Independent Researcher

A successful first Global Dialogue on AI Governance should move beyond general principles and deliver a structured foundation for actionable, human-centred AI governance at the global level. First, it should establish a shared conceptual framework that clarifies the purpose of AI systems in society. This includes recognising that trustworthy AI cannot be reduced to technical performance or safety metrics alone, but must integrate fundamental rights, accountability, and human oversight as core design principles. Second, the Dialogue should produce concrete guidance on how to operationalise these principles across contexts. In particular, it should address how AI systems can support, rather than replace, human decision-making, especially in high-risk domains and in economic ecosystems dominated by SMEs, which have limited capacity to absorb technological risk. Third, it should foster meaningful multistakeholder coordination through a "quadruple helix" approach, bringing together academia, industry, governments, and civil society. This is essential not only for legitimacy, but also for ensuring that governance frameworks are grounded in real-world constraints and societal needs, including those of the cultural and creative sectors. Fourth, the Dialogue should identify priority areas for international cooperation, including mechanisms for transparency, traceability, and evaluation of AI systems, particularly in the context of generative models and their implications for knowledge infrastructures. Finally, success should be measured by its ability to catalyse sustained collaboration. This includes the creation of working groups, pilot initiatives, and shared research and policy agendas that continue beyond the Dialogue itself. In summary, the first Global Dialogue should not only align perspectives, but also define a clear pathway towards AI systems that protect rights, preserve societal value, and empower human agency.

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

I believe my selection can contribute a strongly applied, interdisciplinary perspective to the Global Dialogue on AI Governance, grounded in both technical rigor and real-world deployment of AI systems in high-risk and socially sensitive contexts. My work sits at the intersection of probabilistic AI, uncertainty modelling, causality, and explainability, with a focus on designing systems that support human decision-making under conditions of risk. Over the past years, I have applied this approach across sectors such as finance, autonomous driving, aviation, healthcare, and disinformation, where the limitations of purely technical approaches to "trustworthy AI" become immediately evident. This has led me to develop and lead a multidisciplinary applied research group integrating technical, legal, and philosophical expertise. Through this work, I have engaged directly with questions of governance, accountability, fundamental rights, and societal impact, including contributions to institutional discussions at the European level on generative AI, copyright (European Parliament briefing: https://www.europarl.europa.eu/RegData/etudes/BRIE/2025/776529/IUST_BRI(2025)776529_EN.pdf), and the broader implications of AI systems on knowledge ecosystems (Joint Declaration: https://www.safer-ai.org/u/2026/04/Joint-declaration-Moonshots-in-Reliable-Safe-and-Secure-AI_-A-Call-for-European-Leadership.pdf). In addition, my experience working with diverse stakeholders, including industry actors, SMEs (https://cambraterrassa.org/agenda/regula-ia-programa-dimplementacio-legal-i-etica-de-la-ia-a-les-pimes/), and representatives of the cultural and creative sectors (https://www.aie.es/en/la-relacion-entre-musica-e-inteligencia-artificial-generativa-a-debate-en-la-xxii-edicion-del-seminario-juridico-de-aie/) has reinforced the importance of grounding AI governance in real-world constraints. In particular, I bring a perspective that emphasizes the need for AI systems that are not only safe and compliant, but also usable, adoptable, and aligned with human agency across different economic and social contexts. I would aim to contribute to the Dialogue by bridging technical and governance perspectives (aligned with ongoing multidisciplinary, multi-institutional, and international collaboration, including a position paper entitled "AI as Reflective Civic Infrastructural Shift. Towards a Distributed Renaissance of Human Agency"), helping translate complex AI capabilities and limitations into actionable, human-centred governance approaches that can scale globally while remaining sensitive to local realities.

One critical cross-cutting issue that remains insufficiently captured is the need to move from a predominantly risk-mitigation paradigm towards a more constructive and purpose-driven understanding of AI systems. Current governance approaches often focus on preventing harm (safety, robustness, and compliance) which are necessary but not sufficient. A key emerging challenge is how to design AI systems that actively support and strengthen human agency, decision-making, and responsibility. This requires integrating technical, legal, and societal perspectives already at the design stage, rather than treating governance as a downstream constraint. A second cross-cutting issue concerns the role of uncertainty. While reliability and performance are widely discussed, the ability of AI systems to represent and communicate uncertainty (and the implications this has for human oversight, accountability, and decision-making) remains underdeveloped in governance frameworks. Without this, human supervision risks becoming formal rather than effective. Third, there is a structural gap related to adoption capacity across different economic actors. In particular, SMEs and micro-enterprises, which represent the majority of economic activity in many regions, face limited capacity to absorb technological and regulatory risk. Governance frameworks that do not explicitly account for this asymmetry risk reinforcing existing inequalities and limiting meaningful adoption. Finally, an emerging issue concerns the cultural and knowledge ecosystems. Generative AI is not only a technical development but a transformation of how knowledge, creativity, and authorship are produced and sustained. Governance must therefore address not only protection (e.g., copyright), but also how to preserve and evolve these ecosystems in a way that remains socially sustainable. Addressing these cross-cutting dimensions is essential to ensure that AI governance frameworks remain both effective and aligned with broader societal objectives.

In Europe, and particularly in countries with economic structures dominated by SMEs, current AI governance developments are creating both important opportunities and structural gaps. A first major gap lies in the operationalisation of governance frameworks. While regulatory initiatives such as the EU AI Act represent a significant step forward, there remains a disconnect between high-level requirements and their practical implementation. Many organisations, especially SMEs, lack the technical, legal, and organisational capacity to translate principles such as transparency, risk management, or human oversight into deployable systems. This creates a risk of formal compliance without effective trustworthiness. A second gap concerns the limited integration of technical and governance perspectives. Advances in areas such as uncertainty modelling, explainability, and causal reasoning are highly relevant for building trustworthy AI, yet they are not systematically embedded in governance frameworks. As a result, critical aspects such as the ability of systems to communicate uncertainty or support meaningful human oversight remain underdeveloped. Third, generative AI is exposing new tensions in cultural and knowledge ecosystems. Current governance approaches are still largely reactive, focusing on issues such as copyright and data use, while lacking forward-looking strategies to ensure the sustainability of creative sectors and the integrity of knowledge production systems. At the same time, these gaps create opportunities. Europe has the potential to lead in defining a model of AI governance that is not only risk-oriented, but purpose-driven. By aligning technical innovation with fundamental rights, cultural sustainability, and economic realities, particularly the needs of SMEs, it is possible to foster AI systems that are both trustworthy and widely adoptable. Addressing these challenges requires stronger coordination across disciplines and stakeholders, enabling governance frameworks that are both practically implementable and socially aligned.

The Global Dialogue on AI Governance can play a pivotal role in advancing international cooperation by moving from fragmented, principle-based discussions towards coordinated, actionable frameworks grounded in real-world deployment and shared societal objectives. First, it can serve as a space to align not only regulatory approaches, but also the underlying purpose of AI systems. International cooperation requires more than harmonising rules; it requires a shared understanding of what AI should achieve in society, particularly in relation to human agency, fundamental rights, and long-term societal sustainability. Second, the Dialogue can facilitate the integration of technical and governance perspectives across jurisdictions. Advances in areas such as uncertainty modelling, explainability, and safety are critical for trustworthy AI, yet they are unevenly understood and applied globally. By fostering a common language between technical experts and policymakers, the Dialogue can help translate these capabilities into interoperable governance practices. Third, it can strengthen multistakeholder coordination through a "quadruple helix" approach, bringing together governments, academia, industry, and civil society. This is essential to ensure that governance frameworks are not only aligned internationally, but also grounded in diverse economic and social realities, including those of SMEs and cultural and creative ecosystems. Fourth, the Dialogue can identify priority areas for cooperation, such as transparency, traceability, evaluation of AI systems, and the governance of generative models, where cross-border implications are particularly significant. Finally, its most important contribution may be to catalyse sustained collaboration beyond the Dialogue itself, through working groups, shared research agendas, and pilot initiatives. In doing so, it can help build governance models that are not only internationally coordinated, but also practically implementable and socially aligned.

The Global Dialogue on AI Governance should build upon and connect existing international, regional, and sectoral initiatives, while providing a unifying layer that translates fragmented efforts into coherent and actionable governance frameworks. Key initiatives to build upon include the OECD AI Principles, UNESCO's Recommendation on the Ethics of AI, the EU AI Act, and ongoing work within the G7, GPAI, and standardisation bodies such as ISO and CEN-CENELEC (e.g., we are closely working with the chair of JTC21). These initiatives provide valuable foundations in terms of principles, regulatory approaches, and technical standards. In addition, domain-specific efforts in high-risk sectors (e.g., aviation, healthcare, finance) and emerging practices around generative AI governance and copyright are critical sources of practical knowledge. However, these initiatives often operate in parallel, with limited integration between technical advances, regulatory frameworks, and real-world adoption contexts. The added value of the Global Dialogue lies in its ability to act as a coordination and translation mechanism across these layers. First, it can align high-level principles with operational practices by bridging technical and governance perspectives, ensuring that advances in areas such as uncertainty modelling, explainability, and system evaluation inform governance frameworks in a consistent and interoperable way. Second, it can foster structured multistakeholder collaboration through a "quadruple helix" approach, connecting academia, industry, governments, and civil society. This is particularly important to incorporate perspectives from SMEs and cultural and creative sectors, which are often underrepresented despite being central to sustainable AI adoption. Third, the Dialogue can create shared spaces for pilot initiatives and experimentation, enabling the testing of governance approaches across jurisdictions and sectors. Ultimately, its added value is to transform a landscape of dispersed initiatives into a coordinated ecosystem, capable of supporting AI systems that are not only compliant, but trustworthy, adoptable, and aligned with broader societal objectives

Different stakeholders can contribute to the AI Dialogue by bringing complementary forms of knowledge, experience, and constraints, which are all necessary to design governance frameworks that are both robust and implementable. Academia contributes by advancing the scientific and technical foundations of trustworthy AI, including areas such as uncertainty modelling, explainability, and system evaluation. It can also provide critical perspectives on long-term societal implications. Industry contributes practical experience in deploying AI systems at scale, identifying operational challenges, and translating governance requirements into real-world applications. Governments play a central role in defining regulatory frameworks, ensuring alignment with public interests, and coordinating international efforts. Civil society brings essential perspectives on societal impact, fundamental rights, and inclusiveness, particularly for groups that may be disproportionately affected. To effectively integrate these contributions, the AI Dialogue should adopt a structured "quadruple helix" approach, ensuring balanced and continuous interaction between these stakeholders. In terms of format and structure, three elements are particularly important. First, the Dialogue should combine high-level discussions with working groups focused on specific themes (e.g., generative AI, high-risk systems, cultural and creative sectors), enabling both strategic alignment and technical depth. Second, it should include pilot initiatives and case-based exchanges, allowing stakeholders to test governance approaches in real-world contexts and across jurisdictions. Third, it should establish mechanisms for continuity, such as recurring working groups and shared research and policy agendas, to ensure that the Dialogue leads to sustained collaboration rather than isolated exchanges. Additionally, particular attention should be given to the inclusion of SMEs and actors from cultural and creative ecosystems, whose perspectives are often underrepresented but are critical for ensuring that AI governance is both socially sustainable and broadly adoptable. A well-structured Dialogue should ultimately function not only as a forum for discussion, but as a platform for coordinated action and long-term cooperation.

Several key voices and perspectives remain underrepresented in global discussions on AI governance, despite being critical for ensuring that governance frameworks are both effective and socially aligned. First, small and medium-sized enterprises (SMEs) and micro-enterprises are often underrepresented. Given that they constitute the majority of economic activity in many regions, their limited capacity to absorb technological and regulatory risk is a central factor in the real-world adoption of AI. Without their inclusion, governance frameworks risk being either impractical or exclusionary. Second, the cultural and creative sectors remain insufficiently integrated into governance discussions. These sectors are not only affected by AI, particularly in the context of generative models, but are also essential for shaping how knowledge, authorship, and creativity evolve. Their perspectives are crucial for ensuring the sustainability of knowledge ecosystems and maintaining societal trust. Third, interdisciplinary perspectives, particularly from philosophy, ethics, and social sciences, are often included only at a superficial level. However, these disciplines are essential to address questions of responsibility, human agency, and long-term societal impact, which cannot be resolved through technical or regulatory approaches alone. Additionally, practitioners working directly in high-risk domains (e.g., healthcare, finance, public safety) are not always sufficiently represented, despite their direct experience with the limitations and implications of AI systems in real-world settings. To address these gaps, inclusion must move beyond representation towards structured participation. This includes creating dedicated stakeholder tracks, ensuring balanced representation within working groups, and supporting capacity-building initiatives that enable meaningful engagement, particularly for SMEs and civil society actors. Furthermore, interdisciplinary collaboration should be embedded at the design stage of governance processes, rather than treated as an external input. Ensuring the inclusion of these perspectives is essential for developing AI governance frameworks that are not only globally coordinated, but also practically implementable and socially sustainable.

Innovative engagement formats for the AI Dialogue should move beyond traditional panel discussions and enable structured, interdisciplinary, and practice-oriented interaction that reflects the complexity of AI governance. First, case-based and scenario-driven sessions can foster meaningful engagement by grounding discussions in real-world contexts. These sessions should bring together stakeholders from different domains (e.g., high-risk sectors, SMEs, cultural and creative industries) to analyse concrete use cases, identify governance gaps, and co-develop actionable approaches. This helps bridge the gap between abstract principles and practical implementation. Second, multistakeholder co-creation labs can be an effective format to operationalise the "quadruple helix" approach. In these settings, participants from academia, industry, government, and civil society collaborate in small groups to address specific challenges, such as uncertainty communication, human oversight, or generative AI governance. These labs should be designed to produce tangible outputs, such as guidelines, prototypes, or policy recommendations. Third, technical-policy translation workshops are essential to connect advances in AI (e.g., explainability, uncertainty modelling, evaluation methods) with governance needs. These formats should facilitate a shared understanding between technical experts and policymakers, enabling more informed and interoperable governance frameworks. Fourth, pilot and experimentation tracks can allow stakeholders to test governance approaches across sectors and jurisdictions. By incorporating iterative feedback loops, these formats can help identify what works in practice and support scalable solutions. Finally, continuous and hybrid engagement mechanisms, including digital platforms and recurring working groups, are key to maintaining momentum beyond single events and ensuring sustained collaboration. Together, these formats can transform the Dialogue into an active space for co-design and implementation, fostering governance approaches that are not only inclusive and dynamic, but also actionable and aligned with real-world needs.

Several existing policies, practices, and approaches provide valuable foundations for effective AI governance, particularly when they combine regulatory frameworks with technical and operational mechanisms. At the policy level, the EU AI Act represents a significant step by introducing a risk-based approach and explicit obligations for high-risk systems. Similarly, UNESCO's Recommendation on the Ethics of AI and the OECD AI Principles establish globally recognised normative frameworks centred on human rights, accountability, and transparency. However, their effectiveness depends on their operationalisation. In this regard, impact assessment frameworks, such as Fundamental Rights Impact Assessments (FRIAs), are a promising practice. When properly implemented, they provide a structured way to evaluate risks, societal implications, and mitigation strategies before deployment. From a technical perspective, approaches that integrate uncertainty modelling, explainability, and causal reasoning offer concrete solutions to enhance trustworthiness. For example, methods that allow AI systems to quantify and communicate uncertainty can improve human oversight and decision-making, particularly in high-risk contexts. Similarly, explainability techniques grounded in causal inference can provide more reliable and actionable insights compared to purely correlational approaches. In the context of generative AI, emerging practices around traceability, attribution, and evaluation of outputs are essential to address challenges related to copyright, knowledge ecosystems, and accountability. These approaches remain under development but are critical for future governance. Additionally, multidisciplinary and multistakeholder collaboration models, such as "quadruple helix" approaches, are increasingly recognised as effective mechanisms to align technical innovation with legal, societal, and economic considerations. Finally, practical initiatives that support SMEs in adopting AI responsibly, including training programmes and implementation frameworks, are key to ensuring that governance is not only robust, but also inclusive and scalable. Together, these examples highlight that effective AI governance requires the integration of policy, technical methods, and real-world implementation practices.