The technology is reorienting entire agencies toward collecting and refining data — the foodstuff of AI — creating new positions and systems in all areas of data management and governance.
“AI is just not a software deployment,” Benjamin McCulloch, strategic data scientist at the Texas Department of Transportation (TxDOT), said during a panel April 10 hosted by the Transportation Research Board. “It’s a data maturity program that is going to change how you do your work at your agency or company.”
When thinking of AI, McCulloch said, view it as “you are changing the agency, just like the personal computer and the Internet changed how we do work.” Yes, certain jobs will not be around, he said, offering the example of the office typing pool giving way to a personal computer sitting on every desk.
“But those jobs transformed into other jobs,” said McCulloch. “Certain jobs will change. Undeniable. But new jobs are going to be created.”
Data — which transportation departments have lots of — is at the center of this AI transformation. And its integrity is essential for realizing the full potential of AI systems.
TxDOT has an enterprise data platform built on Snowflake, developed around 2022. It includes data from more than 30 systems, including financial, project development and planning. The data is linked to allow AI tools, like a large language model, to enable a range of searches, McCulloch said.
Maybe an engineer wants to examine a stretch of roadway. That analysis can easily examine data related to crashes, pavement scores, maintenance and more. This is the kind of grunt work an AI agent is good at.
“Using AI with agents to do that work will allow us to unlock potential that we’ve never had previously,” McCulloch said, adding, in the past it would have taken “hours and hours of work to just pull the data together from these systems … just for one location. So instead of taking hours we could do this in seconds.”
In a similar fashion, the Connecticut Department of Transportation (CDOT) developed an internal “DOT bot,” a chatbot which takes DOT documentation, tokenizes it and puts the documentation into a searchable database, Gregory Ciparelli, CDOT chief data officer, said during the panel.
“It’s been successful around the agency. We have hundreds of users every week, unique users that are going and asking very specific questions,” he said, noting a key feature of the technology is that when the chatbot presents its findings, it includes a side panel with the document source material — letting users better understand the source and context of the information.
Elsewhere, the Utah Department of Transportation (UDOT) is involved in a comprehensive “digital modernization journey,” Jennifer Volkening, UDOT director of data analytics and governance, said during the panel.
Its project began in 2024 and is expected to be completed later this year. It will involve transitioning an old legacy technology system toward a more unified platform.
“We’re taking a siloed traditional closed-off system, and changing it, and transitioning it, to a modular system,” Volkening said. “We’re creating a connective tissue that makes UDOT’s new architecture work by moving away from a single closed-off system. And by doing this, we’re using interoperability to ensure that data flows seamlessly.”
The modernization, she said, positions UDOT for more data-driven decision-making.
“With this modular solution, we are taking the data that is needed for that data-driven decision-making, and fostering a more mature data governance framework,” Volkening said.
As part of the process, the state is re-editing its data architecture, giving bronze, silver, gold and platinum categories to data, depending on its refinement. For example, data labeled as bronze is simply raw data. Silver data has been cleaned and conformed.
“This is where you label, you tag, add in the meta data,” Volkening said.
Data labeled as gold is intended for UDOT-wide use and can include public data sets; platinum data, she said, is seen as having “more functionally aligned purposes.”
“As you continue through each layer, the data becomes more and more refined,” Volkening said, indicating the goal is that “reliable data governs every decision.”
