PAHO

Establish a shared understanding that the responsible use of artificial intelligence must advance human well-being, trust, and equity—especially across countries with different levels of digital and regulatory capacity.

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

Working in public health at regional level, I see both the promise of AI to strengthen surveillance, inform emergency response and improve access to health services, and the risk that poorly governed systems could amplify existing social and health inequities. Prioritizing "safe, secure and trustworthy AI" is therefore essential: without robust safeguards, evaluation, and risk management, AI deployments in sensitive domains such as health, social protection, or migration can cause real harm to people and communities. I also selected the "social, economic, ethical, cultural, linguistic and technical implications of AI" because governance cannot focus only on technical performance. It must address how AI systems interact with context: whose data are used, which languages and populations are represented, and how power imbalances shape design and deployment. This is particularly important for countries and communities that have historically been under-represented in digital innovation. The "protection and promotion of human rights" and "transparency, accountability and human oversight" are foundational conditions for building public trust. Human rights norms provide a shared baseline for evaluating AI impacts, while transparency and accountability mechanisms ensure that affected persons and institutions can understand, contest and remedy harmful outcomes. Together, these priorities create an enabling environment in which AI can contribute to public goods, including better health, without undermining rights or deepening inequalities.

From where I sit in public health, there are a few cross-cutting issues that don't quite come through in the existing themes. First, AI governance needs to be much more grounded in how health and social protection systems actually work. We are talking about highly sensitive data, used in messy real-world settings, with people who already face structural inequities. If AI systems used for surveillance, triage or resource allocation are poorly designed or badly governed, they can quickly erode trust in institutions and make inequalities worse, especially in low- and middle-income countries. Second, I would highlight the full data and model lifecycle, across borders. We need to think not only about transparency in the abstract, but about who collects the data, who labels it, whose languages and realities are represented, and how models are monitored and updated once deployed in critical services. Third, I see Global South leadership and long-term capacity as an emerging issue in its own right, not just "capacity-building". Governance will be stronger if countries in the South help set the standards, lead research, and shape how AI is used in their own health systems, rather than simply importing technologies and rules. Finally, AI is evolving quickly. The Dialogue should therefore build in ways to learn from experience, adjust course, and respond to new risks over time, instead of treating governance as a one-off exercise.

In the Americas, AI is arriving in health faster than our ability to govern it. From my vantage point working with countries on data, analytics and products, the main challenges cut across the themes I selected. First, there is a clear trust and accountability gap. Many institutions are piloting AI for surveillance, triage or targeting of benefits, but few have robust processes for validation, documentation, or ongoing monitoring. When a model is a "black box", it is very hard for programmes, patients or communities to understand or challenge its decisions, which undermines confidence in already‑fragile systems. Second, capacity is extremely uneven. Some countries and large institutions can invest in governance, quality assurance and local adaptation; others simply adopt tools developed elsewhere, trained on very different populations and data. That creates a risk that AI quietly reproduces old biases and inequities, while the benefits accrue mainly to those who already have more resources. Despite this, the opportunity is real. Used well, AI can help us detect outbreaks earlier, anticipate service gaps, and use scarce human and financial resources more intelligently. The priority for our region is to close the gap between rapid uptake and slower governance, through shared standards, practical guidance, and spaces where countries can learn from each other's successes and failures.

From my perspective in regional public health, the main value of the dialogue lies in its potential to bring coherence to an increasingly fragmented landscape of AI guidance, frameworks, and standards, and to translate these into something usable for countries. The Dialogue could help in three broad ways. First, by mapping and distilling existing efforts into a small set of practical, interoperable reference points that countries can align with when developing their own policies and regulations. Second, by creating spaces where regulators, technical experts and practitioners from different regions can work through real use cases together – for example, AI in surveillance or clinical decision support – and jointly identify good practices and common pitfalls. Third, by making sure that low‑ and middle‑income countries, and regional organizations that support them, are involved not only as "implementers" but as co‑designers of norms and tools. If it can do this, the Dialogue could move cooperation on AI governance from high‑level statements to something that actually helps decision‑makers who are trying to govern AI in complex health and social systems.

There is already a dense ecosystem around AI and digital governance: UN agencies working on digital public infrastructure and data protection, sectoral initiatives in health and education, regional digital strategies, and technical standards work. These efforts have generated useful norms and tools, but they are often siloed and hard for policymakers to navigate. The AI Dialogue should build on, not duplicate, this work. It can act as a hub that maps key initiatives, highlights synergies and tensions, and makes practical resources easier to find and reuse. It can also provide a common table where different sectors compare how principles like safety, human rights and transparency are being implemented, and where regional and South‑South collaborations bring in grounded experience. The added value of the Dialogue would be its convening power and its ability to connect silos: across sectors, regions, and institutions.

From my perspective working with countries in the Americas, the Dialogue will work best if it mirrors the systems we are trying to govern. Governments should lead on setting priorities and translating outcomes into law and policy. UN entities and regional organizations can bring comparative evidence and help countries test what works in real programmes. Industry and technical experts should contribute concrete use cases, technical constraints and impact assessments, not just high‑level principles. Civil society, communities and frontline workers need structured spaces to share how AI is affecting them in practice. In terms of format, I would favour a small number of focused workstreams (for example, health and social protection; education; humanitarian action), each with diverse stakeholder participation and a mix of in‑person and virtual sessions.

I see three groups that are still under‑represented. First, public sector implementers in low‑ and middle‑income countries – the people in ministries, social security institutions or surveillance units who are actually responsible for data and AI projects. Second, communities who are heavily surveilled or scored by algorithms (for example, migrants, informal workers, or people living in marginalized urban and rural areas). Third, frontline health and social workers who must use AI‑enabled tools and explain them to the public. To include these perspectives, the Dialogue could fund targeted regional consultations and embed practitioners and community representatives as full members of working groups. Partnering with regional organizations and existing networks can help identify people who bring hands‑on experience, not only institutional titles.

1. Scenario‑based exercises where mixed groups (governments, communities, industry, UN agencies) work through realistic AI use cases – for example, an outbreak detection system or targeting of social benefits – and surface practical governance questions. 2. Reverse briefings" where frontline workers, civil society and community representatives present to policymakers and experts on how AI deployments are playing out on the ground. 3.Structured peer‑learning clinics, organized regionally, where countries present real projects and receive feedback on governance, safeguards and evaluation plans. All of these can be supported by simple digital tools for pre‑meeting surveys, anonymous inputs and follow‑up, so that people who cannot travel still shape the agenda. This would help keep the Dialogue grounded in actual practice rather than only in high‑level statements.

https://www.paho.org/en/tag/artificial-intelligence