EPFL / FiberLab

A successful first Global Dialogue on AI Governance should produce a shared understanding of the principles, norms and institutional mechanisms required to guide the development and deployment of artificial intelligence in a manner consistent with human dignity, moral responsibility and fundamental rights. In my view, success would not be measured only by declarations of intent, but by the creation of a practical foundation for international cooperation. This should include common ethical standards for transparency, accountability, safety, non-discrimination, privacy and human oversight, while recognising the different social, economic and cultural contexts in which AI systems are developed and used. The Dialogue should also affirm that AI must serve humanity rather than merely accelerate technological competition. Its governance should therefore be oriented towards the advancement of science, responsible innovation, sustainable development and the reduction of global inequalities. In particular, AI should be directed towards addressing urgent human challenges, including poverty, limited access to education, public health crises and the diagnosis, prevention and treatment of disease. A meaningful outcome would also be the establishment of inclusive channels through which governments, researchers, civil society, industry and representatives of vulnerable communities can participate in shaping AI governance. Without such inclusion, global standards risk reflecting only the interests of the most technologically powerful actors. Ultimately, the Dialogue would be successful if it helped transform AI governance from a fragmented and reactive discussion into a coherent, human-centred international agenda: one that protects rights, encourages innovation and ensures that the benefits of AI are distributed more fairly across societies.

• Transparency, accountability, and human oversight
• Protection and promotion of human rights
• Safe, secure and trustworthy AI

My selection is guided by the belief that the legitimacy of AI governance will ultimately depend on its capacity to protect human dignity, preserve public trust, and ensure that technological progress serves the common good. The protection and promotion of human rights must be the cornerstone of any global AI framework. As AI systems increasingly shape access to healthcare, education, employment, public services, finance and information, their deployment must be anchored in the principles of equality, non-discrimination, privacy, freedom of expression and individual autonomy. AI should expand human capabilities, not diminish human agency. Transparency, accountability and human oversight are equally essential. In domains where algorithmic systems may affect lives, rights or opportunities, decisions must be explainable, auditable and subject to clear lines of responsibility. Human oversight is not a procedural formality; it is a safeguard against the abdication of moral and legal responsibility. Safe, secure and trustworthy AI is an urgent priority because innovation without safeguards can amplify harm at scale. Robust governance must ensure that AI systems are reliable, resilient, secure against misuse, and designed with rigorous risk assessment throughout their lifecycle. Finally, the social, economic, ethical, cultural, linguistic and technical implications of AI require sustained attention. AI will not affect all societies equally. Without inclusive governance, it may reinforce existing inequalities, marginalise underrepresented languages and cultures, and concentrate power among a limited number of actors. Together, these priorities reflect a human-centred vision of AI governance: one that combines ambition with responsibility, safeguards rights while enabling innovation, and ensures that AI becomes not merely a force of efficiency, but an instrument of justice, inclusion and shared human progress.

Yes. One cross-cutting issue that deserves stronger attention is the equitable distribution of AI benefits and decision-making power. Many current debates focus on risk mitigation, safety and technical governance, which are essential. However, AI governance must also address the structural question of who controls AI infrastructure, who has access to high-quality data and computational resources, and who is able to participate meaningfully in shaping the future of this technology. Without this, AI may deepen existing asymmetries between countries, institutions and communities. A second emerging issue is the impact of AI on scientific discovery and healthcare sovereignty. AI has the potential to accelerate diagnostics, drug discovery, personalised medicine and public health responses. Yet these benefits will not be automatic. Governance frameworks should ensure that AI-enabled health innovation remains accessible, ethically validated and adapted to local needs, rather than becoming concentrated in a few technologically dominant ecosystems. A third important issue is the preservation of human judgement and moral responsibility in increasingly automated societies. As AI systems become more capable, there is a risk that institutions may defer excessively to algorithmic outputs. Governance should therefore protect not only formal human oversight, but also the cultivation of human expertise, ethical reasoning and institutional accountability. Finally, AI's environmental footprint should be treated as a cross-cutting concern. The development and deployment of large-scale AI systems require energy, infrastructure and natural resources. Responsible AI governance should therefore integrate sustainability, ensuring that technological progress does not come at the expense of climate and environmental commitments. Overall, emerging AI governance must look beyond compliance. It should address power, access, responsibility, health equity and sustainability as central conditions for a just and trustworthy AI future.

Governance gaps in AI affect Belarus and Switzerland in markedly different ways, yet both contexts demonstrate why human-centred AI governance is urgent. In Belarus, the principal challenge lies in the absence of strong democratic safeguards, independent oversight and effective protection of fundamental rights. In such environments, AI may be used not only for innovation, but also for surveillance, social control, information manipulation and the restriction of civic freedoms. Weak transparency and accountability mechanisms increase the risk that automated systems could reinforce discrimination, limit access to rights and deepen public distrust. At the same time, Belarus has a strong scientific, engineering and IT talent base. With appropriate governance, international cooperation and rights-based standards, this human capital could contribute meaningfully to healthcare, education, digital services and scientific innovation. In Switzerland, the challenges are different. Switzerland benefits from robust institutions, advanced research ecosystems, strong universities, a dynamic innovation sector and a tradition of democratic governance. However, even in such a context, AI raises complex questions around accountability, explainability, data protection, medical validation, liability and public trust, particularly in sensitive sectors such as healthcare, biotechnology and diagnostics. The risk is not only misuse, but also regulatory fragmentation, excessive caution that slows responsible innovation, or insufficient inclusion of smaller actors and underrepresented communities in AI development. For my sector — biomedical innovation and diagnostics — the opportunities are considerable. AI can accelerate disease detection, improve clinical decision-making, support personalised medicine and optimise research and development. However, these benefits depend on trustworthy data, rigorous validation, transparent models, human oversight and equitable access. The key opportunity for both Belarus and Switzerland is to ensure that AI becomes a tool for human development rather than a mechanism of control or inequality. The most significant challenge is to build governance that is not merely technically competent, but morally grounded, inclusive and capable of protecting human dignity across very different political and institutional realities.

The AI Dialogue can play a crucial role in transforming AI governance from a fragmented set of national or regional initiatives into a genuinely cooperative international agenda. At a time of growing geopolitical tension, technological competition and institutional mistrust, such a platform can provide a neutral space where states, international organisations, researchers, industry and civil society can engage in structured dialogue around shared risks and common responsibilities. Its value lies not only in developing principles, but in rebuilding trust. AI is increasingly linked to security, economic power, information integrity, military capability and strategic influence. Without international cooperation, governance gaps may deepen geopolitical divisions and encourage a race for technological dominance. The AI Dialogue can help shift this dynamic by promoting transparency, confidence-building measures, interoperability of governance approaches and shared commitments to prevent harmful uses of AI. It can also contribute to the transformation of diplomacy itself. AI governance requires a new form of diplomacy: more interdisciplinary, evidence-based, inclusive and anticipatory. Traditional diplomatic channels are often too slow to respond to the speed of technological development. The Dialogue can create mechanisms for continuous exchange, early warning, joint risk assessment and coordinated responses to emerging challenges. Importantly, the AI Dialogue can give a stronger voice to countries and communities that are often excluded from technological decision-making. This is essential to ensure that global AI governance does not merely reflect the interests of dominant powers, but responds to the needs of humanity as a whole. Ultimately, the AI Dialogue can become a bridge between technological progress and global stability. Its success would lie in helping AI become not a new source of geopolitical confrontation, but a field for renewed multilateral cooperation, responsible innovation and a more human-centred diplomacy.

The AI Dialogue should build upon existing initiatives while addressing a critical gap: the widening technological divide between advanced and developing economies. Relevant mechanisms include the Global Digital Compact, which calls for inclusive digital cooperation and AI capacity-building for developing countries; UNESCO's Recommendation on the Ethics of Artificial Intelligence, the first global ethical standard on AI, applicable to all UNESCO Member States; and UNESCO's Readiness Assessment Methodology, which helps countries identify institutional and regulatory gaps in AI governance. The Dialogue should also connect with the ITU AI for Good platform, particularly its focus on applying AI to the Sustainable Development Goals in areas such as healthcare, education, food security and disaster risk reduction. The AI Skills Coalition, led by ITU, is especially relevant as a UN-led platform for inclusive AI education and capacity-building. At the regional level, initiatives such as Latam-GPT, an open-source model designed for Latin America, demonstrate how linguistic and cultural inclusion can strengthen technological sovereignty and reduce dependency on a small number of dominant AI ecosystems. The added value of the AI Dialogue would be to connect these fragmented efforts into a coherent global architecture. It could help align ethical principles, technical standards, financing mechanisms and capacity-building programmes, while ensuring that developing countries are not merely recipients of technology but active co-designers of AI governance. Its most important contribution would be to move beyond declarations and create practical pathways for equitable access to data, computing infrastructure, skills, research partnerships and locally adapted AI solutions. Without this, AI risks amplifying global inequality. With effective coordination, however, the Dialogue could help transform AI into a shared instrument of development, inclusion and technological justice.

Different stakeholders should contribute to the AI Dialogue through clearly defined mandates, practical deliverables and shared accountability. The Dialogue should not function merely as a diplomatic forum, but as an operational mechanism for global AI governance. Governments should define public-interest priorities, commit to minimum safeguards on human rights, safety and accountability, and share national regulatory experiences. They should also identify capacity gaps, especially in developing countries. International organisations should coordinate existing initiatives, including UNESCO's AI ethics work and Readiness Assessment Methodology, which helps countries assess preparedness for ethical and responsible AI implementation. They could also connect the Dialogue with the OECD AI Incidents Monitor, which documents AI-related incidents and hazards to support evidence-based policymaking. The private sector should provide structured transparency on AI capabilities, limitations, safety testing and deployment risks. Participation should be linked to concrete obligations: auditability, incident reporting, red-teaming, model documentation and responsible access policies. Academia should lead independent evaluation, benchmark development and interdisciplinary research on social, legal, biomedical, environmental and geopolitical risks. Civil society and affected communities should be full participants, not symbolic consultees. Human rights groups, educators, patient organisations, labour representatives, youth, linguistic minorities and developing-country actors should help define harms, priorities and accountability standards. The Dialogue could be structured around four innovative mechanisms: 1). A Global AI Governance Observatory mapping laws, risks, capacity gaps and best practices. 2). A multilateral AI incident-reporting system, inspired by existing incident-monitoring models. 3). Regulatory sandboxes for developing countries, pairing local institutions with technical and legal experts. 4). A Global AI Capacity Fund supporting access to compute, datasets, skills and public-interest AI infrastructure. Its format should combine an annual high-level summit, permanent expert working groups, regional consultations and open digital participation. The added value would be to transform AI governance from fragmented declarations into measurable cooperation, institutional trust-building and practical tools for inclusive, safe and human-centred AI.

Several voices remain underrepresented in global AI governance, despite being essential to understanding both the risks and opportunities of AI. First, young people should be more systematically included. They will live longest with the consequences of today's AI decisions and are often early users of emerging technologies. Youth councils, fellowship programmes and representation in official working groups could ensure that intergenerational justice becomes part of AI governance. Second, entrepreneurs and start-ups, especially from developing countries and smaller innovation ecosystems, are often absent from global discussions dominated by large technology companies. Their inclusion is crucial because they can bring practical insight into innovation barriers, access to capital, regulatory burdens, data availability and locally adapted AI solutions. Third, scientists and technical experts must be represented beyond computer science alone. AI governance requires cognitive scientists, psychologists, neuroscientists, sociologists, economists, educators, legal scholars, ethicists, medical researchers and public health experts. Cognitive scientists and psychologists can assess how AI affects decision-making, attention, trust, behaviour and mental health. Sociologists can analyse inequality, power and institutional impact. Economists can evaluate labour market disruption, productivity, competition and distribution of benefits. Medical and public health experts can guide safe deployment in healthcare. Fourth, affected communities should be included directly: patients, teachers, workers, linguistic minorities, migrants, persons with disabilities and communities exposed to surveillance or algorithmic discrimination. They should not be consulted only after systems are deployed, but involved in risk identification, design and evaluation. Inclusion could be achieved through funded participation, regional consultations, multilingual access, citizen assemblies, expert panels, youth forums and permanent advisory groups. The AI Dialogue should also reserve seats for underrepresented disciplines and communities, ensuring that AI governance is shaped not only by those who build AI, but also by those who are affected by it.

Different stakeholders should contribute to the AI Dialogue through clearly defined roles, measurable commitments and institutionalised channels of participation. Member States - should provide political legitimacy, articulate national and regional priorities, and commit to minimum safeguards on human rights, safety, transparency and accountability. They should also identify domestic governance gaps and share regulatory lessons, including from both advanced and developing economies. International organisations and NGOs - should act as conveners and coordinators, ensuring coherence between existing initiatives and avoiding duplication. Their role should include technical assistance, capacity-building and support for countries that lack regulatory, computational or institutional capacity. The private sector - should be required to contribute technical evidence on AI capabilities, limitations and risks. Companies should participate in structured mechanisms for model evaluation, safety testing, auditability, incident reporting and responsible deployment, particularly where AI systems affect public services, health, employment, security or information ecosystems. Academia and independent research institutions - should provide rigorous, interdisciplinary analysis. They can support evidence-based policymaking through impact assessments, risk evaluation, benchmark development, and long-term research on social, ethical, economic and geopolitical implications. Civil society and affected communities - must be integrated as substantive participants, not consulted only symbolically. Human rights organisations, patient groups, educators, labour representatives, youth, linguistic minorities and communities from developing countries should help define risks, priorities and accountability needs. The AI Dialogue should be structured around three levels. First, a high-level political forum to establish priorities and adopt commitments. Second, permanent expert working groups on human rights, AI safety, healthcare, education, sustainability, capacity-building and governance interoperability. Third, regional and sectoral consultations to ensure that global norms reflect local realities. The Dialogue should be continuous, hybrid, multilingual and publicly accountable. It should produce annual reports, track implementation, identify governance gaps and create practical tools such as model assessment templates, capacity-building roadmaps and shared incident-reporting mechanisms. Its added value would be to move AI governance from abstract principles to operational cooperation, measurable progress and inclusive global trust-building.

Effective AI governance should combine binding regulation, practical risk-management tools, independent oversight and inclusive capacity-building. Several existing policies and practices offer useful models. One important example is the EU AI Act, which introduces a risk-based regulatory approach. Its value lies in distinguishing between unacceptable, high-risk and lower-risk AI applications, rather than treating all systems alike. This approach is particularly relevant for sensitive sectors such as healthcare, employment, education, law enforcement and public services, where AI can directly affect rights, safety and access to opportunities. The Act also links innovation with safeguards, including obligations for high-risk systems and regulatory sandboxes. A second strong example is the UNESCO Recommendation on the Ethics of Artificial Intelligence, which provides a universal normative framework grounded in human rights, dignity, fairness, transparency and human oversight. Its Readiness Assessment Methodology is especially valuable because it helps countries identify institutional, legal and regulatory gaps before adopting AI at scale. The NIST AI Risk Management Framework is another practical tool. It supports organisations in identifying, assessing and managing AI risks across the lifecycle of AI systems, and is useful because it translates abstract principles such as trustworthiness, safety and accountability into operational practice. The OECD AI Principles and the OECD AI Incidents and Hazards Monitor also provide important mechanisms. The Principles offer an intergovernmental standard for trustworthy AI, while the Incidents Monitor helps build an evidence base on real-world AI harms and failures. Building on these examples, effective AI governance should include mandatory impact assessments for high-risk systems, public registries of AI use in sensitive sectors, independent audits, incident-reporting mechanisms, regulatory sandboxes, and dedicated capacity-building programmes for developing countries. The strongest approaches are those that move beyond ethical declarations and create enforceable, measurable and inclusive systems of accountability.