Universitat Pompeu Fabra UNESCO AI Ethics Without Borders & Women for Ethical AI

For me, the success of this first Global Dialogue will not be measured by consensus, but by whether it produces traction. AI governance today often oscillates between principles that are too abstract to implement and technical debates that are too narrow to guide policy. A meaningful outcome would be to bridge that gap—translating high-level commitments into concrete, testable actions that institutions can actually adopt. One way to do this is by making good practices visible and usable. Across sectors such as education, health, public administration, security, and the development of large language models, there are already valuable experiences. The challenge is not the absence of solutions, but their fragmentation. Systematising these practices and turning them into operational guidance clear enough to be implemented, flexible enough to be adapted would be a significant step forward. I would also see success in a subtle but important shift in influence. If actors who are usually underrepresented, particularly from the Global South or from non-technical disciplines manage to shape the agenda and not only contribute to it, the Dialogue will gain both legitimacy and depth. Equally important is the ability to openly address tensions: between innovation and control, speed and responsibility, economic incentives and rights. Avoiding these trade-offs does not make them disappear; it only weakens governance. Ultimately, AI governance is about deciding how power, risk, and responsibility are distributed. If this Dialogue helps define even a minimal shared framework to navigate those questions and leaves behind mechanisms to continue working on them then it will have achieved something meaningful.

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

First, transparency, accountability, and human oversight are essential to ensure that AI systems remain aligned with human rights and democratic values. In my work, I have seen how opaque systems can reproduce and amplify existing inequalities, particularly affecting women and marginalized communities. Without clear accountability mechanisms and meaningful human oversight, it becomes very difficult to contest harmful outcomes or assign responsibility. Second, the social, economic, ethical, cultural, linguistic, and technical implications of AI are not abstract concerns they are already shaping access to opportunities, public services, and information ecosystems. Coming from Latin America and working across different contexts, I consider it critical to include diverse perspectives in the design and governance of AI systems. Otherwise, we risk reinforcing global asymmetries and excluding entire populations from the benefits of these technologies. Finally, advancing safe, secure, and trustworthy AI is key to building public confidence and enabling responsible innovation. This requires not only technical robustness, but also participatory approaches, explainability, and context-sensitive governance frameworks. From a design perspective, integrating ex-ante assessments that explicitly identify and mitigate social risks can significantly reduce the likelihood of harm. For example, incorporating safeguards against misuse at early stages could help prevent the proliferation of harmful applications such as non-consensual sexual deepfakes, whose impact has been shown to disproportionately affect women and girls globally. In this sense, I believe urgent action should focus on bridging technical development with social responsibility, ensuring that AI governance is inclusive, interdisciplinary, and grounded in real-world impacts.

In my view, one important cross-cutting issue that is not sufficiently captured across these themes is the question of power asymmetries embedded in the development and deployment of AI systems. AI is not only a technical or regulatory matter, it is also a reflection of existing global inequalities. The concentration of data, computational resources, and decision-making power in a small number of actors mostly located in the Global North raises important concerns about who defines the standards, values, and priorities embedded in these technologies. In practice, many countries and communities remain positioned as users rather than co-creators. At the same time, there are several emerging impacts that require more explicit attention. One of them is the growing effect of AI-driven systems on mental health, including addictive design patterns, hyper-personalisation, and continuous exposure to synthetic or manipulated content. While other industries such as tobacco, gambling, television, and cinema have developed regulatory frameworks to mitigate harm, similar approaches are still limited in the AI domain. In addition, there is a need to more directly address harmful and under-regulated applications, such as AI in military contexts, the proliferation of non-consensual sexual deepfakes, and forms of sexualised or exploitative AI systems. These raise complex questions around responsibility, harm, and redress mechanisms that are not yet sufficiently developed Finally, the labour dimension of AI remains largely invisible. The functioning of many AI systems depends on large-scale human labour for data annotation, content moderation, and filtering of harmful material, often under precarious conditions. These hidden workers are a fundamental part of the AI value chain, yet their rights, protections, and working conditions are rarely included in governance discussions. Addressing these issues requires incorporating a clearer analysis of power, labour, and harm into AI governance, alongside more comparative learning from other regulated sectors and international agreements.

From my work across Costa Rica, Argentina, and Uruguay from Spain and Europe, it is clear that governance gaps in AI have tangible consequences, particularly in regions where the capacity to influence technological design is limited. One of the main challenges is the absence of a clear framework: countries and institutions struggle to decide whether to prioritise open systems, more regulated approaches, or a hybrid model. This uncertainty, combined with high dependence on external technologies, exposes governments and organisations to both technical and economic vulnerabilities. Another critical issue is the way data is regulated or left unregulated. Large private companies often gain access to citizen data under the promise of free services, which reinforces dependence and creates asymmetric power relationships. Small populations, such as Uruguay with only three million people, are particularly disadvantaged: large language models frequently fail to capture local culture, language, routines, and objects, resulting in incomplete or inaccurate outputs. Weak governance also limits the ability to anticipate and respond to emerging risks. Without mechanisms to detect and adapt to hazards quickly, public and private actors must often react at the pace dictated by commercial interests rather than societal needs. Security and privacy remain significant concerns, particularly as AI systems increasingly handle sensitive information without sufficient safeguards. Despite these challenges, there are clear opportunities. Regions like ours can lead in designing context-sensitive, socially grounded AI governance frameworks that reflect local realities and values. By strengthening institutional capacity, fostering regional cooperation, and combining regulatory clarity with participatory approaches, it is possible to build AI systems that are safer, more reliable, and better aligned with social priorities while also reducing dependence on external actors.

The AI Dialogue can play a key role in connecting global frameworks with local realities. By establishing shared principles that are actionable, it can help countries like Costa Rica, Chile, Argentina and Uruguay reduce dependence on external technologies, strengthen transparency, accountability, and security, and ensure AI reflects social and cultural contexts. It can also serve as a platform to exchange good practices across sectors as education, health, public administration, and language models helping participants learn from concrete experiences and adapt solutions to their own contexts. In this sense, the Dialogue can highlight power and data asymmetries and foster collaborative mechanisms, such as regional partnerships or working groups, to turn discussion into concrete action. In doing so, it can move international cooperation from abstract principles toward practical, inclusive, and socially responsible governance.

The AI Dialogue should build upon existing initiatives at global, regional, and sectoral levels. International frameworks such as the OECD AI Principles, UNESCO's Recommendation on the Ethics of AI, and the Global Partnership on AI provide foundational guidance. In Latin America particularly Costa Rica, Argentina, Chile and Uruguay regional collaborations are exploring context-sensitive AI governance in education, health, and public administration. From my experience living and working in Barcelona, Spain, I also see the value of connecting with European initiatives and regulatory approaches, which offer lessons in balancing innovation, safety, and social responsibility. The added value of the AI Dialogue is its ability to link these efforts, fostering a global conversation that remains attentive to local realities. Many existing initiatives operate in silos; the Dialogue can create space for exchanging lessons, aligning principles with practice, and amplifying underrepresented voices. It could also catalyse concrete follow-up mechanisms joint research, regional partnerships, or pilot projects bridging discussion and implementation. By connecting normative guidance, operational experience, and local context, the AI Dialogue can help reduce dependency on external technologies and support inclusive, socially grounded, and responsible AI governance globally.

Different stakeholders can contribute in ways that reflect their expertise, perspective, and experience. Governments bring legitimacy and policy authority, but they also need to listen closely to those who live the realities AI affects every day. Academia can provide evidence-based insights, risk assessments, and long-term thinking. Civil society and community organisations can highlight impacts on rights, equity, and inclusion. The private sector contributes technological knowledge and practical constraints, but also faces the responsibility to align innovation with societal needs. For the Dialogue to be effective, it should avoid the usual round-table and instead mix formats. Short plenaries can introduce themes, but the real work happens in breakout sessions, workshops, and scenario exercises where stakeholders can co-create solutions, debate trade-offs, and test ideas. Small, diverse working groups allow participants to speak freely, challenge assumptions, and surface local realities often invisible in global debates. Transparency and follow-up are key. The Dialogue could publish live summaries or shared insights in real time, so participants can see where consensus emerges and where tensions remain. It should also include mechanisms to maintain momentum afterward, like collaborative projects, regional hubs, or thematic task forces. I would recommend deliberately mixing voices and geographies in every session. Pairing private sector technologists with social scientists, local policymakers with global experts, or educators with human rights specialists can generate new perspectives and practical recommendations that wouldn't emerge otherwise. The AI Dialogue works best not as a conference of statements, but as a laboratory of ideas, where different stakeholders actively shape outcomes, experiment with approaches, and leave with tangible pathways for responsible AI governance.

In global AI governance discussions, many of the voices that matter most are still missing. People who are pursuing legal claims against large companies for harms or incidents, artists affected by misuse of their work, families of individuals who have suffered tragedy linked to technology, and those harmed by autonomous systems all bring firsthand knowledge of AI's risks. Equally invisible are workers on the frontlines content moderators, data annotators, and other human filters who face anxiety, insomnia, and trauma as part of the AI production chain. Marginalised populations, including undocumented workers, are often exploited in ways that remain hidden, yet are essential to understanding the social cost of these technologies. Bringing these voices into the conversation requires deliberate, structured approaches. Participatory workshops, consultations, and safe reporting channels can give space to those who experience AI's consequences most directly. Civil society organisations, researchers, and advocacy groups can help amplify these experiences while translating them into ethical, social, and technical considerations for policy and design. Inclusion is more than a principle, it is a practice. By ensuring that the Dialogue hears from people who have experienced harm, risk, or exploitation, we can design governance that is not abstract, but grounded in real-world consequences. This approach makes AI governance more accountable, socially responsible, and attuned to the ethical and human challenges that technology creates, rather than only the opportunities it promises.

Meaningful engagement in the AI Dialogue requires moving beyond traditional panels and presentations. To generate real insight and actionable outcomes, formats should combine dialogue, experimentation, and storytelling. One approach is interactive workshops or labs, where participants co-create solutions to real-world scenarios. For example, multi-stakeholder teams could simulate governance decisions for AI in health, education, or public administration, identifying risks, trade-offs, and mitigation strategies. These exercises turn abstract principles into concrete practice. Another powerful format is participatory storytelling. Individuals directly affected by AI content moderators, workers in AI supply chains, artists, or families impacted by technological harm can share experiences that illustrate the social, ethical, and psychological consequences of AI. These stories humanise discussions and make trade-offs tangible for policymakers, technologists, and civil society alike. The Dialogue could also incorporate rapid policy hackathons or challenge sessions, where interdisciplinary teams propose innovative governance mechanisms in real time. This encourages experimentation, cross-pollination of ideas, and a focus on implementable solutions rather than theoretical consensus. Finally, digital platforms like DECIDIM Spain can expand inclusion and continuity. Virtual consultations, live feedback mechanisms, and collaborative knowledge repositories allow voices from smaller countries, marginalised groups, or remote regions to contribute and maintain momentum beyond the event itself. In short, the Dialogue will be most effective when it becomes a laboratory rather than a conference: a space where evidence, experience, and experimentation intersect, where diverse voices actively shape decisions, and where participants leave with both insights and concrete actions to advance responsible AI governance.

Participatory platforms like Decidim allow citizens to submit problems, opinions, and solutions, and vote on proposals. This creates an active feedback loop, ensuring that AI systems reflect real societal needs and priorities. Encouraging transparency and communication between users and developers is another effective approach, allowing communities to understand AI systems, raise concerns, and suggest improvements. In the health and social sector, associations working on suicide prevention, bulimia, or anorexia have developed AI tools co-designed by psychologists, sociologists, and medical professionals. These systems detect risk behaviors early while embedding ethical, social, and medical oversight. Similarly, inclusive design in robotics and voice-based systems for example, for individuals on the autism spectrum or non-binary users ensures AI reflects diverse gender identities and social needs. In medicine, AI is increasingly applied to create personalized and precision approaches, integrating sex and gender perspectives. Systems are being used to improve care for patients with diabetes, Alzheimer's, and other conditions, showing how AI can move from generic algorithms to context-sensitive, human-centred solutions. These examples highlight that effective AI governance requires more than regulation alone. It depends on inclusive design, participatory engagement, multidisciplinary collaboration, and transparency. Platforms and policies that incorporate feedback from users, address social diversity, and embed ethical and scientific expertise offer tangible ways to develop AI responsibly while addressing real societal challenges.