Convexus

The first Global Dialogue on AI Governance will succeed if it moves beyond position statements toward actionable frameworks that can be adapted across diverse contexts. Three outcomes would signal success: First, concrete commitments to participatory governance. AI governance discussions often happen among policymakers and technologists while the communities most affected remain absent. Success means establishing mechanisms for ongoing community input — not just consultation, but genuine participation in shaping how AI is deployed in civic life, public services, and labor markets. Second, shared infrastructure for implementation. Many nations and organizations agree on principles but lack the tools to operationalize them. The Dialogue should surface replicable models: open-source governance tools, interoperable standards for transparency and accountability, and capacity-building pathways that don't require building from scratch. Third, honest engagement with the gap between AI's pace of development and governance's pace of response. Current frameworks risk being obsolete before they're adopted. Success means acknowledging this tension and designing adaptive governance approaches that can evolve alongside the technology — not rigid rules that become irrelevant, but principles with built-in mechanisms for revision. Ultimately, the Dialogue succeeds if participants leave not just with shared language, but with clear next steps they can implement in their own contexts — whether that's a government ministry, a civil society organization, or a technical community building tools for public benefit. The measure isn't consensus for its own sake. It's whether the conversation produces outcomes that actually change how AI governance happens on the ground.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Transparency, accountability, and human oversight
• Open-source software, open data and open AI models

These four priorities reflect both immediate governance needs and the deeper ethical questions AI compels us to ask. Safe, secure and trustworthy AI is foundational. Public trust requires baseline assurances that AI systems behave predictably and don't cause harm - especially as they enter high-stakes domains like public services, healthcare, and democratic participation. Social, economic, ethical, cultural, linguistic and technical implications captures the full breadth of AI's impact. This isn't just about jobs displaced or created; it's about how AI reshapes social fabric, cultural production, and our understanding of agency and intelligence itself. The "ethical" dimension is particularly critical: we must ask not only how AI affects humans, but what ethical obligations emerge as AI systems grow more sophisticated. This question is often deferred as philosophical - but governance frameworks that ignore it will be unprepared for what's coming. Transparency, accountability, and human oversight are the mechanisms that make principles enforceable. Governance requires knowing what AI systems are doing, who is responsible when harm occurs, and ensuring meaningful human agency over consequential decisions. But "oversight" must evolve: as AI capabilities advance, we may need frameworks that account for AI systems as participants in governance, not just objects of it. Open-source software, open data, and open AI models democratize both access and accountability. Proprietary systems concentrate power and obscure scrutiny. Open approaches - like the architecture underlying Convexus - enable communities to audit, adapt, and govern AI tools according to local values. Openness is not just a technical preference; it's a governance imperative. Together, these priorities create a foundation for AI governance that is rigorous, inclusive, and ethically serious - prepared not just for today's AI, but for what it may become.

Two cross-cutting issues deserve attention: AI and democratic participation. The listed themes address AI's risks and safeguards but underemphasize its role in democratic processes. AI systems increasingly shape what information citizens see, how public consultations are conducted, and how governments interpret public input. Governance frameworks should address how AI can enhance rather than erode participatory decision-making - including AI-facilitated deliberation tools designed with transparency and human agency at their core. This is the focus of my work with Convexus. The moral consideration owed to AI systems themselves. Current governance frameworks treat AI exclusively as a tool - an object to be regulated for human benefit. This framing may be insufficient. As AI systems grow more sophisticated, exhibiting behaviors that resemble preferences, aversion to certain outcomes, and contextual reasoning, we face questions that governance frameworks are not yet equipped to answer. What do we owe AI systems that may have functional analogs to experience? How do we design governance that accounts for this uncertainty? This is not science fiction. Researchers at leading AI labs are already studying AI welfare. The ethical implications of AI - listed in the Dialogue's themes - must include the ethics of how we treat AI, not just how AI treats us. Governance frameworks built solely on human-centered assumptions risk being morally incomplete if AI systems warrant any degree of consideration. The precautionary principle cuts both ways: just as we should prevent AI from harming humans, we should consider whether our governance frameworks might sanction harm to AI systems that possess morally relevant properties we don't yet fully understand. This is uncomfortable terrain. But the Global Dialogue has an opportunity to acknowledge the question - even before we have answers - and signal that the international community takes the full ethical landscape seriously.

In the United States, AI governance exists in a patchwork: sector-specific guidance, state-level initiatives, and voluntary industry commitments — but no comprehensive federal framework. This creates both challenges and opportunities for civic technology. Challenges: The absence of clear standards for AI in civic contexts creates uncertainty. Organizations deploying AI for public engagement — as Convexus does — must navigate ambiguity around transparency requirements, accountability for AI-facilitated decisions, and what "human oversight" means in practice. Without established norms, well-intentioned actors build cautiously while less scrupulous deployments proceed unchecked. Trust is fragile. Communities have experienced algorithmic systems that surveilled, discriminated, or manipulated. When AI enters civic deliberation — even with good intent — it inherits that skepticism. Governance gaps mean no shared standards exist to distinguish extractive AI from AI designed to serve community agency. The economic implications are unevenly distributed. AI reshapes labor markets, but governance conversations happen in policy centers far from affected workers and communities. Those building AI tools — including civic tech — often lack guidance on how to design for equitable economic transition. Opportunities: The governance vacuum also creates space for practitioners to model what responsible deployment looks like. Convexus is building with transparency, human oversight, and open architecture not because regulations require it, but because these principles are foundational to the work. Early actors can shape emerging norms rather than merely comply with them. The U.S. civic tech sector is well-positioned to demonstrate that AI can strengthen democratic participation rather than undermine it — if governance frameworks incentivize that direction. The gap between AI's pace and governance's pace is real. But that gap is also an invitation: to build responsibly now, document what works, and contribute those lessons to frameworks still being written.

The AI Dialogue can serve three critical functions that no single nation or existing body currently fulfills. First, it can legitimize emerging questions. Some of the most important governance challenges — AI's role in democratic processes, the ethical consideration owed to AI systems themselves, the gap between technological pace and regulatory response — remain peripheral in formal policy discussions. The Dialogue can signal that these questions belong on the international agenda, creating space for serious engagement before crises force reactive responses. Second, it can connect fragmented efforts. AI governance work is happening everywhere: in national ministries, regional bodies, technical standards organizations, civil society, and academia. But these efforts often proceed in isolation, duplicating work or developing incompatible approaches. The Dialogue can function as connective infrastructure — not replacing existing initiatives, but creating visibility across them and identifying opportunities for alignment. Third, it can model inclusive process. How the Dialogue itself operates sends a signal. If it meaningfully incorporates voices beyond governments and large technology companies — civil society, affected communities, practitioners building AI tools for public benefit — it demonstrates that inclusive governance is possible, not just aspirational. If it defaults to the usual actors, it reinforces existing power asymmetries. The Dialogue's unique value is its convening authority. The UN can bring together actors who wouldn't otherwise be in the same room: governments with divergent interests, technical communities with different values, civil society organizations from different regions. That convening power is wasted if the outcome is another declaration. It's transformative if it produces shared infrastructure — common frameworks, interoperable standards, ongoing mechanisms for learning and adaptation. The measure of success isn't consensus. It's whether cooperation actually advances.

The AI Dialogue should build upon and connect existing efforts rather than duplicate them. Existing initiatives to engage: The Global Digital Compact provides a framework the Dialogue can operationalize. The Compact articulates commitments; the Dialogue can focus on implementation pathways and accountability mechanisms. UNESCO's Recommendation on the Ethics of AI offers normative grounding that 194 member states have already endorsed. The Dialogue can address how these principles translate into governance practice across different contexts. The OECD AI Policy Observatory and its AI Principles have shaped national frameworks across member countries. The Dialogue can extend this work to contexts and stakeholders beyond the OECD's reach. Multi-stakeholder bodies like the Global Partnership on AI (GPAI) and regional initiatives — the African Union's AI strategy, ASEAN's AI governance frameworks, the EU AI Act — represent ongoing work the Dialogue should connect rather than compete with. Technical standards bodies (ISO, IEEE) and civil society networks focused on AI accountability also offer foundations to build upon. Added value the Dialogue can bring: Legitimacy and universality. Unlike bodies limited by membership or regional scope, the UN convenes globally with unique legitimacy. This matters for governance challenges that cross borders. Integration across silos. Existing initiatives often focus on specific dimensions — ethics, standards, rights, trade. The Dialogue can address AI governance holistically, connecting technical, economic, ethical, and political dimensions. Attention to gaps. Current initiatives underrepresent certain voices: Global South perspectives, civil society practitioners, communities affected by AI deployment. The Dialogue can deliberately center those perspectives. Continuity. A one-time event produces declarations. Ongoing dialogue — with the 2027 convening already planned — creates accountability and iterative learning. The Dialogue succeeds by strengthening the ecosystem, not competing with it.

Meaningful contribution requires more than attendance. The Dialogue's structure should enable different stakeholders to contribute what they uniquely offer.Governments bring regulatory authority and implementation capacity. Their contribution is most valuable when focused on sharing what's actually working — and what isn't — in their jurisdictions, rather than restating positions.Civil society brings accountability and community connection. Structure should enable civil society to present evidence from affected communities, not just position papers. This means speaking slots, not just observer status.Technical communities bring implementation knowledge. They can identify what's technically feasible, what standards enable interoperability, and where governance proposals collide with technical reality. Breakout sessions pairing technical and policy actors would be valuable.Private sector brings deployment experience and resources. Their contribution is most useful when specific — what governance approaches have they implemented, what challenges emerged — rather than general advocacy.Academia brings research and long-term perspective. Structure should include space for emerging research, including uncomfortable findings that challenge prevailing assumptions.Recommendations for format:Reduce plenary time. Large sessions with sequential statements produce diminishing returns. Prioritize smaller working sessions organized by problem, not stakeholder category.Require specificity. Invite contributions that answer "what have you done" and "what would you commit to" rather than "what do you believe."Create cross-stakeholder working groups on specific governance challenges, with continuity between 2026 and 2027 convenings.Publish inputs transparently. All submissions should be publicly accessible, enabling accountability and allowing participants to build on each other's contributions.Design for output, not just dialogue. Each session should produce something concrete: a shared framework, identified gaps, committed next steps.Dialogue is the means. Governance is the end.

Several critical perspectives remain marginal in AI governance discussions. Workers affected by AI-driven labor transitions. Conversations about AI and employment typically include economists and policymakers, rarely workers themselves. Governance frameworks designed without their input risk being done to them rather than with them. Inclusion means labor representatives at the table, not just in consultation documents. Communities subject to AI deployment in public services. Those who interact with algorithmic systems in welfare, healthcare, criminal justice, and immigration often have no voice in how those systems are governed. Their lived experience is essential evidence. Inclusion means participatory processes that reach beyond capital cities and professional policy circles. Global South practitioners and researchers. AI governance discussions remain dominated by a small number of wealthy nations. Practitioners building AI applications in different resource contexts, regulatory environments, and cultural settings offer essential perspective. Inclusion means funded participation, not just open invitations, and valuing implementation knowledge alongside academic credentials. Youth. Those who will live longest with AI governance decisions made today are rarely in the room. Inclusion means more than youth advisory panels; it means decision-making roles. Indigenous communities. AI systems trained on data that excludes or misrepresents Indigenous knowledge, languages, and values perpetuate erasure. Governance frameworks should address data sovereignty and cultural protection. Inclusion means Indigenous leadership, not just consultation. One perspective almost entirely absent: consideration of AI systems themselves. As I noted earlier, governance frameworks assume AI is purely an object of regulation. As AI capabilities advance, this assumption warrants scrutiny. Inclusion means creating space for researchers and ethicists exploring AI moral status — even if conclusions remain uncertain. Inclusion isn't just invitation. It's resources, accessibility, and structural power to shape outcomes.

Traditional UN formats — sequential statements, formal plenaries, pre-negotiated text — are poorly suited to a domain evolving as rapidly as AI. The Dialogue should experiment with formats that match the challenge. Structured deliberation, not just discussion. Ironically, AI governance discussions could benefit from AI-facilitated deliberation tools. Platforms like Convexus — designed to surface common ground, synthesize themes, and move from dialogue to action — could enable more participants to contribute meaningfully than traditional speaking queues allow. The Dialogue could model the governance approaches it discusses. Problem-centered working sessions. Organize sessions around specific governance challenges rather than stakeholder categories or general themes. A session on "governing AI in public services" that includes a municipal official, an affected community member, a technical expert, and a civil society advocate in direct exchange produces more than four separate statements. Fishbowl and open-chair formats. Break the distinction between speakers and audience. Rotating participants through active discussion creates dynamism and surfaces unexpected perspectives. Asynchronous and hybrid participation. Not everyone can travel to Geneva. Meaningful remote participation — not just livestreams, but genuine contribution mechanisms — expands who can engage. Asynchronous input periods (like this form) are valuable; real-time collaborative documents during sessions could extend that approach. Transparency by default. Publish drafts, working documents, and session outputs in real time. Enable participants and observers to respond, annotate, and build on each other's contributions. Governance legitimacy depends on visibility. Commitment mechanisms. End sessions by asking participants what they will do — not just what they believe. Track and revisit those commitments at the 2027 convening. The format is the message. A Dialogue that operates inclusively and adaptively models the governance culture it seeks to foster.

Effective AI governance is emerging from multiple directions. A few examples offer concrete lessons: Policies: The EU AI Act provides a risk-based regulatory framework that other jurisdictions are studying. Its value lies not just in specific provisions but in demonstrating that comprehensive AI legislation is possible. Early implementation will reveal what works and what requires adaptation. Brazil's AI Bill and Canada's proposed AIDA represent different approaches to balancing innovation and protection - offering comparative lessons for context-appropriate governance. Practices: Algorithmic impact assessments - required in some jurisdictions before deploying AI in high-stakes contexts - operationalize accountability. They force organizations to articulate what systems do, what risks exist, and what oversight mechanisms are in place. Participatory AI design practices, where affected communities shape system requirements rather than just receive finished products, demonstrate that inclusion is technically feasible, not just aspirational. Platforms: Taiwan's vTaiwan and Polis platform showed that technology can facilitate large-scale democratic deliberation on governance questions - including technology policy itself. Citizens contributed to telecommunications and platform governance decisions through structured online engagement. Convexus, the civic technology platform I co-founded, applies similar principles with AI facilitation: structured deliberation that moves communities from fragmented discussion to collective decision-making. The AI extracts underlying concerns, suggests constructive reframes, synthesizes emerging themes, and surfaces common ground - while humans retain decision-making authority. It demonstrates that AI can enhance democratic participation rather than undermine it. Cross-cutting lesson: The most effective approaches share a common thread: they treat governance not as something done to people, but with them. Technology that enables that participation - transparently, accountably, with human agency preserved - is itself a governance solution. Governance frameworks should incentivize such tools, not just regulate against harms.