US eyes AI to fix its tired roads

America’s roads are tired. Ageing asphalt, corroded guardrails, faded paint and potholes that feel practically geological in origin have become part of the national driving experience. For decades, public works departments have struggled with limited staff, inconsistent funding cycles and a repair backlog that grows faster than crews can fill it. Now, artificial intelligence is quietly emerging as the most promising tool in a generation to reverse that decline. Not because AI is glamorous, but because the technology is proving remarkably good at something the country has long found difficult: seeing what’s wrong, everywhere, all the time, in enough detail to act intelligently.

From Hawaii to Texas, from San Jose to Washington, states are starting to treat AI not as a futuristic experiment but as the new set of eyes on their roads — and increasingly, as the system that decides what gets fixed first. The early results tell a story policymakers should pay close attention to: AI is helping identify dangers faster, cheaper and more comprehensively than any human-led system ever could. This shift doesn’t solve America’s longstanding infrastructure funding challenges. But it does something equally important: it ensures every maintenance dollar goes further. And at a time when federal and state budgets are being stretched across multiple priorities, efficiency may be the most valuable currency of all.

Hawaii offers perhaps the clearest example of why AI-enabled road inspection isn’t simply a technological novelty but a public-safety necessity. The state is giving away 1,000 AI-enabled dashboard cameras — each worth nearly $500 — through a new “Eyes on the Road” campaign. The cameras automate inspections of guardrails, signs and lane markings, providing continuous feedback instead of monthly check-ups that often leave dangerous conditions unnoticed far too long. The backdrop is sobering. Hawaii recorded more traffic deaths in 2025 than in the whole of the previous year. It is impossible to say how many of those tragedies were linked to infrastructure failures, but the risks are well-known. The state recently paid a $3.9 million settlement to a family whose relative died after crashing into a guardrail that had remained damaged for 18 months.

AI offers a clear break from that pattern. By feeding thousands of daily images into mapping and diagnostic systems, Hawaii’s transport officials now receive targeted alerts on deteriorating guardrails — every single day, across the entire network. It is precisely the kind of routine, reliable attention that transport agencies have long wanted but could never realistically deploy with human staff alone. The system also reflects Hawaii’s unique logistical constraints. Shipping equipment to islands is costly; terrain limits physical inspections; manpower is finite. For remote or geographically challenging states, AI isn’t simply convenient. It is the only scalable option.

On the mainland, San Jose provides an equally compelling model, albeit for a different reason: the city has embraced AI not as a stand-alone tool but as part of an emerging public-sector data ecosystem. The city began modestly by mounting cameras on street sweepers. The results were stunning: the system correctly identified potholes 97 per cent of the time. That level of accuracy transformed the programme from an experiment into an increasingly mainstream municipal service, now expanding to parking enforcement vehicles. But San Jose’s mayor, Matt Mahan, believes the real breakthroughs will come from sharing data — not hoarding it. The city helped create the GovAI Coalition, a group of US governments working to pool imagery and best practices. The idea is simple but powerful: if an AI system in Minnesota sees a particular kind of road debris on Monday, the same system in California should recognise it instantly on Tuesday.

That is how AI improves: through volume. A single city might only encounter a specific road hazard once every five years. A shared national database might encounter it thousands of times a month. The lesson for policymakers is that AI infrastructure requires more than hardware; it requires collaborative protocols. Cities that embrace shared learning will see faster improvements than those that try to reinvent the wheel.

Cameras are not the only pathway to smarter road maintenance. In Massachusetts, Cambridge Mobile Telematics has demonstrated that the ordinary smartphone itself can act as a sensor network. Its StreetVision system aggregates data on driving behaviour — hard braking, sudden swerves, inconsistent speeds — and uses that pattern to infer where infrastructure may be contributing to unsafe conditions. One example illustrates the elegance of this approach: at a conference in Washington, D.C., company staff noticed a high concentration of aggressive braking at a nearby junction. The culprit was not poor driving but an overgrown bush obscuring a stop sign. The fix required nothing more than a pair of garden shears. This is the beauty of AI-enabled diagnostics: not all solutions require multimillion-dollar engineering projects. Texas, which has one of the largest and most complex road networks in the US, has adopted this behavioural-data approach at scale. The state recently scanned more than 400,000 kilometres of roadway and flagged countless ageing or outdated road signs — the kinds of items that often slip through the cracks when maintenance records are held on paper or buried in archived work orders. This is where AI becomes a true force multiplier. It doesn’t replace workers; it ensures they are deployed precisely where needed.

The most forward-looking insight from these early pilots is that AI’s real value is long-term. Today’s systems detect potholes and faded lines. Tomorrow’s will help build an infrastructure suitable for a mixed fleet of human-driven and autonomous vehicles. Experts anticipate that within a decade, nearly every new vehicle — driverless or not — will come equipped with a camera. The result will be billions of data points flowing daily into traffic management systems. Departments of transportation will no longer guess where problems are emerging; they will see them, in real time, from the perspective of every vehicle on the road.

This raises policy questions the US must start addressing now: how states should coordinate data governance across thousands of local agencies; what privacy protections should apply to citizen-contributed imagery or vehicle telemetry; how AI results can be translated quickly into actionable work orders, funded consistently enough to prevent further backlog; and what standards must be created for AI-verified road conditions in the age of autonomous vehicles. These issues are not incidental. They will shape how safely America transitions to an automated future.

The emerging consensus across states is not that AI will replace traditional road maintenance; rather, that without AI, traditional maintenance simply cannot keep up. America’s infrastructure challenges are too vast, its networks too sprawling, its repair backlogs too extensive. What AI offers is a new social contract: in exchange for the steady flow of data from vehicles and sensors, governments promise faster repairs, safer roads and more transparent prioritisation. And unlike many other uses of AI, the road-repair applications are relatively uncontroversial. They do not affect employment directly. They do not threaten creative professions.

With inputs from AP