In August, a Pennsylvania state lawmaker began drafting a bill that would require a licensed human driver aboard any automated transit vehicle weighing more than 10,000 pounds, even though the transit agency the legislation is aimed at says the technology is nowhere close to arriving in the city. The proposal would not ban automated buses outright. It would simply keep a human in the seat until trust catches up with the technology. 

That single bill says more about the current moment in transportation than any product launch could. The public conversation about artificial intelligence (AI) and traffic has become almost entirely a conversation about robots that drive. It is not the conversation about the AI that is already driving results. 

A crowded, noisy narrative 

The unease is not irrational. In Gothenburg, Sweden, a new autonomous bus service was rear ended by a tram during its second passenger trip, an incident that made international headlines despite causing no serious injuries. In Austin, Texas, one automaker has disclosed more than a dozen crashes involving its robotaxi fleet since the service launched in 2025 and another operator recalled over a thousand vehicles after software caused collisions with gates and chains. Each incident is minor on its own. Together, they are shaping a public narrative that AI behind the wheel, whether in a car or a bus, is reckless or unready. 

That narrative is a reasonable response to a genuinely hard problem: teaching software to make split second physical driving decisions with no fallback. But it is a narrative about one specific application, not about AI and transportation as a whole. And while that debate plays out, congestion keeps getting worse almost everywhere. Chicago now leads the nation in time lost to traffic, with drivers there losing 112 hours a year, and nearly nine in 10 U.S. cities saw delays increase over the previous year. Traffic is not waiting for the robotaxi question to be settled. 

Two different jobs, one confusing label 

It helps to separate what is actually being asked of AI in these two cases. An autonomous vehicle is AI making continuous, real time physical decisions, replacing a human driver entirely. It is one of the hardest open problems in engineering, and it is still, visibly, a work in progress. AI-powered adaptive traffic signal control, by contrast, uses AI for doing something far less dramatic: adjusting how long a light stays green based on real conditions, and increasingly, predicting what traffic will look like several intersections ahead. 

One is trying to replace a driver. The other is trying to replace a stopwatch. Not every signal timing system delivers as promised, and poorly integrated sensor data can leave even a smart intersection performing like a dumb one, but the specific, narrow, already proven uses of AI in traffic management should not be judged by the struggles of a much harder and unrelated problem. 

The quiet version that already works 

This is where the less exciting story deserves more attention. AI-powered adaptive signals have cut travel times in many regions and reduced emissions by cutting the idling that comes with unnecessary stops. Federal research simulating an intersection in Charlotte, N.C., found that AI-based transit signal priority reduced average delay per person by nearly 19% during peak hours—a measured outcome rather than a projection. 

None of this requires a vehicle that drives itself. It requires software that gets a little better at anticipating what is about to happen at an intersection and nudging a signal a few seconds ahead of time. It is the traffic equivalent of a thermostat that learns a household's schedule, not a robot that walks the dog. And it does not require new fleets, new safety regulations, or years of public trust building. Cities already own the signals. Most of the infrastructure consideration is already won. 

What is actually at stake 

None of this is an argument against caution on autonomous vehicles. The safety questions raised by recent incidents are legitimate and deserve serious scrutiny. But public trust in "AI and traffic" is currently being shaped by the hardest, highest risk version of the technology—at precisely the moment cities need to make calm, evidence-based decisions about the easiest, lowest risk version. Congestion and emissions are not pausing while that debate continues. Every year, a city delays adopting smarter signal software is a year of avoidable delay and fuel burned that a decades old traffic light, running better software, could have prevented. 

The future of traffic that actually arrives on time will not look like science fiction. It will not get rear ended by a tram on its second day in service. It will be a traffic light that already knew a driver was coming and turned green when it needed to. No one will post a video of that. It will just work.