Humanitarian Documentation

A successful first Global Dialogue on AI Governance should not feel like another room where responsibility is deferred. At the Internet Governance Forum last year, I watched governments, international bodies, and technology companies point at each other to take the lead on governance, while avoiding real accountability. If this Dialogue repeats that dynamic, it will fail, regardless of how strong the language of its outcomes appears. Success starts with shifting who is in the room and whose realities shape the conversation. The people most impacted by AI systems, particularly in conflict-affected regions and the Global South, are largely absent from these spaces. Instead, governance is being shaped by a small set of actors, often reinforcing what increasingly feels like "digital colonialism," where systems are built in a few contexts and exported globally without meaningful local agency. It must also expand how we define risk. The focus is often on misinformation and disinformation, but the deeper shift is in how AI is reshaping trust: how people perceive institutions, who they believe, and what they expect from authority. In many places, this is already altering relationships between citizens and governing bodies in ways that are not being captured in current frameworks. Bias is not only a technical issue, it is embedded in language, culture, and perspective. When governance conversations are dominated by a narrow set of voices, those biases scale globally through AI systems. A successful Dialogue would confront these realities directly: redistribute voice, define clear accountability, and focus on how AI is reshaping power and trust, not just information. If it cannot do that, it risks becoming another forum that describes the problem without changing it.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight
• AI capacity-building

AI systems are shaping how people see the world, but the people most affected by those systems are largely absent from how they are built, trained, and governed. That gap matters. When entire regions, languages, and lived experiences are missing from the process, their absence shows up in the outputs: what is visible, what is normalized, and what is ignored. This is why the social, cultural, and linguistic dimensions of AI are not secondary concerns. They are where power shows up. AI is not just reflecting reality, it is constructing a version of it, often based on a narrow set of perspectives that then scale globally. Human rights, in this context, is not only about protection from harm. It is about participation, who has the ability to shape the systems that increasingly shape them. Without that, governance risks reinforcing existing inequalities under the appearance of neutrality. At the same time, these systems are influencing perception and decision-making at scale, without clear lines of accountability. It is often unclear who is responsible for how these systems represent people, cultures, or events, even as their impact grows. Capacity-building is therefore essential, not as a development goal, but as a condition for meaningful participation. Without it, we are not creating inclusive governance, but extending a model where a small number of actors define the informational environment for everyone else.

One issue that cuts across all of these themes but is not explicitly named is representation, specifically, who is included in shaping AI systems and who is not. Current discussions often focus on outcomes: safety, fairness, or risk mitigation. But much less attention is given to who defines those standards in the first place. AI systems are trained on data, language, and perspectives that are unevenly distributed, and this imbalance becomes embedded in how these systems generate and prioritize information. As a result, certain worldviews are reinforced at scale, while others are underrepresented or excluded entirely. This is not only a technical or ethical issue, but a structural one. When participation in building and governing AI is concentrated among a small set of actors, it shapes not just how systems function, but what they consider relevant, valid, or true. A related emerging issue is the growing gap between influence and accountability. AI systems are increasingly capable of shaping perception, decision-making, and public understanding, while making it more difficult to trace origin or responsibility. This creates an environment where impact is clear, but ownership is diffuse. Finally, there is a need to recognize AI as a form of infrastructure for knowledge and perception. Governance frameworks often treat AI as a tool or product, but in practice it is becoming part of how reality is constructed and understood. This has implications that extend beyond existing categories and require new ways of thinking about governance. Addressing these gaps will require shifting from a focus on outputs to a deeper consideration of participation, power, and representation within AI systems themselves.

The most significant gap I see is between where AI systems are being built and where they are being experienced. Across many of the regions I have worked in, particularly in conflict-affected and politically sensitive contexts, AI is already shaping how information is produced, translated, and circulated. But the people in these environments are not meaningfully involved in how these systems are designed or governed. That disconnect shows up in very practical ways: misrepresentation of local realities, loss of nuance in language, and outputs that reflect external perspectives more than lived experience. This creates a broader challenge around trust. When systems consistently fail to reflect people accurately, it affects not only how information is received, but how institutions and external actors are perceived. Over time, this can deepen skepticism toward both technology and governance structures. There is also a growing accountability gap. AI systems are influencing perception and decision-making at scale, but it is often unclear who is responsible when those systems misrepresent and falter. This is particularly difficult in cross-border contexts, where systems developed in one region are deployed in another without clear mechanisms for recourse. At the same time, there is a clear opportunity. If governance efforts focus on inclusion and capacity-building, AI could support more accurate representation, enable local actors to shape their own narratives, and strengthen trust rather than erode it. The challenge is whether governance frameworks can move quickly enough to close the gap between those building these systems and those living with their consequences.

The AI Dialogue can play an important role, but only if it is clear about what it is trying to do differently from existing forums like the Internet Governance Forum. The IGF has been valuable as an open space for exchange, but it is not designed to produce alignment or accountability. In practice, it often becomes a place where different actors (governments, companies, and international organizations) articulate positions without clear follow-through or ownership. That model has limits, especially in a space like AI where the pace of development is outstripping the ability of fragmented approaches to keep up. The opportunity for the AI Dialogue is to move beyond discussion and make cooperation more operational. That does not require binding agreements, but it does require continuity. A recurring forum can track how positions evolve, where approaches are diverging, and whether previous discussions have led to any form of action. It can also create a level of visibility that is currently missing. When differences in approach remain implicit, coordination becomes difficult. Making those differences explicit and revisiting them over time creates a basis for more practical cooperation. The value of the Dialogue is if it can introduce continuity, visibility, and a clearer link between discussion and action, then it can serve as a meaningful complement to existing spaces rather than another parallel track.

There is no shortage of initiatives working on AI governance, but they are often operating in parallel and shaped by a relatively narrow set of actors. Efforts like the United Nations High-Level Advisory Body on AI, the OECD AI Principles, the Global Partnership on AI, and regulatory approaches such as the EU AI Act are all contributing to the global landscape. At the same time, forums like the Internet Governance Forum create space for dialogue. The challenge is that these efforts are not only fragmented, but often shaped without meaningful participation from many of the regions and communities most affected by AI systems. As a result, coordination is not just a technical issue, but a question of whose perspectives are being connected and whose are still missing. The AI Dialogue could add value by acting as a connective layer across both initiatives and participants. That means not only making existing frameworks more legible and comparable, but also creating clearer pathways for perspectives that are currently peripheral to influence how these efforts evolve. It also has the opportunity to introduce continuity in a space where many initiatives are episodic. By revisiting key issues over time and making gaps more visible, it can help ensure that discussions translate into something more than parallel progress.

Different stakeholders can only contribute if they are actually able to show up in a meaningful way and right now, many of the people most affected by AI systems are not in these rooms at all. So the starting point isn't just format, it's access. If the Dialogue wants a broader set of perspectives, it has to actively bring them in and support their participation. That means being intentional about who is invited, but also recognizing that showing up requires time, resources, and context that not everyone has. Once people are in the room, the structure needs to make their presence matter. A lot of these forums create space for input, but not for influence. You hear perspectives, but they don't shape what happens next. If that continues, participation will remain surface-level. What would make a difference is grounding the Dialogue in real situations rather than abstract themes. When conversations are tied to what is actually happening, how AI systems are being used, where they are failing, what people are experiencing; it becomes much harder to ignore certain perspectives, and much easier to see where they should influence outcomes. The format should support that. Large, open discussions tend to reinforce existing dynamics. Smaller, mixed groups, where people who build systems, govern them, and live with their consequences are in the same space, are more likely to produce something useful.

The most underrepresented voices are not hard to identify. They are the people living with the consequences without having any role in shaping them. Sure, this includes communities in conflict-affected regions, smaller language groups, and parts of the Global South where AI systems are often deployed but not developed, but it also includes people working outside formal policy and technical spaces such as journalists, local organizers, artists, andhonestly, just the every day person. They are often closest to how these systems show up in everyday life. Their absence matters because AI systems are not neutral. They reflect the data, language, and assumptions they are built on. When those inputs come from a narrow set of contexts, the outputs carry those same limitations, and those limitations scale globally. Inclusion, in this case, is not just about inviting more people into the room. Many of these communities are not in a position to engage through traditional policy formats, and even when they are present, their input does not always carry weight. If the Dialogue is serious about inclusion, it needs to rethink how participation happens. That could mean working through local partners who already have trust and context, supporting participation with resources and preparation, and creating formats that allow people to contribute from where they are rather than requiring them to fit into existing structures. It also means making it clear how these perspectives influence outcomes. Without that, inclusion remains symbolic.

If the Dialogue follows familiar formats such as panels, statements, open discussion, it will reproduce the same dynamic seen in spaces like the Internet Governance Forum, where perspectives are shared but very little changes. More effective engagement would come from structuring the Dialogue around moments of decision instead of discussion that are grounded in the realties of those who are not traditionally in the room. Instead of asking participants to describe positions, the format should require them to respond to real points of divergence and experienced consequnce. The goal is not to debate, but to surface what each actor is actually willing to do in order to make clear improvements. This could be done through working sessions where participants are asked to put forward concrete positions, constraints, and trade offs, and where those positions are made visible across groups. What matters is not consensus, but clarity in understanding where alignment is possible and where it is not. If each session produces a clear record of positions, gaps, and areas of movement, the Dialogue becomes cumulative rather than repetitive. Over time, this creates pressure toward coordination in a way that open ended discussion does not.

There are already strong foundations for AI governance, but they tend to operate at different levels and don't always connect in practice. Frameworks like the OECD AI Principles and the UNESCO Recommendation on AI Ethics are valuable because they establish shared baselines: human rights, accountability, and inclusion, etc., that many countries have aligned around. They've been effective in creating a common language, which is a necessary starting point for international cooperation. More operational approaches, like the EU AI Act, show what it looks like to move from principles to enforcement. Its risk-based model where higher-risk systems face stricter requirements around transparency, testing, and oversight, is one of the clearest attempts to translate governance into practice. There are also practical frameworks like the NIST AI Risk Management Framework, which focus on how organizations identify and manage risk across the lifecycle of AI systems, helping bridge the gap between policy and implementation. What these examples show is that progress is happening, but in layers: principles, regulation, and implementation tools. The challenge is that they are often disconnected. The opportunity is to link them more intentionally. Principles without enforcement remain abstract, regulation without implementation becomes difficult to apply, and technical frameworks without broader context risk being too narrow. Effective governance seems to come from combining these approaches, while remaining flexible enough to adapt as the technology evolves.