EPFL

A successful first Global Dialogue on AI Governance should achieve outcomes that are both principled and operational, bridging high-level ethical commitments with implementable frameworks. First, it should establish a shared baseline of governance principles across jurisdictions. While complete regulatory harmonisation is unrealistic, convergence around core values—such as transparency, accountability, safety, and human oversight—would provide a foundation for international cooperation. Importantly, these principles should be translated into actionable guidelines rather than remaining purely declarative. Second, the dialogue should catalyse the creation of interoperable regulatory approaches. Given the global nature of AI development and deployment, fragmented regulatory environments risk inefficiencies and ethical inconsistencies. Mechanisms for mutual recognition, regulatory sandboxes, and cross-border compliance standards would represent tangible progress. Third, success would involve meaningful inclusion of multidisciplinary perspectives. AI governance cannot be shaped solely by policymakers and technologists; it requires input from healthcare, economics, ethics, and civil society. Drawing on interdisciplinary expertise ensures that governance frameworks are both context-aware and socially robust. Fourth, the dialogue should prioritise capacity-building and equitable access. There is a clear risk that governance frameworks may disproportionately reflect the interests of technologically advanced economies. Supporting emerging and developing countries in building regulatory, technical, and institutional capacity would enhance legitimacy and long-term effectiveness. Finally, a successful outcome would be the establishment of a roadmap for continued collaboration. Rather than a one-off event, the dialogue should initiate a sustained process with defined milestones, working groups, and measurable objectives. In essence, success lies not only in consensus, but in the ability to translate that consensus into coordinated, inclusive, and adaptable governance structures that can evolve alongside the rapid development of AI technologies.

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

From my perspective, these priorities reflect the most urgent and impactful areas for engagement, particularly at the intersection of healthcare innovation, data-driven technologies, and responsible entrepreneurship. Ensuring safe, secure and trustworthy AI is fundamental, especially in high-stakes domains such as medical diagnostics and clinical decision-making, where system reliability directly affects patient outcomes. Closely linked to this is the need for transparency, accountability, and human oversight, which are essential to maintain trust, enable auditability, and prevent unintended harm in complex AI systems. At the same time, AI capacity-building represents a critical lever for reducing global inequalities in access to AI technologies and governance capabilities. Supporting knowledge transfer and institutional development is key to fostering inclusive innovation ecosystems. Finally, addressing the social, economic, ethical, and technical implications of AI ensures that technological advancement remains aligned with broader societal values. A multidisciplinary perspective is particularly important to anticipate long-term impacts and guide responsible deployment.

Yes, while the listed themes capture many foundational dimensions of AI governance, several cross-cutting and emerging issues warrant more explicit attention. First, the integration of AI into high-stakes, domain-specific contexts, particularly healthcare, remains underrepresented as a distinct governance challenge. AI systems in such settings require not only general principles of safety and accountability, but also domain-adapted validation standards, clinical interpretability, and continuous post-deployment monitoring. The risks associated with misdiagnosis, bias in medical datasets, and overreliance on automated systems highlight the need for specialised regulatory pathways. Second, the issue of human-AI interaction and cognitive dependency is emerging as a critical concern. As AI systems become more capable, there is a growing risk of automation bias, deskilling, and erosion of human judgement. Governance frameworks should therefore address not only system performance, but also how humans engage with and rely on these systems over time. Third, data provenance and epistemic integrity deserve greater emphasis. Beyond questions of access and openness, there is a need to ensure the traceability, quality, and contextual relevance of data used to train AI systems. This is particularly important in scientific and medical applications, where flawed or non-representative data can lead to systemic errors. For concluding ,the translation of ethical principles into operational practice remains a persistent gap. While many frameworks articulate high-level values, fewer provide concrete mechanisms for implementation, auditing, and enforcement. Bridging this "ethics-to-practice" gap is essential for ensuring that governance is not merely aspirational, but effective. Addressing these issues would strengthen the robustness, applicability, and long-term resilience of global AI governance efforts.

In the European and Swiss context, as well as within the healthcare and biotechnology sector, governance gaps in AI are creating both significant challenges and strategic opportunities. One of the primary challenges lies in the fragmentation of regulatory approaches. While Europe is advancing comprehensive frameworks such as the AI Act, inconsistencies in implementation and alignment with global standards create uncertainty for innovators and early-stage companies. This is particularly relevant in health-tech, where startups must navigate overlapping requirements related to medical devices, data protection, and AI-specific regulation, often slowing down development and market entry. A second challenge concerns data governance and access. High-quality, representative clinical data are essential for developing reliable AI systems, yet they remain difficult to access due to privacy constraints, institutional silos, and lack of interoperable infrastructures. This limits both innovation and the ability to validate AI tools across diverse populations. At the same time, there are important opportunities. The strong regulatory environment in Switzerland and Europe creates a foundation for trustworthy AI, which can become a competitive advantage globally—particularly in sensitive domains such as healthcare. Clear standards for safety, transparency, and accountability can enhance user trust and facilitate adoption. Moreover, the growing emphasis on AI capacity-building and interdisciplinary collaboration aligns well with the region's academic and innovation ecosystems. Institutions such as EPFL and the University of Geneva foster cross-sector collaboration, enabling the integration of medical, technical, and economic expertise. Finally, addressing the ethical and societal implications of AI presents an opportunity to shape innovation in a more human-centric direction. By embedding ethical reflection and human oversight into system design, the sector can move towards more sustainable and responsible technological development. Overall, while governance gaps introduce complexity, they also create space for leadership in developing robust, ethical, and scalable AI solutions.

The AI Dialogue can play a pivotal role as a bridge between principles and practice in international AI governance. To begin with, it can foster shared understanding and trust among diverse stakeholders. Given that AI development progresses at different speeds across regions—often shaped by distinct political, economic, and cultural contexts—the Dialogue offers a neutral space to align perspectives, clarify expectations, and ultimately reduce fragmentation. Furthermore, it can support the development of interoperable governance frameworks. Rather than pursuing full harmonisation, which may prove unrealistic, the Dialogue can encourage convergence around core standards while enabling mutual recognition across regulatory systems. In doing so, it would facilitate cross-border innovation without compromising safety or accountability. Equally important, the AI Dialogue can strengthen inclusion and capacity-building. By actively engaging countries at different stages of technological development, it helps ensure that governance frameworks are not defined by a limited set of actors. In this regard, promoting knowledge exchange, training, and institutional support becomes essential for building a more balanced and legitimate global landscape. In addition, it can serve as a platform for multidisciplinary collaboration, bringing together policymakers, researchers, industry leaders, and experts from fields such as healthcare, ethics, and economics. Such diversity is crucial to addressing the complex and interconnected nature of AI systems. Finally, its long-term value lies in establishing sustained mechanisms for cooperation—including working groups, shared benchmarks, and iterative policy processes. As AI continues to evolve, governance must remain adaptive rather than static and the AI Dialogue has the potential to shift international cooperation from fragmented discussions towards coordinated, inclusive, and forward-looking action.

A meaningful way to approach this question is to consider, first, the existing ecosystem of AI governance initiatives, and second, the distinct value that the AI Dialogue can add by connecting and amplifying these efforts. A number of important initiatives already shape the global AI governance landscape. For instance, the OECD AI Principles and the work of the Global Partnership on AI (GPAI) have contributed to establishing shared ethical and policy foundations. Similarly, UNESCO's Recommendation on the Ethics of AI provides a globally endorsed normative framework, while the Council of Europe's work on AI and human rights advances legally grounded approaches. In parallel, the European Union's AI Act represents one of the most comprehensive regulatory models currently being developed. Beyond public governance, industry-led initiatives and research collaborations—particularly in areas such as healthcare and biotechnology—also play a key role in advancing technical standards and responsible innovation. However, despite this rich ecosystem, these efforts often remain fragmented, with limited coordination across regions, sectors, and levels of implementation. This is where the AI Dialogue can bring unique added value. Rather than duplicating existing frameworks, the AI Dialogue can act as a connector and integrator. It can facilitate alignment between normative principles and regulatory practices, while also bridging the gap between policy discussions and real-world applications. By creating structured spaces for exchange between governments, academia, industry, and civil society, it can enhance coherence and mutual understanding. Moreover, the Dialogue can contribute by promoting interoperability and practical coordination, for example through shared benchmarks, best practices, and collaborative pilot initiatives. It can also strengthen inclusivity by ensuring that underrepresented regions and sectors are meaningfully involved in shaping global governance. In this sense, the added value of the AI Dialogue lies not in creating new principles, but in connecting, operationalising, and scaling existing ones.

Different stakeholders can contribute to the AI Dialogue most effectively when their roles are clearly defined and when the structure allows for both expertise and practical experience to be meaningfully integrated. From my perspective, working at the intersection of healthcare, biotechnology, and entrepreneurship, it is essential that technical experts and domain practitioners—particularly from high-stakes fields such as medicine—are actively involved. Their contribution is critical to ensure that governance discussions remain grounded in real-world constraints, such as clinical safety, data limitations, and implementation challenges. At the same time, policymakers play a key role in translating these insights into coherent and enforceable frameworks. Equally, academic institutions can contribute through independent research, methodological rigour, and long-term thinking, while industry actors and startups can bring an innovation-driven perspective, highlighting feasibility, scalability, and emerging risks. Importantly, civil society and ethical experts should remain central to the process, ensuring that societal values and human impact are consistently reflected. In terms of structure, the AI Dialogue would benefit from a multi-layered format. First, high-level plenary sessions could define shared priorities and strategic direction. These should be complemented by thematic working groups, focused on specific sectors such as healthcare, where more detailed and technical discussions can take place. Additionally, I would strongly recommend incorporating case-based discussions. Drawing on concrete use cases—such as AI in medical diagnostics—can help bridge the gap between abstract principles and operational realities. This approach also enables stakeholders to collectively identify risks, trade-offs, and best practices. Finally, the Dialogue should not be a one-time event but an ongoing, iterative process, supported by continuous exchanges, pilot collaborations, and measurable follow-ups.

Global discussions on AI governance continue to be shaped by a relatively narrow group of actors, mainly large economies, major technology companies, and policy institutions; consequently, several critical perspectives remain underrepresented. In particular, domain practitioners in high-stakes environments, such as clinicians and biomedical researchers, are rarely central to governance design. Yet, these professionals engage daily with uncertainty, responsibility, and real human outcomes. Their experience offers essential insight into where AI systems can succeed, where they may fail, and where caution is required. At the same time, early-stage innovators and startups are often overlooked. Although they operate with limited resources, they are frequently at the forefront of disruptive solutions. Including their perspective is key to ensuring that governance frameworks remain practical and do not unintentionally hinder innovation. Moreover, emerging economies and diverse cultural contexts are not sufficiently represented. Given that AI systems are deployed globally, governance must reflect a wider range of social, ethical, and linguistic realities in order to remain legitimate and effective. We know that interdisciplinary profiles—those working across science, economics, ethics, and technology—are still underutilised, despite the fact that many of today's most impactful innovations emerge at the intersection of these fields. To address these gaps, the AI Dialogue should move beyond formal representation and actively integrate lived experience and applied knowledge. This could involve embedding practitioners into thematic working groups, creating structured entry points for startups, and developing regional dialogues that feed into global decision-making. Ultimately, inclusion should be understood not as a procedural obligation, but as a driver of both better governance and meaningful innovation.

To foster meaningful and dynamic engagement, the AI Dialogue should move beyond traditional formats and adopt more interactive, inclusive, and practice-oriented approaches. First, incorporating case-based policy labs would allow participants to work collaboratively on real-world scenarios—such as AI in healthcare or public services. By engaging with concrete use cases, stakeholders can better understand trade-offs, identify risks, and co-develop actionable solutions. This format helps bridge the gap between abstract principles and implementation. It interesting to notice that co-creation sessions could bring together policymakers, researchers, industry representatives, and civil society in small, diverse groups. Structured around specific challenges, these sessions would encourage dialogue across disciplines and ensure that different perspectives are not only heard, but actively integrated into outcomes. To enhance inclusivity, the Dialogue could introduce rotating regional forums and hybrid participation models, enabling contributions from stakeholders who may not be physically present. Digital platforms can support asynchronous input, ensuring that voices from different time zones and resource contexts are meaningfully included. Another innovative approach would be the use of "reverse panels", where decision-makers primarily listen while practitioners, community representatives, and early-stage innovators share lived experiences and practical insights. This can help rebalance traditional power dynamics and surface perspectives that are often overlooked. Finally, the Dialogue would benefit from iterative engagement cycles, combining plenary sessions with ongoing working groups and follow-up collaborations. This ensures continuity, accountability, and the ability to refine ideas over time. In essence, the most effective formats will be those that combine structure with openness—creating a space where diverse actors can engage not only in discussion, but in shared problem-solving and collective learning.

Effective AI governance is best illustrated through concrete applications where ethical principles are translated into real-world impact. In healthcare, patient-centred digital platforms-such as AI-enabled monitoring applications for chronic conditions (e.g., wound care or post-operative recovery tools)-demonstrate how human oversight, safety, and usability can be embedded into system design. For instance, applications that allow patients to track symptoms while clinicians retain decision-making authority reflect a balanced model of augmentation rather than replacement. Such approaches ensure both clinical reliability and patient trust. In biomedical research, AI-driven platforms for protein structure prediction and drug discovery-as seen in recent breakthroughs in protein-folding mechanisms-highlight the transformative potential of AI when combined with strong scientific validation frameworks. These systems accelerate discovery while relying on reproducibility, peer review, and data transparency, offering a model for responsible innovation in high-impact domains such as oncology and personalised medicine. Within the pharmaceutical industry, AI-supported quality control systems provide another example. Machine learning models are increasingly used to monitor manufacturing processes, detect anomalies, and ensure compliance with regulatory standards. When combined with audit trails and explainability requirements, these systems enhance both efficiency and accountability. More broadly, regulatory sandboxes-such as those developed in Europe-offer a practical governance tool. They allow innovators to test AI systems in controlled environments under regulatory supervision, enabling iterative learning while managing risks. We observe the fact that collaborative platforms that bring together academia, industry, and public institutions-particularly in ecosystems like Switzerland-illustrate the value of interdisciplinary governance models. These platforms facilitate knowledge exchange, align technical development with ethical standards, and support scalable, trustworthy solutions. Together, these examples show that effective AI governance emerges not only from principles, but from integrated systems where innovation, oversight, and human values evolve in parallel.