Centro de Rehabilitacion Neurotecnologica

A successful first Global Dialogue on AI Governance should not be measured by declarations alone, but by whether it meaningfully bridges the gap between how AI is perceived, used, and governed in practice. First, success would mean grounding governance in real human experience. Evidence from large-scale interview research (e.g., Anthropic's 1,250 professional interviews) shows that people are broadly optimistic about AI's productivity benefits, yet simultaneously anxious about job security, identity, and trust. A meaningful outcome would be governance frameworks that explicitly address this duality—supporting augmentation while mitigating displacement and psychological uncertainty. Second, the dialogue should confront the perception–reality gap in AI use. Workers report primarily "augmenting" their work, yet actual usage shows much higher automation. This mismatch has deep implications: policies built on incorrect assumptions risk failure. A successful dialogue would therefore commit to evidence-based governance, including continuous measurement of real-world AI deployment and impacts. Third, success would involve establishing trust and accountability mechanisms. Across sectors—especially in science—adoption is limited not by capability but by verification costs and lack of trust. Outcomes should include standards for auditing, validation, and transparency (e.g., risk reporting, third-party evaluation), ensuring AI systems are reliable enough for high-stakes use. Fourth, it should embed human-centered and participatory governance. Workers want to retain control over meaningful tasks while delegating routine ones, signaling a desire for agency in human–AI collaboration. A successful dialogue would include mechanisms for public input and iterative feedback loops between users, developers, and regulators. Finally, success would mean operational commitments, not just principles: clear timelines, shared safety benchmarks, and international coordination to avoid fragmented regulation. In short, the dialogue succeeds if it transforms governance from abstract principles into adaptive, evidence-based, and human-centered systems aligned with how AI is actually reshaping society.

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

Transparency, accountability, and human oversight is a priority because AI systems are already embedded in high-stakes decisions. Without auditability, clear responsibility, and human-in-the-loop mechanisms, governance cannot be operationalized. This is essential to ensure that systems can be evaluated, challenged, and corrected in real-world contexts. Protection and promotion of human rights is equally urgent. AI can amplify existing inequalities or create new forms of exclusion if left unchecked. Safeguarding rights such as privacy, non-discrimination, and freedom of expression ensures that technological progress does not come at the expense of fundamental societal values. Safe, secure and trustworthy AI focuses on the technical and procedural backbone of governance. Trust must be earned through rigorous testing, continuous monitoring, and resilience against misuse. Without safety and reliability, adoption in critical sectors-such as healthcare, education, and public administration-will remain limited or risky. Finally, social, economic, ethical, cultural, linguistic and technical implications of AI are central because governance must reflect how AI is actually experienced by people. Evidence from large-scale qualitative research shows a gap between perceived and real AI use, particularly regarding automation and job transformation. Addressing these impacts requires inclusive, context-aware policies that consider diverse societies and labor realities. Together, these priorities emphasize that effective AI governance must be accountable, rights-based, technically robust, and grounded in lived human experience.

Yes-while the listed themes are comprehensive, several cross-cutting and emerging issues deserve explicit attention: 1. Human-AI psychological and cognitive impact Beyond economic effects, AI is reshaping how people think, learn, and perceive their own agency. Emerging evidence from large-scale interview research (such as Anthropic's work) shows growing reliance on AI for reasoning and decision support, alongside concerns about skill atrophy and loss of professional identity. Governance should therefore address cognitive dependency, not just labor displacement. 2. Measurement and evidence gaps A critical issue is the lack of reliable, real-time data on how AI is actually used. There is often a mismatch between reported and observed use (e.g., augmentation vs. automation). Without standardized measurement frameworks, policies risk being based on assumptions rather than reality. Continuous monitoring and shared metrics should be treated as global public infrastructure. 3. Concentration of power and infrastructure asymmetry AI capabilities are increasingly concentrated in a small number of actors with access to compute, data, and talent. This creates structural dependencies for countries and institutions, particularly in the Global South. Governance must address not only access (capacity-building) but also structural inequalities in AI infrastructure and control. 4. Human-AI collaboration design (not just safety) Most frameworks focus on risk mitigation, but less on optimizing collaboration. Questions of task allocation, decision boundaries, and human dignity in AI-augmented work are still underdeveloped. This is key to ensuring AI enhances-not erodes-meaningful human roles. 5. Temporal governance (pace of change) AI evolves faster than regulatory cycles. A major emerging issue is how to design adaptive, iterative governance systems that can respond in near real-time, rather than relying on static regulations. Together, these issues highlight the need for governance that is not only protective, but also adaptive, evidence-driven, and attentive to deeper transformations in human cognition and power structures.

In my view, governance gaps in these priority areas are already producing uneven outcomes across countries and sectors, particularly in emerging economies. A key challenge is the lack of transparency, accountability, and oversight mechanisms in real-world deployments. Many organizations are adopting AI faster than they can audit or govern it, creating risks in areas such as automated decision-making, cybersecurity, and misinformation. This is especially acute where regulatory capacity is still developing, leading to asymmetries between global technology providers and local institutions. In terms of human rights, gaps are evident in data governance and algorithmic bias. Limited local datasets and weak enforcement frameworks can result in systems that do not adequately reflect linguistic, cultural, or social diversity, increasing the risk of exclusion or discrimination. For safe, secure, and trustworthy AI, the main challenge is the absence of standardized evaluation and certification mechanisms. Organizations often lack the tools and expertise to assess system reliability or security, which slows adoption in high-stakes sectors like healthcare and public services. At the same time, the social and economic implications present both risk and opportunity. There is growing evidence that AI is augmenting work, but also silently automating tasks at a higher rate than perceived. This creates uncertainty in labor markets, particularly for knowledge workers, and exposes gaps in reskilling systems. However, these challenges also open significant opportunities. Countries and sectors that invest early in governance capacity, local talent development, and context-aware AI systems can leapfrog legacy constraints. There is also an opportunity to shape more inclusive models of AI that reflect regional priorities—particularly in language, culture, and public service delivery.

The AI Dialogue can play a pivotal role as a coordination and alignment mechanism in an otherwise fragmented global landscape. First, it can function as a bridge between principles and implementation. Many countries and organizations already endorse high-level AI ethics frameworks, but lack pathways to operationalize them. The Dialogue can facilitate convergence around practical standards—such as shared risk taxonomies, evaluation protocols, and audit requirements—making governance interoperable across jurisdictions. Second, it can enable evidence-sharing and collective learning. AI is evolving faster than most regulatory systems can track. By creating a structured space for exchanging empirical data—on real-world use, failures, and impacts—the Dialogue can reduce the current reliance on assumptions and help countries design more adaptive, evidence-based policies. Third, it can address global asymmetries in capacity and influence. Many regions lack the technical, institutional, and financial resources to fully participate in AI governance. The Dialogue can support more equitable participation by promoting capacity-building partnerships, knowledge transfer, and inclusion of underrepresented voices, ensuring that governance is not shaped solely by a small group of actors. Fourth, it can foster trust and confidence-building measures. International cooperation on AI depends on reducing uncertainty and perceived risk. Mechanisms such as voluntary disclosures, shared safety benchmarks, and coordinated incident reporting can help build mutual trust among states, companies, and civil society.

The AI Dialogue should build on and connect existing multilateral, technical, and multi-stakeholder initiatives rather than duplicating them. First, intergovernmental frameworks such as the OECD AI Principles and the UNESCO Recommendation on the Ethics of AI provide widely endorsed normative baselines. Similarly, the Global Partnership on AI has advanced practical work on responsible AI and data governance. The Dialogue can align these efforts and help translate them into interoperable policy tools. Second, emerging regulatory approaches—such as the EU AI Act—offer concrete models for risk-based governance. The Dialogue can act as a forum to compare such approaches, identify common elements, and reduce regulatory fragmentation across jurisdictions. Third, technical and safety-oriented initiatives led by organizations like ISO and research-driven efforts from companies such as Anthropic contribute to evaluation, auditing, and safety methodologies. These are essential building blocks for operational governance. Fourth, multi-stakeholder platforms like the World Economic Forum and open communities working on open-source AI and data ecosystems provide spaces for collaboration across sectors. The added value of the AI Dialogue lies in integration and coordination. It can: Bridge the gap between norms and implementation, translating principles into shared metrics and practices. Promote interoperability across regulatory regimes to avoid fragmentation. Enable inclusive participation, especially from underrepresented regions. Facilitate real-time knowledge exchange on risks, incidents, and best practices. The Dialogue's unique contribution is not to create new principles, but to connect, harmonize, and operationalize the ecosystem of existing AI governance efforts.

Different stakeholders can contribute to the AI Dialogue by leveraging their distinct capabilities, while a well-designed structure ensures these contributions translate into actionable outcomes. Stakeholder contributions: Governments should provide regulatory perspectives, public policy priorities, and national experiences, while committing to policy experimentation and sharing outcomes. Private sector actors (including leading AI developers and SMEs) can contribute technical expertise, safety practices, and real-world deployment data, including lessons from failures and risk mitigation. Academia and research organizations can offer independent evaluation, methodological rigor, and foresight on emerging risks and societal impacts. Civil society and affected communities play a critical role in representing public interest, identifying rights-based concerns, and ensuring inclusivity—particularly for marginalized groups. International organizations can act as coordinators, ensuring continuity, comparability, and alignment across initiatives. Recommended format and structure: Multi-layered architecture Combine high-level plenary sessions (for political alignment) with technical working groups focused on specific themes (e.g., safety standards, human rights, measurement frameworks). Evidence-based working tracks Each thematic area should include a requirement to present empirical evidence (case studies, usage data, incident reports), reducing reliance on abstract principles. Iterative and continuous process Move beyond one-off meetings toward a standing mechanism with regular cycles (e.g., annual summits + quarterly working sessions), enabling adaptive governance. Output-oriented design Each cycle should produce concrete deliverables: shared metrics, model policies, best-practice toolkits, or voluntary commitments. Inclusive participation mechanisms Ensure structured input from underrepresented regions and communities, including funding support and hybrid participation formats. Transparency and public communication Publish outcomes, methodologies, and progress indicators to build trust and global legitimacy.

Several voices remain underrepresented in global AI governance discussions, limiting both legitimacy and effectiveness: 1. Global South and low-resource regions Many AI governance debates are dominated by North America, Europe, and a few East Asian countries. This skews priorities toward contexts with abundant data, infrastructure, and regulatory capacity, often neglecting local needs, languages, and cultural practices. Inclusion could be achieved through funded participation, capacity-building programs, and regional hubs that feed insights directly into the Dialogue. 2. Marginalized and vulnerable communities Groups disproportionately affected by AI—such as indigenous populations, workers in precarious sectors, and people with disabilities—rarely have direct representation. Structured mechanisms like community advisory councils, participatory consultations, and co-design workshops could ensure their lived experiences inform standards and policies. 3. Non-technical disciplines While technical experts dominate, perspectives from social sciences, humanities, ethics, and law are underrepresented. These voices are crucial to anticipate societal, psychological, and cultural impacts. Inclusion could involve interdisciplinary working groups and incentives for cross-sector research collaboration. 4. Small and medium-sized enterprises (SMEs) and civil society organizations Global discussions often reflect the priorities of large tech companies. SMEs and advocacy organizations can highlight practical constraints, emerging risks, and public-interest concerns. Mechanisms such as open calls for evidence, stakeholder panels, and rotating representation can broaden participation. 5. Future-oriented perspectives AI governance debates rarely include foresight on long-term societal and existential risks. Engaging futurologists, scenario planners, and AI safety researchers ensures policies are adaptive rather than reactive. Inclusion strategies should emphasize funding support, multilingual access, hybrid participation, and formal integration of feedback into decision-making. By diversifying participation, the Dialogue can produce equitable, context-sensitive, and forward-looking governance frameworks that reflect the full spectrum of AI's global impacts.

To foster meaningful and dynamic engagement, the AI Dialogue should move beyond traditional panel discussions and adopt innovative, interactive formats that encourage collaboration, reflection, and evidence-based decision-making. 1. Thematic "Deep Dive" Workshops Small, facilitated sessions focused on specific governance challenges (e.g., auditability, human rights, safety standards) allow participants to co-create solutions. These can combine case studies, scenario exercises, and role-playing to simulate real-world consequences and trade-offs. 2. Multi-stakeholder Hackathons and Simulation Labs Bringing together policymakers, researchers, technologists, and civil society in structured problem-solving exercises encourages rapid prototyping of governance tools, such as risk assessment frameworks, auditing protocols, or transparency dashboards. Simulation labs can model AI deployment impacts across social, economic, and cultural contexts. 3. Evidence-Sharing and Data Showcases Dedicated sessions where stakeholders present empirical findings—such as AI usage patterns, bias audits, or social impact studies—help ground discussions in real-world evidence rather than abstract principles. Interactive dashboards or live polling can make data more accessible and actionable. 4. Participatory Deliberation Circles Structured small-group dialogues allow underrepresented voices—workers, indigenous communities, and SMEs—to directly influence policy recommendations. Using facilitators and digital collaboration tools ensures contributions are captured and amplified in plenary discussions. 5. Continuous Virtual Engagement Platforms A hybrid model combining in-person summits with an online platform enables ongoing collaboration between meetings. Features could include discussion forums, collaborative documents, and asynchronous workshops, ensuring that momentum and feedback loops are maintained. 6. "Policy Lab" Challenges Participants tackle specific governance questions with real-time expert mentorship and iterative feedback. Outcomes are directly linked to tangible outputs like draft guidelines, interoperable standards, or capacity-building toolkits.

Several existing policies, practices, and platforms provide concrete models for advancing effective AI governance: 1. Risk-based regulatory frameworks The EU AI Act classifies AI systems by risk level, imposing proportionate obligations for high-risk applications. This approach aligns oversight with potential societal harm, offering a structured template for governments seeking adaptable yet enforceable rules. 2. Independent auditing and certification Standards like those under ISO (e.g., ISO/IEC 42001 on AI management systems) provide mechanisms for evaluating safety, reliability, and ethical compliance. Third-party audits build trust and accountability while encouraging organizations to adopt robust internal governance practices. 3. Multi-stakeholder platforms The Global Partnership on AI demonstrates how governments, industry, academia, and civil society can collaboratively address technical and societal challenges, from fairness assessments to capacity-building, ensuring governance reflects diverse perspectives. 4. Open-source and transparent approaches OpenAI, Anthropic, and other organizations have developed public model cards, transparency reports, and benchmark datasets that allow independent evaluation, increasing system accountability and fostering cross-institutional learning. 5. Human-centered design and participatory methods Policies that integrate stakeholder input-such as community consultations, co-design workshops, or citizen assemblies-ensure AI systems respect local contexts, human rights, and social norms. This approach has been applied in municipal AI ethics guidelines in countries like Canada and Finland. 6. Iterative, adaptive governance Regulatory sandboxes, such as the UK's Financial Conduct Authority Sandbox, allow experimentation under supervision, enabling continuous learning and adjustment as technology evolves. Together, these approaches illustrate that effective AI governance requires a combination of risk-based regulation, independent oversight, inclusive participation, transparency, and adaptability, creating a system capable of safeguarding human rights, promoting trust, and fostering innovation.