If you asked a hundred people about "AI in transportation" in 2020, ninety-five would have said "self-driving cars." Five years later, autonomous vehicles are real but narrower in scope than the hype suggested, and the actual story of AI reshaping how we move is happening somewhere else entirely.
The real story is routing. Scheduling. Matching. The unglamorous mathematics of getting more people and goods to their destinations with less waste.
Let me tell you where it's working.
The obvious: autonomous vehicles
Let's get this out of the way. Self-driving cars exist and operate in a handful of cities — Waymo in Phoenix and San Francisco, Baidu in China, a few others. The technology works, mostly, in constrained geographies, with a lot of careful engineering around the edges.
They are not about to replace human drivers at scale. The remaining 1-2% of driving — the unusual weather, the novel construction zone, the once-in-a-decade edge case — turns out to be extraordinarily hard, and that last mile of reliability is what separates demonstration from deployment.
Meanwhile, highly-automated trucking is arriving faster than passenger vehicles. Long-haul routes with predictable conditions are easier to automate, and the economics are compelling. Several companies are operating commercially on limited routes.
The quiet revolution: routing
This is where the real impact is happening.
Delivery route optimisation. UPS, FedEx, Amazon, and countless smaller logistics companies now use AI routing systems that dynamically adjust delivery sequences based on traffic, weather, customer availability, and driver fatigue. The cumulative effect is enormous — UPS famously saved around 10 million gallons of fuel a year just by eliminating left turns. Modern systems do far better than that.
Ride-hailing matching. Uber, Lyft, and their international competitors match drivers to riders with AI systems that have to balance many factors — distance, driver income, rider wait time, surge pricing, expected demand in the next ten minutes. The quality of this matching is invisible to users, but it's why modern ride-hailing works at all.
Public transport optimisation. Cities are increasingly using AI to redesign bus routes, adjust metro frequencies based on real-time demand, and coordinate connections. Not glamorous. Directly changes people's commutes.
Flight and rail scheduling. Airlines and rail operators use AI for crew scheduling, maintenance planning, and disruption management. When a storm grounds flights, the AI helps decide which cancellations minimise downstream disruption.
Port and freight optimisation. Container ports use AI to decide which ship to unload when, which container goes where, which truck picks up what. Ports that used to handle, say, 800 TEU (containers) an hour now handle 1,200 — with the same physical infrastructure. Pure software.
Where AI is genuinely helping safety
Advanced driver assistance. Modern cars have lane-keeping, automatic emergency braking, adaptive cruise control, blind-spot monitoring. These aren't "self-driving"; they're "really good safety features." And they work. Insurance data shows meaningful reductions in claims for cars with these features.
Fatigue detection in trucks. AI systems watching the driver (eye closure, posture, steering patterns) can alert drivers — or their dispatchers — when fatigue is setting in. This has prevented accidents. It's also, some drivers point out, a bit surveillance-y; the best implementations are designed with driver input.
Predictive maintenance for fleets. The same predictive-maintenance stories from manufacturing apply to trucking and rail. Fewer breakdowns, safer vehicles, longer service life.
Traffic management. AI-driven traffic light systems can adapt to real-time conditions, reducing congestion and — in some cities — emergency response times. Unglamorous. Saves lives.
The public-interest angle
Transportation is one of the most public-interest-loaded areas of AI. A few things I care about:
Accessibility. AI-driven transport can be a huge win for people with disabilities — better paratransit routing, accessible wayfinding, real-time information in accessible formats. The tech industry has historically underinvested here. This is changing, slowly.
Equity. AI routing systems can reinforce historical biases if they're trained on historical demand — routing drivers away from lower-income neighbourhoods, for example. Good practitioners explicitly design against this.
Climate. Transportation is about 20% of global emissions. Anything that reduces vehicle-miles (better routing, better matching, fewer empty legs) has real climate impact. Anything that encourages more driving (e.g. self-driving cars used to avoid parking costs) has the opposite.
Workers. Many of the AI systems we're talking about change the work of drivers, dispatchers, and schedulers. Some of these changes are for the better (fewer painful shifts, better information). Some aren't (more surveillance, lower wages, less autonomy). The difference is mostly in how the deployment is designed.
What I keep watching
- Eclectic autonomy. Not a single autonomous technology, but many of them — self-driving shuttles on university campuses, autonomous container trucks in ports, autonomous tractors in fields. The deployment is quiet, incremental, and largely working.
- Logistics AI at mid-sized companies. The big players already have good systems. The middle of the market is where the next wave of productivity gains is happening.
- Urban mobility. European cities rolling out integrated AI-driven mobility-as-a-service platforms. Not perfect, but the vision is compelling: one app for bikes, buses, trains, scooters, car-share — with AI-optimised routing across modes.
If you're working in transport or logistics and thinking about where AI fits, we'd love to help you think. Operations research is a favourite.
