The way artificial intelligence is being incorporated into transportation has an almost anticlimactic quality. No grand opening, no ribbon-cutting ceremony in front of a brand-new, shiny terminal. Rather, it takes place in the background, buried in the traffic signal logic of a mid-sized city in North Carolina, embedded in the routing software of freight trucks traversing the Australian outback, and inside scheduling algorithms at bus depots. If you want to call it that, the revolution is quieter than anyone anticipated.
The situation was exposed in a seminal study published late last year at CoMotion GLOBAL in Riyadh. The report, which was created by the MIT Mobility Initiative and the Kearney Advanced Mobility Institute, mapped real-world AI applications across mobility systems in Europe, the Americas, Asia-Pacific, and the Middle East with input from 55 major organizations, including Google, Lyft, Uber Freight, Deutsche Bahn, and NEOM. The results were about equally exciting and sobering. The use of AI in transportation is still very dispersed. Pilots still make up the majority of deployments. The gap between what AI is actually accomplishing at scale and what it can theoretically accomplish is not getting smaller. It’s getting bigger.
Right now, that paradox is at the heart of the discussion. The greater the potential benefit, such as cleaner air, safer roads, or more equitable access, the more difficult it is to implement. According to John Moavenzadeh, Executive Director of the MIT Mobility Initiative, governments, regulators, operators, and tech companies will need to work together internationally in ways they have never done before in order to fulfill the promise of AI. Perhaps the algorithm was never the bottleneck. It might have been the coordination all along.
Nonetheless, small-scale achievements are building up all over the world in ways that seem genuinely encouraging. GoDurham, a public bus system in Durham, North Carolina, began with a simple traffic signal priority system that used onboard AI to make it easier for buses to navigate crowded hallways. Travel times decreased. The pilot spread throughout the entire fleet. There, Michael Hutchins, a data analyst, compared his agency to a baby learning to crawl in an honest and disarming description of their methodology. In a field full of lofty claims about autonomous everything, that level of candor is uncommon.
Moventis, a Spanish mobility operator, started with driver safety—cameras, sensors, and sophisticated driver-assistance systems intended to lower accidents—rather than passenger-facing flashiness. They added automation over the course of five years, doing away with drivers’ need to manually enter data, before moving on to predictive analytics. Before they even arrive at the stop, passengers in Pamplona can now check an app to see the anticipated bus occupancy. The level of engagement increased by 40%. One gets the impression that the true story isn’t about a single breakthrough as these small victories mount. The idea is to keep making minor adjustments until the system starts acting differently.
Another perspective is provided by Australia. There, the problem is geographical: long distances, few people, and a thin infrastructure spanning a huge area. While drone delivery makes headlines, real-time supply chain modifications, demand forecasting, and AI-powered route optimization offer more useful benefits. For example, FedEx protects temperature-sensitive cargo traveling from Japan to Australian ports by using AI to evaluate network data and dynamically reroute shipments. The majority of customers cannot see the work. In a way, that’s the point.
One noteworthy term from the MIT-Kearney report is the “jagged frontier.” AI doesn’t always fail. In certain tasks, such as demand prediction and pattern recognition across large datasets, it performs remarkably well, but in other tasks, it falters erratically. That unevenness is not a small annoyance in transportation systems that are safety-critical. It’s a basic design problem. Sometimes using AI in conjunction with human operators makes things safer. In others, reliability is actually decreased by human intervention. No one has been able to fully solve the strategic imperative of determining which scenario applies and when.
The tension that permeates everything is difficult to ignore. There is a great deal of ambition. Though piecemeal, the progress is genuine. Governments, businesses, and cities are all developing their own AI-powered mobility systems, frequently lacking compatible data infrastructure or common standards. A future in which regional systems are unable to communicate with one another—a patchwork of clever but separate solutions—is the danger, according to several researchers. It’s still genuinely unclear if the partnership will keep up with the technology. However, a bus simply made its light somewhere in Durham, and no one on board gave it a second thought.
