Verein Innovation und Soziales

A successful first Global Dialogue on AI Governance should move beyond general principles and deliver practical, inclusive, and measurable outcomes. First, it should establish a shared baseline of actionable governance standards—not just high-level ethics, but concrete guidance on transparency, accountability, data protection, and human oversight that countries at different levels of development can realistically implement. These standards should be adaptable, allowing innovation while safeguarding fundamental rights. Second, success requires inclusion of underrepresented voices, particularly from the Global South and vulnerable populations. AI already shapes access to services, employment, and rights; therefore, governance frameworks must be informed by those most affected. Mechanisms for continuous participation—not one-off consultation—should be created. Third, the Dialogue should prioritize real-world pilot initiatives that demonstrate how AI governance works in practice. For example, in areas like migration and public service delivery, AI systems can reduce administrative burden and improve access to rights when properly governed. Evidence-based pilots, supported by multi-stakeholder partnerships, can translate principles into scalable solutions and generate trust through measurable impact. Fourth, there should be agreement on data governance for public good, including responsible data sharing, interoperability, and safeguards. Today, fragmented systems limit effective policymaking; better data coordination can enable faster, more targeted, and more equitable responses. Finally, the Dialogue should result in a clear roadmap with accountability mechanisms—including timelines, follow-up platforms, and indicators of progress. Without implementation pathways, even the strongest commitments risk remaining symbolic. In essence, success means shifting from dialogue to delivery: aligning global principles with local action, building trust through evidence, and ensuring that AI governance tangibly improves people's lives.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Transparency, accountability, and human oversight
• Protection and promotion of human rights

My selection reflects the need to ensure that AI governance is both human-centered and operational in real-world contexts, particularly in public services and for vulnerable populations. Safe, secure and trustworthy AI is fundamental, as AI systems increasingly influence access to essential services such as legal aid, employment, and social support. Without strong safeguards, risks such as misinformation, bias, or data misuse can disproportionately affect those already in vulnerable situations. The protection and promotion of human rights is equally critical. AI systems must reinforce-not undermine-rights such as access to information, non-discrimination, and due process. In migration and integration contexts, this includes ensuring fair treatment, accessibility across languages, and respect for dignity. I also prioritized transparency, accountability, and human oversight, as these are essential to building trust in AI systems. Users and institutions must understand how decisions are made, be able to challenge outcomes, and ensure that humans remain responsible for critical decisions. Finally, I selected social, economic, ethical, cultural, linguistic, and technical implications of AI to emphasize that AI governance cannot be purely technical. AI systems interact with complex human realities, and their effectiveness depends on how well they reflect diverse user needs, cultural contexts, and systemic inequalities. Together, these priorities aim to ensure that AI is not only innovative, but also fair, inclusive, and practically beneficial, delivering measurable improvements in people's lives while strengthening public institutions.

Yes-several important cross-cutting and emerging issues deserve greater attention. First, AI in public service delivery remains underrepresented. As governments increasingly use AI in areas such as migration, social protection, and employment services, governance must address not only risks but also operational effectiveness, fairness in access, and measurable social outcomes. This includes ensuring that AI systems reduce-not reinforce-administrative burdens and inequalities. Second, there is a growing need for evidence-based governance through real-time data. Current policymaking often relies on delayed or fragmented information, limiting responsiveness. AI creates an opportunity to generate anonymized, real-time insights into system gaps and user needs, but governance frameworks must define how such data can be responsibly used for public good while safeguarding privacy. Third, inclusion through digital and linguistic accessibility is critical. Even well-designed AI systems can exclude users if they are not accessible across languages, literacy levels, and digital skills. Governance should explicitly address accessibility standards to ensure equitable participation. Fourth, human-AI collaboration in decision-making is an emerging issue. Rather than focusing only on "human oversight," more attention is needed on how AI can effectively support professionals (e.g., social workers, legal advisors) without deskilling them or shifting responsibility in unclear ways. Finally, trust through demonstrated impact is often overlooked. Beyond principles and compliance, AI governance should incorporate mechanisms to evaluate real-world outcomes-such as improved access to services, reduced inequalities, or increased efficiency-especially in high-impact sectors. Addressing these cross-cutting issues would help ensure that AI governance evolves from a primarily risk-focused approach to one that also enables responsible, inclusive, and high-impact use of AI in society.

In Switzerland and similar decentralized systems, the main governance gaps lie not in the absence of principles, but in their practical implementation across fragmented institutions, particularly in public service delivery. A key challenge is fragmentation of data and services. Responsibilities for integration, social services, and employment are distributed across federal, cantonal, and local levels, each operating with partial information. This limits transparency, weakens accountability, and makes it difficult to apply consistent human rights standards in practice. As a result, individuals—especially refugees and other vulnerable groups—often face unequal access to services and must navigate complex systems without clear guidance. Another gap concerns operational transparency and accountability of AI systems. While there is strong awareness of ethical AI in Switzerland, there is limited guidance on how to implement explainability, auditability, and human oversight in everyday administrative processes. This creates uncertainty for public institutions and slows responsible adoption. A third challenge is ensuring safe and trustworthy AI in multilingual, high-stakes contexts. In sectors such as legal information or social support, inaccuracies or bias can have direct consequences on people's rights. Governance frameworks need to better address quality assurance, localization, and continuous monitoring in real-world use. At the same time, these gaps create significant opportunities. AI can help reduce administrative burden, improve coordination, and provide real-time insights for policymaking, enabling more efficient and equitable allocation of resources. In Switzerland, where systems are already well-developed but complex, AI offers the potential to connect existing actors rather than replace them—enhancing both efficiency and human-centered service delivery. Overall, the opportunity lies in moving from high-level governance principles to practical, evidence-based implementation models that strengthen trust, improve outcomes, and ensure that AI meaningfully benefits society.

The AI Dialogue can act as a bridge between principles and implementation, aligning global standards with practical use across countries. It should facilitate knowledge-sharing on real-world applications, especially in public services, where governance challenges are most tangible. By promoting common frameworks for transparency, accountability, and human rights, the Dialogue can reduce fragmentation and support interoperability between national approaches. Importantly, it can create multi-stakeholder partnerships—bringing together governments, academia, private sector, and civil society—to co-develop and test solutions. Establishing pilot initiatives and shared evaluation metrics would allow countries to learn from each other and scale what works. The Dialogue can also amplify voices from underrepresented regions, ensuring more inclusive and globally relevant governance.

The AI Dialogue should build on existing efforts such as OECD AI Principles, UNESCO's Recommendation on AI Ethics, the EU AI Act, and UN-led initiatives (e.g., UNHCR, ITU, and Global Digital Compact processes). These provide strong normative foundations but often lack coordination and practical implementation pathways. Its added value would be to connect these frameworks and translate them into actionable tools—for example, guidelines for AI use in public services, shared data governance models, and impact measurement standards. It could also serve as a platform for exchanging pilot results and best practices, reducing duplication of efforts. Ultimately, the Dialogue can become a coordination hub, ensuring that global AI governance evolves as a coherent, inclusive, and impact-oriented system rather than a set of fragmented initiatives.

Different stakeholders should contribute based on their strengths, but within a shared, practice-oriented framework. * Governments can define regulatory standards and enable pilot environments in public services. * Private sector can provide technical expertise and scalable solutions. * Academia can ensure methodological rigor and independent evaluation. * Civil society and NGOs can bring real user needs and safeguard human rights. * End users, especially vulnerable groups, should be directly involved to reflect lived realities. From the perspective of ERIDA, meaningful contribution comes from co-creating and testing solutions in real contexts, not only discussing principles. For example, multi-stakeholder pilots (e.g., in refugee integration) can demonstrate how AI improves coordination, reduces administrative burden, and strengthens access to rights. The Dialogue should therefore be structured around thematic working groups + pilot labs, where stakeholders jointly design, test, and evaluate AI applications, supported by shared metrics and continuous feedback loops.

Key underrepresented voices include refugees, migrants, low-income communities, and frontline practitioners (e.g., social workers, legal advisors). These groups are directly affected by AI in public services but rarely shape governance discussions. To include them, the Dialogue should move beyond formal consultations and create structured participation mechanisms, such as: * Co-design workshops with affected communities * Partnerships with NGOs working on the ground * Multilingual, accessible participation formats * Compensation for participation to ensure equity ERIDA's experience shows that these voices are essential: without them, AI systems risk reinforcing fragmentation and exclusion. With them, AI can become a practical tool for inclusion, better service delivery, and evidence-based policymaking.

To foster meaningful engagement, the AI Dialogue should focus on a few high-impact formats: 1. Multi-stakeholder pilot labs – small, practical working groups where governments, NGOs, tech actors, and users co-design and test real AI applications (as done in ERIDA). This ensures governance is grounded in real-world use. 2. User journey workshops – mapping real experiences (e.g., accessing services) to identify gaps and ensure AI governance reflects lived realities, especially for vulnerable groups. 3. Data-driven dialogue sessions – discussions based on anonymized, real-time insights rather than assumptions, enabling more evidence-based and actionable outcomes. These formats shift the Dialogue from discussion to co-creation, testing, and measurable impact.

One concrete example is ERIDA (Efficient Refugee Integration through Data Analytics), a Swiss-based AI platform designed to improve governance in complex public service systems. ERIDA addresses key governance challenges-such as fragmentation, lack of transparency, and limited real-time data-by providing a single, multilingual digital entry point for refugees and integration actors. Through AI-powered agents, it offers personalized guidance (e.g., legal, employment, housing) while automating routine inquiries and referrals across institutions. From a governance perspective, ERIDA introduces several good practices: * Transparency and accountability: The system provides clear, traceable guidance and supports human oversight by referring complex cases to qualified professionals. * Human rights and inclusion: It ensures equitable access through multilingual, user-centered design, particularly for vulnerable groups navigating complex systems. * Data governance for public good: ERIDA generates anonymized, aggregated data that enables real-time insights into system gaps, supporting evidence-based policymaking while respecting privacy. * Operational efficiency: By reducing repetitive administrative tasks, it allows institutions to focus more on personalized, human-centered support. Importantly, ERIDA is developed through multi-stakeholder collaboration (public authorities, NGOs, academia), ensuring that governance principles are embedded directly into system design and tested in real-world contexts (e.g., pilot in the Canton of Bern). This approach demonstrates how AI governance can move from abstract principles to practical, measurable impact, improving both service delivery and policy effectiveness.