Addis Ababa Science and Technology University

In my opinion, for the first Global Dialogue on AI Governance to be a success, 3 concrete outcomes are essential. 1st, set a clear roadmap for inclusive AI governance that moves beyond high‑level principles to actionable commitments this including measurable milestones for capacity building in represented regions like Africa. 2nd, focus on the establishment of a permanent multi stakeholder mechanism that ensures the technical community, Civil Society, and youth voices have a structured seat at the table, as I saw most of the time in observer so not just observer status. 3rd, the launch of one or more pilot projects that demonstrate international cooperation on AI for public goods such as a shared framework for AI‑driven projects. for example one of my reseach AI help for climate resilience in telecommunications infrastructure, drawing from my published ITU use case. Success should be measured not by declarations but by follow‑through, a clear calendar of working groups, dedicated funding for global South participation and commitment to reconvene with evidence of progress.

• Safe, secure and trustworthy AI
• AI capacity-building
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Open-source software, open data and open AI models

I selected these four priorities because they directly address the gaps I have observed as an AI researcher in Ethiopia, an NDSS fellow, and as an ITU AI for Good Scholar. 1st, Safe, secure, and trustworthy AI is foundational. My NDSS Fellowship exposed me to the fragility of distributed AI systems, in all of time without security and trust, at any time, no governance framework can succeed. 2nd, AI capacity building is urgent for Africa. Most AI governance discussions take place in developed countries, but this AI era should not pass the Global South. After some time, this gap will widen further. Therefore, without targeted investment in education, infrastructure, and local research, these regions will remain rule takers rather than rule makers. 3rd, social, economic, ethical, cultural, linguistic, and technical implications cannot be separated. AI systems deployed in any country without local language support or cultural context fail these dangers consequnces after some years, so governance must account for linguistic diversity and local economic realities. It should also include community participation, inclusive design practices, and continuous evaluation to ensure that AI systems remain relevant, fair, and effective over time. 4th, open source software, open data, and open AI models democratize access. Proprietary AI widens the digital divide, so there must be a stronger emphasis on openness. Open ecosystems allow African researchers, students, and startups to innovate without paying high licensing fees, enabling locally relevant solutions. This approach helps address existing gaps in the current system. Together, these four priorities form a coherent framework for secure AI, built by a capable global workforce, respectful of local contexts, and powered by open tools. I urge the Dialogue to treat them as interlinked, not separate tracks.

Yes, two critical issues are underrepresented in the listed themes. 1. AI for climate resilience in critical infrastructure. My published ITU use case (AI-driven predictive beamforming for 5G/6G networks) demonstrates how AI can directly mitigate climate-induced disruptions in this case, signal degradation from heavy rain. Yet climate adaptation is not explicitly listed. I recommend adding a thematic track on AI for climate-resilient infrastructure and disaster response, with a focus on telecommunications, energy grids, and early warning systems sectors where AI can save lives. 2. Meaningful youth participation mechanisms. Many UN processes include "youth" as an afterthought, a single speaker or a side event. The AI Dialogue should institutionalise youth engagement through a rotating Youth Advisory Council with voting-like input on agenda setting, and by funding 10-15 young technical experts from developing countries to attend each Dialogue session as full participants, not observers. Tokenism is not governance. These two issues, climate resilience and genuine youth inclusion, cut across all listed themes and deserve explicit attention in the Dialogue's outcomes.

Ethiopia / East Africa / Telecommunications & AI Research Sector In Ethiopia and across Africa, the absence of coordinated AI governance creates both acute challenges and untapped opportunities. Challenges: First, security gaps in our telecommunications infrastructure are rapidly adopting network optimization techniques, but without clear security standards, these systems are vulnerable to adversarial attacks. My NDSS Fellowship highlighted how distributed AI systems lack robust defense mechanisms. Second, capacity gaps most Ethiopian universities have no AI governance curriculum, and policymakers lack technical literacy, leading to reactive, often copy‑pasted regulations from Europe/India that do not fit local contexts. Third, data and language gaps, proprietary AI models trained on Western data fail to recognise Amharic, Oromo, or Tigrinya, excluding millions from digital services. Opportunities: First, leapfrogging Africa can build AI governance frameworks that are agile, inclusive, and context‑aware, avoiding the bureaucratic over‑regulation seen elsewhere. Second, open‑source ecosystems and affordable AI tools allow Ethiopian startups and researchers to innovate without large capital, as demonstrated by my work on low‑cost IoT vehicle safety systems. Third, with 70% of Ethiopia's population under 30, targeted AI capacity‑building can turn a governance gap into a talent engine. The ITU AI for Good initiative proved that African youth can contribute at the highest level when given the opportunity.

The AI Dialogue can play a catalytic and coordinating role that no existing forum currently fills. First, it can bridge the North‑South divide by ensuring that AI governance is not dictated solely by wealthy nations. As an AI researcher from Ethiopia, I have seen how global discussions often exclude African voices. The Dialogue should mandate balanced regional representation and fund participation from low‑income countries. Second, it can establish minimum interoperability standards for AI governance frameworks. Today, the EU, US, China, and the African Union are developing divergent approaches. The Dialogue should not impose uniformity but should identify common principles of security, transparency, and accountability that allow cross‑border AI systems to operate safely. Third, it can create an early warning mechanism for emerging AI risks. The speed of AI development outpaces treaty negotiations. A standing technical working group, co chaired by the technical community and governments, could issue non-binding advisories on threats such as AI powered cyberattacks or algorithmic bias in critical infrastructure. It should also establish clear communication channels, share timely risk intelligence across regions, and support rapid response coordination to mitigate potential harms effectively. Finally, the Dialogue should connect to existing implementation mechanisms the ITU, UNESCO, and regional bodies rather than creating parallel structures. Its added value is political convening power at the UN General Assembly level, which can translate technical agreements into high‑level commitments.

The AI Dialogue should build upon several proven initiatives rather than start from zero. 1. ITU AI for Good platform. My experience as an ITU AI for Good Scholar showed that this platform successfully convenes technical experts. The Dialogue could adopt ITU's use‑case library and scholar network as a technical resource, adding a political layer for standardisation and treaty discussions. 2. UNESCO Recommendation on AI Ethics. UNESCO has done foundational work on ethical frameworks, but implementation remains weak. The Dialogue could transform those principles into measurable national action plans with reporting mechanisms, similar to the Paris Agreement model. It could also introduce benchmarking tools and peer review systems to track progress and encourage accountability. 3. African Union AI Working Group. The AU is developing a continental AI strategy. The Dialogue should formally partner with the AU to ensure African priorities such as capacity building, local languages, and open data are not afterthoughts. This would provide a strong model for regional integration and ensure that global governance reflects diverse development contexts. 4. Internet Society (ISOC) chapters and IGF. The Internet Governance Forum has decades of multistakeholder experience. The Dialogue should adopt the IGF's inclusive model while adding binding outcomes rather than limiting itself to discussion. It could also strengthen participation from underrepresented groups to ensure more balanced global input.

An effective AI Dialogue depends on meaningful contributions from all stakeholder groups, each bringing a distinct perspective shaped by their experience and responsibility. The technical community, for instance, plays a critical role in grounding discussions in evidence. Their expertise allows them to assess emerging AI risks, evaluate the feasibility of governance options, and identify new capabilities before they become widespread. Establishing a standing Technical Advisory Panel that reports directly to the Dialogue's co chairs would ensure that decision making remains informed by up to date and credible analysis rather than speculation. Civil society and youth groups bring a different but equally essential dimension. They act as a bridge between policy and lived experience, ensuring that governance frameworks remain accountable to the people they affect. In many cases, marginalised communities are the first to experience the unintended consequences of new technologies. Creating a Youth Delegate Program with dedicated, funded participation would allow voices from the Global South to be heard consistently, not occasionally. This inclusion strengthens legitimacy and ensures that policies are not detached from reality. The private sector, as the primary developer and deployer of AI systems, holds valuable practical knowledge. Companies understand operational challenges, limitations, and risks that may not be visible from the outside. However, this information is often sensitive. Structured, closed door roundtables organised by sector such as telecommunications or finance can create a trusted environment for sharing insights. This approach encourages honest dialogue while protecting confidentiality, leading to more realistic and implementable solutions. Academia contributes through long term, systematic research. Unlike other stakeholders who may focus on immediate concerns, academic institutions can analyse the broader social and economic impacts of AI over time. Their work helps identify patterns, unintended consequences, and policy gaps. By soliciting pre Dialogue working papers and summarising them for policymakers, the process becomes more informed and forward looking, rather than reactive. Governments, on the other hand, provide the political authority needed to translate ideas into action. Without their involvement, even the most well designed frameworks remain theoretical. At the same time, governments can learn from each other. Peer learning sessions, where countries present their national AI governance experiments, can build trust and accelerate progress by sharing both successes and failures in an open setting. Finally, the Dialogue should move away from long, formal speeches and instead prioritise interactive formats. Workshops focused on problem framing, collaborative solution design, and drafting concrete commitments are more likely to produce meaningful outcomes. This shift from passive listening to active participation reflects the urgency and complexity of AI governance, where progress depends not just on ideas, but on collective action.

A meaningful AI Dialogue must begin by recognising who is not in the room. At present, several groups remain consistently underrepresented, and this absence shapes the outcomes of global discussions in subtle but significant ways. Young people from developing countries are often invited to participate, but many of them come from elite universities or well connected organisations. This leaves out a much larger group of youth who lack institutional backing but have valuable perspectives grounded in everyday realities. Similarly, Indigenous communities are rarely included, even though their knowledge systems and concerns around data sovereignty are highly relevant to AI governance. Their exclusion limits the diversity of ethical frameworks being considered. Small island developing states face a different set of challenges. Their exposure to climate risks and their limited digital infrastructure create unique vulnerabilities that are often overlooked in global AI policy discussions. Persons with disabilities also remain on the margins, despite the fact that AI systems can either enhance accessibility or reinforce exclusion depending on how they are designed. In many cases, accessibility and bias are treated as secondary concerns rather than core design principles. Language is another critical gap. Most AI systems are developed for a narrow set of dominant languages, leaving the vast majority of the world's linguistic communities unrepresented. Speakers of languages such as Amharic, Swahili, or Bengali are often excluded not only from the technology itself but also from the conversations that shape its future. In addition, grassroots technologists, including self taught developers, repair technicians, and community network operators, play a key role in maintaining digital systems on the ground. Yet their contributions are rarely acknowledged because they do not fit traditional definitions of expertise. Addressing these gaps requires deliberate and structured action. One practical step is to establish fully funded fellowship programmes that support participants from underrepresented groups. Covering travel, visas, and even childcare can remove barriers that would otherwise prevent meaningful participation. Regional pre Dialogues can also play an important role. By hosting discussions in Africa, Asia Pacific, and Latin America, and formally integrating their outcomes into the main event, the process becomes more inclusive and grounded in diverse realities. Language accessibility must also be expanded. Providing interpretation in a wide range of languages, including major African and Asian languages, allows more participants to engage directly rather than through intermediaries. In addition, asynchronous participation mechanisms such as online discussion forums and document commenting enable contributions from those who cannot attend live sessions due to time zone differences or connectivity limitations. Another effective approach is the creation of a community rapporteur programme. By training local organisers to host listening sessions and submit structured inputs, the Dialogue can capture perspectives that would otherwise remain invisible. This decentralised model ensures that participation is not limited to those who can travel or access formal platforms. Ultimately, inclusion should not be treated as an optional component or a symbolic gesture. It must be embedded into the governance structure of the Dialogue itself. This includes reserving speaking opportunities, ensuring representation in decision making roles, and designing processes that value diverse forms of knowledge. Only then can the Dialogue move beyond representation and toward genuine participation that shapes outcomes in a meaningful way.

To foster dynamic engagement, the AI Dialogue should move beyond traditional panels. 1. Policy prototyping labs. Mixed teams (gov, tech, civil society) spend 90 minutes drafting a mini‑agreement on a concrete issue e.g., "common reporting format for AI security incidents." The best prototypes are refined and adopted as pilot commitments. 2. AI‑powered real‑time synthesis. Use privacy‑preserving LLMs to summarise breakout discussions instantly, identify conflicting positions, and suggest compromise language. Humans vote on the AI's suggestions – accelerating negotiation. 3. Silent deliberation followed by dot‑voting. Participants read short proposals silently, then use coloured dots to indicate support, concerns, or need for modification. This prevents loud voices dominating and surfaces genuine consensus. 4. "Fishbowl" conversations. A small diverse group discusses on stage while outer circle listens. Outer circle members can tap in to replace someone. Ensures rotating voices and focused dialogue. 5. Pre‑committed action pledges. Each stakeholder group submits one concrete action they will take if a certain principle is adopted. Pledges are displayed publicly. Creates accountability and transforms words into commitments. These formats prioritise doing over talking. The Dialogue should be judged by the number of actionable agreements, not the length of its final declaration.

Based on my experience as an ITU AI for Good Scholar and NDSS Fellow, I highlight three effective examples: 1. ITU AI for Good use-case library and scholar programme. The ITU's open repository of 20+ AI use cases (including my own on predictive beamforming for climate-resilient 5G/6G) provides evidence-based, peer-reviewed examples of AI addressing societal challenges. The scholar programme embeds young technical experts directly into governance discussions. Why it works: It grounds abstract policy in real applications and ensures technical voices are heard. 2. The African Union's emerging AI strategy with a focus on local languages and open data. The AU is developing a continental framework that prioritises capacity-building, open-source models, and data sovereignty. Unlike top-down approaches, it involves grassroots stakeholders. Why it works: It acknowledges that governance must be context-specific AI deployed in Ethiopia cannot ignore Amharic or local infrastructure constraints. 3. NDSS (Network and Distributed System Security) Symposium's adversarial review model. NDSS uses rigorous, double-blind peer review and red-teaming to identify vulnerabilities before publication. Applying this model to AI governance e.g., "pre-mortem" exercises where experts attack proposed policies would surface hidden flaws. Why it works: Security and governance both benefit from assuming failure is possible and designing defences accordingly. Recommendation: The AI Dialogue should establish a living repository of governance experiments national laws, corporate codes, community protocols with clear success/failure indicators. This would turn policy into an evidence-based, iterative practice.