation tools, empirical findings, or case-based governance lessons in a short and accessible format. This would be especially valuable for emerging issues such as answer-first AI systems, source visibility, external auditability, and representational imbalance. Such sessions could help participants compare methods and identify shared governance needs. Structured multistakeholder roundtables would also be useful if they are designed around specific questions and produce short written outputs. Rather than open-ended discussion alone, each roundtable could be asked to identify one governance gap, one promising practice, and one area requiring international cooperation. Interactive digital participation tools could further support remote engagement, including live multilingual input channels, moderated Q&A, and post-session feedback collection. Finally, the Dialogue should include a synthesis mechanism that captures not only areas of consensus but also unresolved disagreements. Dynamic engagement is most meaningful when participants can see how their input is reflected in next steps. For that reason, innovative formats should be paired with visible documentation, follow-up workstreams, and opportunities for continued participation after the meeting ends.

Examples already exist across several levels of AI governance. UNESCO's Recommendation on the Ethics of AI provides a global normative baseline centered on human rights, dignity, transparency, fairness, and human oversight, and it applies to all 194 UNESCO Member States. The OECD AI Principles and the OECD.AI platform provide practical policy guidance, shared terminology, and comparative information on national AI initiatives. NIST's AI Risk Management Framework offers an operational approach for managing AI risks across the lifecycle, helping organizations translate trustworthiness into governance, measurement, and monitoring practices. The Council of Europe Framework Convention on AI and Human Rights, Democracy and the Rule of Law is also important because it connects AI governance to legal accountability and a risk-based approach. The AI Dialogue should build on these examples by encouraging implementation-oriented practices that are comparable across borders. In particular, effective governance should not rely only on high-level principles. It should also support practical mechanisms such as documentation, post-deployment monitoring, external evaluation, and independent auditing of deployed systems. This is especially important for answer-first and generative AI systems, where governance challenges often appear at the interface level, including source selection, visibility allocation, and representational imbalance. In this area, the Dialogue could add value by promoting interoperable audit methods and evidence-based governance tools that help make transparency and accountability more operational in real-world systems. This final point is my inference from the frameworks above and how they can be applied to deployed AI systems.