experta en inteligencia artificial

The first Global Dialogue on AI Governance will be a success if it achieves something that global frameworks have systematically avoided: listening to those who have the most to lose when these technologies arrive without mediation, without context, and without accountability. As the author of Colombia's Ethical and Sustainable AI Adoption Roadmap —which became the main technical input for CONPES 4144 of 2025, the national artificial intelligence policy— I can affirm that AI governance is not a technical debate. It is a debate about who has the right to participate in the decisions that these technologies are already making about communities that were never consulted to build them. A successful dialogue would produce at least three concrete outcomes. First, an explicit recognition that closing capacity gaps does not mean exporting the same Global North models to developing countries with better user manuals. It means funding situated research, rooted in local epistemologies and in the languages of the communities that will inhabit them. Second, a real commitment that human oversight mechanisms for AI include voices from ethnic and rural communities and historically marginalized territories — voices that are currently absent from every space where these decisions are made. Third, clear guidelines on interoperability that do not penalize countries that opt for open-source and low-computational models, which are often the only viable option for territories with precarious connectivity. AI governance cannot continue to be designed exclusively by those who already have guaranteed access to it. The success of this dialogue will be measured by how much it changes that equation.

• AI capacity-building
• Interoperability of governance approaches
• Safe, secure and trustworthy AI
• Transparency, accountability, and human oversight

Selecciono las cuatro áreas temáticas, porque desde mi experiencia construyendo política pública de IA en Colombia, ninguna puede abordarse de manera aislada. Sin embargo, explico el orden de urgencia desde la realidad de los países del Sur Global. El desarrollo de capacidades en IA es la prioridad más urgente. No porque las demás sean menos importantes, sino porque sin capacidad instalada en los territorios, todas las demás áreas se convierten en debates que ocurren sin nosotros. Mi contribución a la construcción del CONPES 4144 de 2025 como política pública nacional de inteligencia artificial en Colombia me permitió comprobar que los avances reales en gobernanza de IA ocurren cuando hay personas que construyen conocimiento técnico y político desde adentro de sus países y sus territorios. Ese proceso debe replicarse en comunidades y naciones que hoy no tienen voz en estos espacios. La transparencia, responsabilidad y supervisión humana es la segunda prioridad. Los sistemas de IA ya están tomando decisiones sobre poblaciones vulnerables, en educación, salud y seguridad, sin que esas poblaciones tengan mecanismos reales para cuestionarlos. La supervisión humana no puede ser un principio declarativo, debe traducirse en estructuras concretas de participación ciudadana. La IA segura, protegida y fiable cobra sentido real cuando se pregunta: ¿segura para quién? Los estándares de seguridad diseñados en contextos de alta conectividad y estabilidad institucional no responden a las realidades de territorios con infraestructura precaria. Esta área temática debe incorporar una perspectiva territorial diferenciada. La interoperabilidad de los enfoques de gobernanza es fundamental para que países como Colombia puedan construir marcos propios sin quedar subordinados a estándares diseñados en otros contextos. Interoperabilidad real significa reconocer que existen múltiples formas legítimas de gobernar la IA, y que el multilateralismo en este campo solo funciona si parte de esa pluralidad, no de la imposición de un modelo único.

In Colombia, as in much of the Global South, governance gaps are not technical problems. They are democratic deficits. Safe, secure and trustworthy AI: We have ethical frameworks — like the roadmap I built, which became the main input for CONPES 4144 of 2025 — but there is no binding regulation for high-risk systems. Predictive algorithms affect decisions in education, credit, and public services with no accountability mechanism and no real right of appeal for the people affected. Capacity development: Training courses designed in the North do not solve local gaps. Colombia has more than 70 indigenous languages and not a single public dataset in those languages. Real capacity means funding situated research, rooted in local epistemologies — not exporting models built in other contexts for realities that were never consulted. Transparency, accountability and human oversight: Algorithmic decisions are presented as neutral to legitimize outcomes that affect vulnerable communities. When something fails, there is no transparency, no person with real authority to intervene, and no mechanism for redress. Human oversight cannot exist only in policy documents. Interoperability of governance approaches: Interoperability cannot mean compliance with Northern frameworks. Colombia built its own roadmap and its own national AI policy. Real interoperability means mutual recognition of different approaches, not forced convergence toward a single model. The communities most affected by AI decisions — rural, indigenous, Afro-descendant — are systematically excluded from the spaces where those decisions are made. Closing governance gaps means closing participation gaps. That is the real measure of success.

The Dialogue can play a role that no other forum has achieved so far: legitimizing the Global South as a producer, not just a consumer, of AI governance norms. Too often, international cooperation means Northern frameworks that Southern countries are expected to adopt. The result is a one-way street: standards designed in high-resource contexts, for high-resource contexts, with no room for local adaptation. The Dialogue can change this by doing three things. First, establish a binding rule that at least half of all speakers, panelists, and contributors come from developing countries, indigenous communities, and marginalized territories. Representation cannot be decorative. Second, create a dedicated fund for Global South-led research on AI governance , not for implementing Northern models, but for producing situated knowledge, datasets in local languages, and low-computational architectures that work in low-connectivity environments. Third, promote a framework of equivalency, not convergence. Countries like Colombia have already built their own AI roadmap and national AI policy. International cooperation should recognize these domestic frameworks as valid alternatives, not force compliance with external standards. The Dialogue will succeed if it produces fewer declarations and more mechanisms that redistribute the power to govern AI. The countries and communities most affected by AI decisions must have a seat at the table, not as advisors, but as decision-makers.

The Global Dialogue on AI Governance does not start from zero. It must build on existing mechanisms to avoid duplication and accelerate real impact. First, the Independent International Scientific Panel on AI. This Panel was created alongside the Dialogue under the Global Digital Compact. It will produce annual, evidence based assessments of where we stand with artificial intelligence. The Dialogue can take those scientific findings and translate them into political action, giving them real weight through intergovernmental consensus. That is a unique role that no other forum can play. Second, UNESCO's Women4Ethical AI platform. This platform brings together 17 leading female experts from academia, civil society, and the private sector. Their goal is to advance gender equality in AI governance. The Dialogue should give this platform a permanent seat at the table. Gender perspectives cannot be an afterthought. They must be central to every discussion. Third, the Global Partnership on Artificial Intelligence, or GPAI. Hosted by the OECD and launched by the G7, GPAI brings together more than 40 countries to shape international AI standards. The Dialogue can bridge the technical work of GPAI with the universal legitimacy of the United Nations. That connection does not exist today. Fourth, regional frameworks like the São Paulo Charter on AI Governance. This Charter was developed by CEBRI in Brazil. It outlines principles for responsible AI governance in Latin America, including human rights, transparency, and accountability. The Dialogue should recognize these regional efforts as valid governance frameworks, not force everyone to converge toward a single model. Fifth, UNESCO's Recommendation on the Ethics of AI. This Recommendation was adopted unanimously by 193 Member States in 2021. It remains the only universal normative instrument on AI ethics. The Dialogue should not reopen it. Instead, it should focus on accelerating its implementation, especially in developing countries. The unique value of the Dialogue is not creating new norms from scratch. It is connecting what already exists. It can link science, gender expertise, technical standards, regional frameworks, and universal principles into one coherent ecosystem. The success of this Dialogue will be measured by how well it listens to voices from the Global South. Voices like mine, from Tumaco, Colombia. Voices that have been systematically excluded from designing the AI systems that now govern their lives.

Different stakeholders bring different strengths. The key is designing a structure where each can contribute what they do best. Governments can commit to binding rules on high risk AI systems, not just voluntary principles. They can also allocate domestic resources to implement international agreements at the local level. Without implementation, declarations are just words. The private sector and technical community can open their black boxes. They can provide independent auditors with access to datasets and algorithms. They can also invest in low computational models that work in low connectivity environments, not just in data centers. Civil society and academia can bring the voices that are systematically excluded from governance spaces. Indigenous communities, rural populations, and people affected by algorithmic decisions must have direct representation, not just be studied or cited. As for format and structure, I have three recommendations. First, ensure that at least half of all speakers and panelists come from developing countries, indigenous communities, and marginalized territories. Representation cannot be symbolic. Second, create parallel working groups organized by region. Global dialogues often become Northern conversations with Southern observers. Regional groups would allow Latin America, Africa, and Asia to develop positions grounded in their own realities before bringing them to the plenary. Third, dedicate specific sessions to listening, not just presenting. Too many dialogues are a series of monologues. Set aside time for affected communities to share their experiences without interruption or filtering. The Dialogue will succeed if it stops treating the Global South as an audience and starts treating it as a co author of the rules that will govern all of us.

The underrepresented voices are not hard to identify. They are the ones systematically excluded from technological design. Rural communities and marginalized territories are almost always absent. When they do appear, it is as objects of study or beneficiaries of programs designed elsewhere. They do not need someone to speak for them. They need a seat at the table. Indigenous and Afro descendant peoples hold knowledge that does not fit into current datasets. Colombia has more than seventy indigenous languages, yet not a single public dataset in those languages. An AI model trained exclusively on Global North data cannot serve these communities. Including them means funding the creation of situated data repositories, rooted in their own epistemologies. Women from developing regions are also underrepresented. In global AI panels, most female voices come from Europe and North America. The women leading AI policy in Latin America, Africa, and Southeast Asia are rarely invited. People directly affected by algorithmic decisions in credit, education, or criminal justice systems are almost never consulted. Their experiences are the most valuable data that exists, but no one asks them. How to include them? Through concrete mechanisms. A dedicated fund to finance the participation of underrepresented communities. A binding rule that at least half of all speakers on any panel must come from the Global South. And specific sessions where affected communities share their experiences without intermediaries or filters. Including is not inviting people to watch. Including is inviting them to decide.

Traditional panel formats and prepared statements are not enough. The Dialogue needs structures that allow listening, not just speaking. First, deliberative citizen assemblies have proven their effectiveness. The Global Coalition for Inclusive AI, launched at the Paris AI Action Summit in 2025, is already organizing citizen assemblies in more than one hundred countries, with ten thousand participants who are statistically representative of their populations. These are not experts. They are ordinary citizens who learn, deliberate, and produce recommendations. The Dialogue should integrate these results as binding input, not as decorative consultation. Second, sessions with real human oversight. An innovative format is one where communities affected by algorithmic systems, for example in credit, education, or criminal justice, can test and evaluate models in real time, accompanied by experts who translate their experience into concrete recommendations. It is not about someone speaking for them. It is about them being the ones who test and question. Third, regional working groups before the plenary. Global dialogues often become Northern conversations with Southern observers. The solution is to dedicate a full day to closed sessions by region, such as Latin America, Africa, and Asia, where countries develop positions based on their own realities before bringing them to the global space. This prevents Southern voices from being diluted. Fourth, the Dialogue must set aside specific time for listening, not just for presenting. Too many international meetings are a series of monologues where each delegation reads a prepared statement. A more dynamic format is structured small group dialogues, with guiding questions and neutral facilitators, that produce written recommendations in real time. The Dialogue will succeed if it stops treating participants as an audience and starts treating them as coauthors of the rules that will govern all of us.

I can share the example I know from the inside: Colombia's AI Adoption Roadmap, which I had the honor of building and which later became the main technical input for CONPES 4144 of 2025, the country's national artificial intelligence policy. This effort positioned Colombia as the fourth adopter of artificial intelligence in Latin America. What made this policy work was not just the document. It was the process. Regional dialogues were convened in different cities across the country with the participation of more than three thousand people. There was a citizen consultation with dozens of participants from various sectors. These dialogues were not a decorative exercise. Their results directly fed into the UNESCO RAM Report for Colombia, which served as the diagnostic base for the national policy. The Roadmap was not imposed from the desk of a few officials in Bogotá. It was built from the territories. CONPES 4144 of 2025 is organized around six strategic axes: ethics and governance, data and infrastructure, research and innovation, digital talent development, risk mitigation, and AI use and adoption. It includes more than one hundred concrete actions to be implemented by 2030, with an investment of approximately 479 billion Colombian pesos. The lesson is clear. Effective AI governance is not just a matter of laws and regulations. It is a matter of participatory processes, sustained investment, and putting communities at the center, not at the margins.