Originally published September 2025 in the ITS International AI in Transportation publication

In late 2022, everything changed. ChatGPT hit the market and AI became real for the world. In the months that followed, we saw an explosion of innovation: copilots for developers, AI productivity tools, generative design platforms, all arriving at a pace that we haven’t seen in human history. Whole industries have been seismically shifted as they’re forced to adapt. AI wasn’t just theoretical anymore, it was in people’s hands and it was unlocking problem after problem.

But in transportation…the silence was deafening. No new class of tools. No radical rethink of how we design, operate, and improve infrastructure. While other sectors redefined what AI could do, transportation got left behind. Where was our ChatGPT moment?

A Flood of Hype, and a Failure to Deliver

After ChatGPT we saw a deluge of companies, both upstarts and big tech, developing similarly groundbreaking tools like Gemini, Claude, Perplexity, Grok, even Deepseek. Every month we see these companies shattering benchmarks, delivering new breakthroughs, pushing innovation and each other forwards. But in transportation we got something different.

Companies suddenly rushed to slap the letters AI onto pre-existing products, riding the hype-wave even when the technologies used were barely related to AI. Companies that couldn’t even measure the past were claiming to now be able to predict the future.

Then tech giants entered the transportation space, backed by trillion dollar market caps, they jumped in with big wallets and even bigger PR engines. They arrive with limited data, little domain understanding, and software teams trained to build consumer tools, not critical infrastructure. The results were underwhelming: models trained on narrow datasets, AI tools detached from regulatory reality built without even the most basic grasp of traffic engineering.

Then we saw companies chasing novelty over necessity, deploying generative AI for tasks that demanded precision and repeatability, but instead simply allowed users to have conversations with bad data. They built interfaces that sounded futuristic, but didn’t align with how transportation professionals plan projects, follow standards, or secure funding. What looked like innovation quickly became a distraction.

Then legacy hardware players pushed not for AI but on deploying even more hardware than we already have. At a cost that is unprecedented, impractical, and would drag us into deployment timelines that would last decades. Masquerading as innovations of the future, they’ve promoted the same iterations of past failures.

By the looks of it we are far away from our ChatGPT moment. So how can we close the gap?

Opening the Internet for Transportation

ChatGPT and other LLMs benefit from a massive, open, and accessible internet providing petabytes of diverse and complete training data. Transportation is hampered by the opposite problem - 250PB of transportation data is generated each year in the US, but it sits trapped in siloes across multiple organizations, in multiple standards, or held hostage by protocols.

Today most AI tools in transportation rely on a single stream of data, and they expect it to tell the whole story. That’s not how real systems work. Speed data alone won’t explain why delay is happening. Travel time can’t account for signal phasing. Crashes aren’t random—they’re shaped by volumes, behaviors, geometry, and policy.

And then there are legacy vendors—particularly in the hardware space—who’ve built closed, proprietary ecosystems that make integration nearly impossible. These systems claim to champion openness, but often block agencies from accessing their own data. That asymmetry slows progress. AI thrives on interoperability.  

When critical infrastructure data is trapped inside black boxes, even the best AI systems are left flying blind. Data from multiple sources needs to be utilized to weave a complete picture of what is happening, to give AI the visibility it needs to make decisions.

Making Regulation Work with Innovation

As organizations have struggled to keep up with the rapid advances in AI, the ones that have blocked tools entirely have fallen behind those that chose to embrace them, albeit with safeguards.

Innovation in transportation doesn’t stall because of a lack of ideas—it stalls because the rules haven’t kept up. Many of the regulatory frameworks that agencies still operate under were created before modern computing, before spreadsheets, before real-time detection, and certainly before machine learning. These processes, often grounded in studies from the 1970s and 1980s, were built for a world where traffic counts were manual, plans were static, and engineering decisions were made on paper. They weren’t designed to block innovation, but today, that’s what they do.

Even when a clearly better tool exists—more accurate, faster to deploy, easier to scale—agencies often can’t use it. The rules say no. Professionals are forced to decline cutting-edge technologies, instead using old processes that struggle with the demands of the fast-moving modern world. The system protects the status quo, not better outcomes.

Meanwhile, much of academia continues to analyze transportation through simulation and theory. Models are developed with idealized assumptions: clean data, perfect deployments, cooperative jurisdictions, zero political friction. These abstractions rarely survive contact with the field. The real world runs on legacy firmware, patchy detection, and policy that often overrides logic. Theory breaks down the moment it hits the tarmac.

And yet, this is how most innovation is still judged: not by what it delivers in the field, but by how well it aligns with legacy methodologies or academic models. That mindset has to change.

Perfection is the Enemy of Progress

AI is not perfect by any means. We’ve all seen it hallucinate, or come up with bizarre answers to often simple questions. But the progress of these models is unquestionable, while many flaws remain countless more have been eliminated. Organizations that took the risk of early adoption and have stayed patient, have seen major benefits as models have improved. The most successful have been organizations who have spent time determining the limitations of these tools, accepting that they can’t perform everywhere but recognizing and utilizing them in the areas where they can.

Real innovation is messy. It doesn’t happen by waiting until something is perfect. It happens by testing, learning, iterating, and improving. It happens by validating ideas in practice, not just publishing them in journals or pitching them in slides. Engineers need the space to experiment—not recklessly, but responsibly. Not for the sake of novelty, but in pursuit of better outcomes.

The path forward isn’t about abandoning standards. It’s about updating them to reflect what’s possible. The burden shouldn’t be on agencies to defend every deviation from outdated practices. It should be an opportunity to prove what works—and to do it transparently, with data, in the open, and in collaboration with those who know the system best.

Innovation doesn’t need to be perfect to begin. It just needs to be better—and provable. We’ve given the world new tools. Now we need to give professionals permission to use them.

Making AI Work for People, Not Making People Work for AI

But the biggest myth in transportation right now is that AI can do it all on its own. That it can replace engineers, planners, and public decision-makers. A mindset that tries to work around engineers rather than with them, is a recipe for failure.

We’ve seen this story play out before. Adaptive signal control was the first major foray into AI which promised to take retiming off engineers’ plates. But instead we saw engineers having to spend more time monitoring, maintaining, and managing these systems. Engineers shouldn’t be working for AI, AI should be working for engineers.

Engineers aren’t bottlenecks, they’re the ones holding the system together. AI shouldn’t be here to displace them—it should be here to make them faster, sharper, and more effective. It can bring insights to the  surface that would take months to analyze, simulate changes in real time, and move decisions from reactive to proactive. AI should give people superpowers—not take away their agency.

This Is the Moment

We’ve seen what real transformation looks like in other industries. Transportation is overdue. 

The principles we’ve championed over the years - integrated data, aggressive innovation, real-world experimentation, collaboration with engineers - are finally breaking through. Just this year, they’ve delivered breakthrough after breakthrough—from generating insights at scale across North Carolina in a matter of weeks without deploying a single piece of hardware, to launching Florida’s largest real-time integrated system, which has broken new benchmarks for real-time volume accuracy. And our biggest breakthrough innovation is yet to come later this year.

These are the principles that will lead transportation to its ChatGPT moment. It is not a possibility, it is an inevitability. In a world of decreasing budgets, increasing complexity, and a growing despondency amongst the public we serve, the choices are simple. Who wants to build the future, and who wants to stay in the past? Who wants to build things that work, and who wants to cling onto processes that don’t?

We know where we stand. The real question is, who’s coming with us?

‍