MezTeck

A successful outcome, in my view, is not just another set of high-level principles, but something that translates into how AI systems are actually built and used in the real world. From what I've seen while working on AI-driven systems, the biggest gap today is between policy discussions and implementation. So for this dialogue to be meaningful, it should produce practical guidance that developers, companies, and institutions can actually apply, especially those building decision-making systems. I would consider it a success if the dialogue leads to: Clear, actionable recommendations for building transparent and accountable AI systems Shared understanding of how to handle AI-driven decision-making in sensitive areas like hiring, procurement, and public services Stronger inclusion of voices from regions like Africa, where AI is being adopted in very different contexts and constraints Initial steps toward alignment across countries, even if full standardization is not immediately possible Most importantly, the outcome should reduce ambiguity. Right now, many builders want to do the right thing, but lack clarity on what responsible AI actually looks like in practice. Bridging that gap would be a real success.

• Safe, secure and trustworthy AI
• AI capacity-building
• Transparency, accountability, and human oversight
• Social, economic, ethical, cultural, linguistic and technical implications of AI

My priorities are strongly shaped by working on applied AI systems that directly influence decisions. First, transparency, accountability, and human oversight are critical because AI is increasingly being used to make or support decisions about people and organizations. Without clarity on how decisions are made, trust breaks down very quickly. Second, safe, secure, and trustworthy AI is foundational. In practice, even small errors or biases in AI systems can scale very quickly, especially in automated pipelines. Ensuring reliability and robustness is not optional; it is essential. Third, AI capacity-building is especially important in regions like Africa. There is a growing interest in AI, but limited access to infrastructure, training, and implementation support. Without capacity-building, governance risks becoming theoretical rather than practical. Finally, the broader social and economic implications cannot be separated from technical design. AI systems behave differently depending on context, data availability, and local realities. Governance needs to reflect that diversity instead of assuming a one-size-fits-all model.

One key issue that I believe is still under-discussed is the governance of AI-driven decision systems, especially those used in operational workflows like hiring, procurement, lending, and public service delivery. These systems are not just tools they actively shape outcomes. However, there is still limited clarity on: Who is accountable when an AI-assisted decision causes harm How to audit systems that continuously learn or adapt How to balance automation with meaningful human oversight Another emerging issue is the gap between AI development and deployment environments. Many governance discussions assume ideal conditions, but in reality, systems are often deployed in low-resource settings with limited monitoring and control. There is also a growing need to address practical implementation standards, not just principles. Builders need concrete frameworks they can follow during system design, not only after deployment. Addressing these gaps would make AI governance more grounded, inclusive, and effective in real-world settings.

From my perspective, working on applied AI systems in Ethiopia and more broadly in Africa, the governance gaps are very visible at the implementation level. One of the biggest challenges is that AI adoption is moving faster than the structures needed to guide it. Organizations are starting to use AI in areas like hiring, procurement, and service delivery, but there is very limited guidance on how to ensure these systems are fair, transparent, and accountable. This creates real risks, not necessarily because of bad intent, but because of unclear standards. Another challenge is the capacity gap. Many teams are interested in building or using AI, but lack access to training, infrastructure, and practical frameworks. As a result, systems are often deployed without proper evaluation, monitoring, or understanding of long-term implications. There is also a context gap. Most existing governance models are developed in very different environments, and they don't always translate well to local realities, especially where data is limited, informal systems are common, and digital infrastructure is still evolving. At the same time, there are strong opportunities. The region is not deeply locked into legacy systems, which creates space to build AI systems more responsibly from the start. There is also growing interest from young professionals and builders who are eager to apply AI to real-world problems. If supported with the right guidance and capacity-building, regions like ours can move quickly not just in adopting AI, but in shaping how it is governed in a more practical and inclusive way.

I see the AI Dialogue as an opportunity to close the gap between fragmented national approaches and the need for more coordinated global direction. Right now, countries and organizations are moving at very different speeds, with different priorities and levels of capacity. Without some level of alignment, this creates confusion for builders and risks widening the gap between regions that can shape AI and those that mainly adopt it. The Dialogue can play a practical role by creating a shared space where not only policies are discussed, but also real implementation experiences are exchanged. For example, what works in deploying AI systems in constrained environments, how organizations are handling accountability, and where current approaches are failing. It can also help establish a baseline understanding of key principles, not necessarily strict global standards, but common reference points that different regions can adapt to their own context. Another important role is inclusion. For international cooperation to be meaningful, it has to go beyond a small group of countries and institutions. The Dialogue should actively bring in perspectives from regions like Africa, where the challenges and opportunities are different but highly relevant. If done well, the Dialogue can move the conversation from abstract alignment to practical cooperation, where countries and organizations learn from each other and build more compatible, responsible AI systems over time.

There are already several important initiatives shaping AI governance, including efforts by the OECD, UNESCO, the African Union, and various national AI strategies. There are also growing communities in open-source AI and research that are contributing to transparency and shared learning. However, many of these efforts operate in parallel, and from a builder's perspective, they can feel disconnected. One framework focuses on ethics, another on policy, and another on technical standards, but there is limited integration between them in practice. The AI Dialogue can add value by acting as a connecting layer, not replacing existing initiatives, but bringing them into a more coherent conversation. This includes aligning principles with real-world implementation and linking policy discussions with the realities faced by developers and organizations. Another important contribution would be amplifying underrepresented voices and surfacing practical insights from regions that are often not central in global discussions but are actively adopting AI. Finally, the Dialogue can help translate high-level frameworks into more actionable guidance. Many organizations want to build responsibly, but need clearer direction on how to apply these principles in real systems. Bringing these elements together would make the overall AI governance ecosystem more connected, practical, and inclusive.

For the AI Dialogue to be meaningful, participation should go beyond formal statements and create space for practical input from different types of stakeholders. Governments can contribute by sharing policy approaches, but also by being open about challenges and gaps in implementation. The private sector and technical community should bring real-world experience what is actually being built, what is working, and where things are breaking down. Academia can help structure these insights and provide deeper analysis, while civil society plays a critical role in highlighting societal impacts and accountability concerns. In terms of format, the Dialogue should not rely only on panels or prepared remarks. There should be structured ways to capture applied experiences, such as: Short case-based submissions from builders and organizations Thematic working groups focused on specific problems (e.g., AI in hiring or public services) Open consultation sessions where participants can react to each other's ideas A useful structure would combine high-level discussions with smaller, more focused sessions where participants can go deeper into real issues. This balance would make the Dialogue both inclusive and practically valuable.

One of the most underrepresented groups in AI governance discussions is practitioners who are actively building and deploying systems in real-world environments, especially in low- and middle-income countries. There is also limited representation from small and emerging companies, independent developers, and local innovators who are working with constrained resources but solving very relevant problems. Their perspectives are important because they often deal directly with the challenges of data limitations, infrastructure gaps, and contextual differences. Additionally, communities that are most affected by AI systems, such as job seekers, small suppliers, and users of public services, are rarely part of these conversations in a meaningful way. To include these voices, the Dialogue should: Lower barriers to participation (simple submission formats, remote access, flexible engagement options) Actively reach out beyond established networks and institutions Create space for practical, experience-based contributions, not only formal policy input Inclusion should not be symbolic — it should shape the discussion itself.

To make the Dialogue more dynamic and impactful, the format should move beyond traditional panels and allow for more interactive and problem-focused engagement. One effective approach would be case-driven discussions, where participants present real AI systems or use cases, including both successes and failures. This helps ground the conversation in reality and makes it easier to identify concrete governance needs. Another format could be live problem-solving sessions, where mixed groups (policy, technical, and civil society) work together on specific challenges, such as designing accountability mechanisms for AI-driven decisions or improving transparency in automated systems. Short, focused formats like builder briefings could also be valuable, quick presentations from practitioners explaining what they have built, what challenges they faced, and what guidance they need. Finally, incorporating hybrid and asynchronous participation (online contributions, recorded inputs, and follow-up discussions) would allow broader global engagement, especially from regions that cannot easily attend in person. The goal should be to create a Dialogue that is not only inclusive but also interactive, practical, and continuously evolving.

From my perspective, effective AI governance comes from approaches that connect high-level principles with how systems are actually built and used. One good practice is embedding transparency and auditability directly into AI systems. For example, in decision-support tools, keeping clear records of how inputs are processed and how outputs are generated makes it easier to review and challenge outcomes. This is especially important in areas like hiring or procurement, where decisions have real consequences. Another practical approach is maintaining human-in-the-loop systems, not just as a formality, but as a meaningful checkpoint. AI can assist in filtering and analysis, but final decisions, especially high-impact ones, should remain accountable to humans with clear responsibility. There is also growing value in modular and explainable system design, where different components of an AI pipeline (data processing, scoring, ranking) can be understood and evaluated independently. This makes systems easier to govern and improve over time. On the policy side, context-aware frameworks are essential. Instead of rigid, one-size-fits-all rules, guidelines should allow adaptation based on local realities, data availability, and sector-specific risks. Finally, capacity-building as part of governance is critical. Policies alone are not enough if organizations lack the skills and tools to implement them. Training, shared resources, and practical guidelines should be considered part of the governance approach itself. Overall, the most effective practices are those that are actionable, adaptable, and grounded in real-world use, not only principles, but systems that can actually be applied and sustained.