AI Won’t Make Cities Resilient on Its Own; Institutions Have to Deliver

Artificial intelligence (AI) can alert infrastructure operators to emerging faults, help them assess risks and support faster decisions during disruption, but a city gains little from those capabilities if its agencies cannot share information, act on a warning or sustain the system beyond a pilot.

In the study "AI Adoption Enablers and Institutional Governance for Sustainable Infrastructure Resilience in Smart Cities," published in Sustainability, Wael Alattyih and Mohamed T. Elnabwy examine what makes AI appear valuable for resilient urban services. Their evidence comes from Saudi Arabia, but the question extends to any city investing in digital infrastructure.

The researchers find that stakeholders' belief in AI's practical benefits is the strongest predictor of their resilience assessments. Institutional capacity and integration, followed by trust and ethics, are the leading governance factors. These are findings about perceived resilience, not proof that AI has prevented outages or reduced losses. Even with that boundary, the study offers city leaders a useful challenge: explain how a technology will improve an operational decision, then build the institutional ability to use it.

The hardest work begins after deployment

Cities are turning to digital systems as heat, flooding and other pressures test essential services. AI could help a utility identify equipment at risk of failure or help an operator interpret changing conditions across a network. Whether either application improves resilience depends on what follows the analysis. Someone must judge the information, have authority to respond and coordinate action across the organisations involved.

The chain is easy to overlook when smart city progress is measured by projects launched or systems installed. A technically capable model can sit outside routine work if its outputs arrive too late, conflict with existing procedures or lack a clear owner. Agencies may also struggle to maintain a system after its initial funding ends. The distance between a successful demonstration and a dependable public service is therefore an organisational question as much as a technical one.

The authors examined that question through an online survey of 170 stakeholders involved in Saudi smart city projects. Participants included engineers, managers, officials, consultants, technology providers and researchers. They rated conditions associated with AI adoption, aspects of institutional governance and infrastructure resilience across environmental, economic and social dimensions. The survey ran for four weeks in May and June 2026, providing a snapshot of professional assessments at a time of substantial smart city investment.

The authors compared three machine learning models and used additional analyses to examine which factors drove predictions and whether their relationships were consistent across respondents. The best-performing model explained a modest share of variation in perceived resilience. Its value lies less in predicting an individual city's future than in identifying conditions that warrant closer attention. This is important for anyone tempted to treat a model ranking as a ready-made spending plan.

Practical value leads, but governance keeps services working

Perceived value and performance benefits was the strongest predictor in the study. It contributed roughly twice as much to the model's predictions as the next-ranked factor. Stakeholders who saw clearer practical benefits from AI tended to rate infrastructure resilience more highly. The result suggests that a credible use for the technology carries more weight in these assessments than a general commitment to digital transformation.

For policymakers, the finding changes the first question asked of an AI proposal. A city might examine whether a tool helps staff identify maintenance needs earlier, assess a developing risk more accurately or coordinate a response more effectively. It would then need to specify what evidence could establish that improvement. Without an operational goal and a way to check results, claims about resilience remain difficult to distinguish from confidence in the technology.

Institutional capacity and integration ranked next, while institutional trust and ethics was the other leading governance contributor. These factors bring the analysis back to the people and organisations responsible for public services. A warning may be accurate but ineffective if staff cannot interpret it or agencies cannot combine the information needed to respond. Trust affects whether an AI-supported recommendation is examined, challenged when necessary and ultimately used.

The researchers also looked at whether strong governance consistently magnified the apparent benefits of AI. They found modest interactions rather than a broad amplification effect. One exploratory pattern suggested that regulatory readiness was associated with a higher baseline of perceived resilience when stakeholders saw less value in AI; the difference narrowed as perceived value increased. The authors describe a possible compensating relationship, while cautioning that their data cannot establish it as a general rule.

This nuance has practical consequences. Governance should not be treated as paperwork added after an AI system proves itself, but neither can it make an unhelpful application valuable. City governments need to consider both sides of an investment: the decision a tool might improve and the arrangements that allow people to use it responsibly. Clear roles, staff capability, information sharing and oversight become especially consequential when several agencies depend on the same infrastructure.

Different institutions face different obstacles

The analysis suggests that progress may be uneven. Four of the five leading predictors showed nonlinear patterns, with predicted resilience rising more sharply once respondents' ratings reached approximately 3.3 to 4.0 on the study's five-point scales. These figures are exploratory features of the survey models, not universal readiness thresholds. Still, they raise the possibility that incremental investment has limited effect when an organisation lacks several basic conditions needed to put AI into routine use.

The researchers also identified four stakeholder profiles with different average resilience ratings. A small group of 11 rated both adoption conditions and governance highly and reported the strongest perceived resilience. Two much larger groups reported almost identical resilience ratings despite taking different routes to that position: one saw greater value in AI but rated aspects of governance more weakly, while the other reported a more balanced governance profile. A fourth group rated most conditions lower and reported the weakest resilience.

These differences suggest why a uniform smart city programme may disappoint. An organisation that has yet to identify a useful application may benefit more from testing a narrowly defined service problem than from acquiring a wider platform. Another may understand what AI could do but need stronger coordination, staff skills or safeguards before it can use the system confidently. The study did not test these responses, so they should guide programme design and evaluation rather than be presented as proven remedies.

The issue is particularly relevant where cities must balance digital investment against immediate infrastructure needs. Governments and development agencies may face pressure to fund visible technology while less visible work, such as maintaining reliable data and training staff, competes for the same resources. Businesses supplying systems also have a role: they can demonstrate how a product fits an agency's decisions, rather than relying on broad claims of efficiency. Civil society can ask whether improved services reach the people most exposed to disruption.

A global question that needs stronger evidence

The study helps broaden the definition of success in urban AI. Resilient infrastructure has environmental, economic and social dimensions: services must cope with disruption, remain viable to operate and serve the people who rely on them. Those concerns connect the research to international goals for sustainable cities and resilient infrastructure. The connection is strategic, however; the survey does not measure progress against global development targets.

Its evidence also has clear limits. Recruitment through professional networks and referrals may have favoured people already involved in, or receptive to, AI initiatives. All the main measures came from one survey at one point in time, leaving room for shared perceptions to influence the relationships observed. The sample was modest for detailed machine learning analysis, performance varied across validation groups, and two measurement scales fell below a conventional reliability benchmark.

Most consequentially, respondents assessed resilience rather than supplying independently verified records of infrastructure performance. The study cannot tell us whether a system reduced service interruptions, improved recovery after a flood or delivered savings large enough to justify its cost. Those questions require follow-up research that tracks projects over time and compares institutional readiness with operational results. Replication across countries with different resources, climate risks and governance systems would show which patterns travel beyond the Saudi case.

For now, the most useful conclusion is a discipline for decision-makers, not a promise about AI. Cities should identify the public service decision they want to improve, establish how improvement will be measured and assess whether their institutions can act on the result. Saudi Arabia provides an example of the challenge at a moment of major investment. The test facing cities everywhere is whether digital capability becomes reliable capacity to protect essential services.