Why Governance of AI Matters
Artificial intelligence is no longer a future prospect—it is already reshaping productivity, business models, and societal structures. Yet the scale, speed, and complexity of these changes often overwhelm our capacity to direct them towards good. We know that AI solutions will be key in achieving the global goals. But we also know that it brings new risks and challenges. The central question is no longer whether to govern AI, but how to do so in ways that balance innovation, inclusion, and long-term societal benefit.
International organizations, governments, businesses, and civil society face a core tension: how to advance AI as a driver of innovation, economic growth and security while mitigating its risks. Without aligned standards and safeguards, AI applications in healthcare, education, or sustainability may also deepen bias, inequality, and mistrust. This is not a technological failure, but an institutional one.
In this article we explore two examples and draw lessons for the future of AI. First, past experiences in internet governance—particularly the multistakeholder model developed over decades—offer guidance. Institutions like the ITU have shown how collaborative processes can produce standards, norms, and oversight that respond to complex technological shifts. Second, healthcare, offers a particularly instructive case: it combines high public sensitivity, complex data flows, and a wide range of actors with competing mandates. The following section explores possible ways forward, by examining how governance structures shape power and legitimacy in practice
Centralization vs. Decentralization
AI governance is being shaped under conditions very different from those that guided the internet’s early development. The internet grew out of decentralized principles: open protocols, distributed technical communities, and arenas like the Internet Governance Forum (IGF) where governments, firms, and civil society engage on relatively equal terms. By contrast, AI today is defined by concentration. A handful of firms and state-backed actors control the compute, data, and talent required to set the pace of development.
This difference matters because governance structures crystallize power relations. If the future of AI is steered primarily by a few dominant players, global governance risks becoming fragmented and unbalanced. Unilateral initiatives may advance quickly but undermine legitimacy, while multistakeholder efforts risk tokenism if they lack real authority. The São Paulo Guidelines, endorsed at NetMundial+10 in 2024, were a recent attempt to set principles for multistakeholder cooperation. Yet geopolitical tensions are affecting consensus, leaving broad and inclusive efforts like these vulnerable to being sidelined. The lesson for AI is clear: without durable mechanisms of shared stewardship, governance will either entrench inequality or collapse under competing pressures. These challenges invite us to look at traditions where diverse actors have successfully worked together to shape common rules.
Multi-Stakeholder Governance Traditions
With 20, and more, years of experience, the internet governance setup offers useful lessons in how different governance traditions can be combined. The World Summit on the Information Society (WSIS), which shared the space during the yearly AI for Good Global Summit, did not produce binding laws, yet it influenced global priorities by convening diverse actors into a deliberative space. While the Internet Corporation for Assigned Names and Numbers (ICANN), on the other hand, exercises executive authority over technical coordination. Taken together, these examples show the value of layering deliberative forums with executive bodies, where spaces for dialogue shape agendas and institutions implement concrete decisions.
Yet multistakeholder platforms are never free of tension. Inclusivity can slow processes down while efficiency can sideline important voices. Even mundane details—such as who speaks, how interventions are summarized, or how meetings are structured—carry weight. But the key lesson is that legitimacy does not come from formal authority alone. It comes from sustaining workable compromises across heterogeneous actors. For AI, this means building a governance arena that is at once open and decisive: forums to debate priorities broadly, and institutions capable of translating those priorities into action while guarding against capture by powerful interests.
Why Multi-Stakeholder Partnerships Matter (the Case of Health)
Healthcare is one of the clearest domains where AI governance cannot be left to a single actor. Health systems are inherently interdependent: patients, clinicians, regulators, hospitals, researchers, and technology providers all influence outcomes. Decisions around data, algorithms, or care delivery directly affect lives, rights, and trust. In this context, governance must reflect the perspectives and responsibilities of all actors involved.
As digital tools become central to diagnosis, treatment, planning, and surveillance, critical questions emerge: Who controls the data? How are decisions made? What values guide the system? No single institution can answer these alone. Governments have legitimacy but often lack technical agility. Tech companies bring innovation but are not democratically accountable. Hospitals understand patient needs but don’t set global standards. And patients, while most affected, have historically had little voice. Multistakeholder governance is therefore essential—not optional.
The European Health Data Space (EHDS) exemplifies this approach. It seeks to enable both primary (care) and secondary (research, innovation, policy) uses of health data through a federated, trust-based model. Patients retain data control, regulators ensure safeguards, technical experts ensure interoperability, and governance is shared across the ecosystem. This approach is being operationalised through TEHDAS2 (Towards the European Health Data Space), a Joint Action involving 25+ countries. Through structured workstreams and consultations, TEHDAS2 develops practical governance mechanisms. Its core message: legitimacy is not declared—it is earned, by treating each stakeholder as essential.
Three key lessons emerge:
- Shared responsibility builds legitimacy. Inclusive systems earn trust. Exclusion leads to resistance and failure.
- Aligned incentives prevent fragmentation. TEHDAS2 shows how transparent, consensus-driven processes can bridge diverse goals.
- Coordination prevents capture. Neutral forums—like those in internet governance—are essential to ensure AI in health serves collective, not concentrated, interests.
