AI governance must account for sustainability, not just safety

The global conversation around Responsible AI has largely revolved around fairness, transparency, accountability, and safety. These are necessary discussions. But as artificial intelligence becomes increasingly embedded across healthcare systems, energy infrastructure, logistics networks, and public administration, a broader question is beginning to emerge – one that still receives comparatively little attention:

How sustainable is societal dependence on intelligent systems themselves?

This is no longer simply a conversation about software. AI is gradually evolving into infrastructure.

Across sectors, intelligent systems are increasingly functioning as operational layers sitting beneath critical systems that societies rely upon daily. Hospital diagnostics, electricity demand forecasting, insurance processing, digital health records, transport coordination, supply-chain management, and public-service delivery are all becoming progressively intertwined with algorithmic systems and computational infrastructure.

The efficiencies are undeniable. So are the opportunities. But infrastructure changes the nature of dependence. And dependence changes the governance question.

Beyond the megacity: AI, governance, and the rise of intelligent peri-urban regions

When efficiency becomes dependency

Modern societies rarely notice infrastructure when it functions smoothly. Roads, electricity grids, cloud networks, and hospital information systems become visible primarily when they fail.

AI may increasingly follow the same trajectory.

The current phase of AI adoption is still often discussed through the language of innovation and deployment. Yet the deeper structural shift underway is that societies are beginning to reorganise essential services around the assumption of uninterrupted computational capability.

This distinction matters.

A recommendation engine failing inside an entertainment platform is inconvenient. A disruption affecting AI-enabled healthcare systems, energy management infrastructure, or emergency response coordination is fundamentally different because the consequences extend beyond technology into institutional continuity itself.

Healthcare perhaps illustrates this most clearly.

Healthcare was never built for this level of interdependence

Across Asia, healthcare systems already operate under immense structural pressure. If one includes Northeast Asia, Central Asia, and West Asia, the region contains well over 150,000 hospitals, alongside hundreds of thousands of primary care centres, health posts, community clinics, and GP networks. Yet despite this scale, the underlying challenges remain remarkably similar across large parts of South and Southeast Asia.

Primary healthcare systems frequently remain under-resourced. Preventive care is uneven. Non-communicable diseases continue rising rapidly while many systems remain historically oriented toward acute and infectious disease management. Urban-rural inequities persist. Workforce shortages remain chronic. Public spending often struggles to keep pace with demographic and epidemiological transition.

The result is that tertiary hospitals across many countries increasingly absorb pressures that primary and preventive systems were originally intended to manage.

It is within this already fragile environment that AI is now being introduced as a systems-level solution.

AI-enabled diagnostics. Algorithmic triage. Predictive imaging systems. Remote monitoring. Automated decision support. Digital public health infrastructure.

Many of these systems may significantly improve care delivery. In some settings, they already are.

But healthcare also remains uniquely sensitive to disruption because failures are not abstract. They affect patients directly.

A cloud outage affecting a retail platform may delay transactions. A systems disruption affecting emergency care, imaging workflows, or medication access introduces an entirely different category of societal vulnerability.

This is where the governance conversation begins changing shape.

AI systems do not operate independently of context

There is growing consensus that healthcare AI systems must be trained and validated on populations reflecting their intended environments of use. On the surface, this appears like a technical requirement. In reality, it reflects something much deeper.

AI systems do not function independently of the environments they enter.

Their reliability depends on infrastructure stability, workforce capability, interoperability, institutional readiness, energy continuity, and governance quality as much as model performance itself.

A diagnostic model functioning effectively inside a highly digitised urban hospital may behave very differently in fragmented healthcare environments with incomplete electronic records, unstable connectivity, workforce shortages, or inconsistent diagnostics infrastructure.

This becomes particularly important across the Global South, where AI is increasingly being asked not merely to optimise systems, but to compensate for institutional scarcity itself.

That distinction fundamentally changes the deployment challenge.

In many developing regions, AI systems are entering healthcare environments where continuity of care remains uneven, digital interoperability remains weak, health data ecosystems are fragmented, and institutional capacity itself remains under strain.

Under such conditions, excessive dependence without sufficient resilience may gradually create new forms of systemic fragility.

The invisible infrastructure beneath AI

Much of today’s AI discourse still treats AI systems as though they exist primarily inside applications and interfaces.

In reality, modern AI ecosystems depend upon extraordinarily complex physical and geopolitical infrastructures: hyperscale data centres, stable electricity networks, semiconductor supply chains, rare earth minerals, cloud providers, undersea cables, cooling systems, satellite infrastructure, and increasingly concentrated computational ecosystems.

This dependency chain matters more than many governance discussions currently acknowledge.

Recent semiconductor shortages during the COVID-19 pandemic disrupted everything from automobiles to medical devices.

Large-scale cloud outages have temporarily affected hospitals, financial systems, and logistics operations simultaneously across multiple countries. Rising energy demands associated with hyperscale AI infrastructure are already triggering debates around electricity consumption, water use, and environmental sustainability.

As AI systems become more deeply embedded across critical sectors, disruptions in one layer increasingly risk cascading across others.

Energy instability affects compute infrastructure. Compute instability affects digital health systems. Supply-chain disruption affects medical technology availability. Cyber incidents affect public-service continuity.

Over time, these systems become interdependent in ways societies may not yet fully appreciate institutionally.

Responsible AI cannot be separated from sustainable AI

None of this suggests societies should retreat from AI adoption. That would neither be realistic nor desirable. The opportunities AI presents for healthcare, public systems, and scientific advancement are substantial. But Responsible AI discussions may increasingly need to evolve beyond questions of fairness and algorithmic safety alone.

Because a healthcare AI system cannot be considered fully responsible if the surrounding infrastructure required to sustain it remains institutionally fragile, environmentally unsustainable, or geopolitically vulnerable.

Similarly, public systems optimised entirely around automation without sufficient redundancy, local capability, or human oversight may improve short-term efficiency while quietly increasing long-term systemic dependence.

Historically, resilient societies built safeguards into critical systems precisely because disruptions were inevitable. Healthcare systems maintained manual pathways. Infrastructure systems built redundancy. Institutions preserved human capability even during technological transition.

The challenge ahead is ensuring that AI integration does not gradually erode resilience faster than societies develop new forms of institutional preparedness around it.

Beyond adoption

The defining challenge ahead may therefore not simply be whether AI systems become more intelligent. It may be whether societies retain the institutional, environmental, and governance capacity required to depend on them responsibly over time.

Because eventually, the conversation around Responsible AI may need to confront a more difficult question altogether:

Not whether societies can deploy intelligent systems at scale but whether societies are building systems resilient enough to live with that dependence for decades to come.