Optimizing Your Data With the Power of AI

Edward Dowgiallo, Architect at the Transportation Department’s Office of the Chief Information Officer and Federal Transit Administration’s (FTA) Office of Information Technology, is helping deploy artificial intelligence (AI) for pattern recognition at the agency. At a recent GovLoop Virtual Summit, we spoke to him about how AI is helping with datasets on everything from public transit to road safety.

This interview has been lightly edited for clarity and length.

GOVLOOP: Can you explain how data optimization is laying the groundwork for AI at your agency?

DOWGIALLO: AI is not yet at a point where it can replace people and the jobs they do. It’s more specific to tasks that people can do. It can help automate those tasks, but when I look at AI, what it helps do is identify patterns in my data that I can use in a repeatable manner. Given that AI’s about pattern recognition in your data, if you have optimized data, it would naturally make sense that you would be able to do some of those activities more efficiently. And so, in my agency, the thing we’re very invested in is really building a data capability so that we can eventually build machine learning algorithms.

How is AI helping with the day-to-day work being done at FTA?

We started with AI around something we call the National Transit Database, which collects transit statistics for the whole country on public ridership. We were really concerned about the quality of the data, so we started paying people to do data analysis to make sure that information was accurate, and then we started building a rules-based system, to make sure that that data was clean. As you know now, with AI or machine learning, what you can do instead is you can make the data more accurate.

Another area AI is helping us with our data is safety. We want to start benchmarking our data and understand situations where, for example, if we drop maintenance for a transit agency by $1 million, do accidents increase.

Finally, we want to use AI to understand our data well enough eventually to be able to do predictions on transportation itself, and be able to say, as an example, “If I put a bus route on 14th Street in D.C., that will provide X percentage increase of accessibility to jobs for the people that live there.” So those are the types of use cases we’re doing or looking at potentially implementing with AI.

There’s a lot that goes into determining who is involved with agencies’ AI efforts. How do you decide who is part of those conversations?

Having the business lines involved is very important. Since I work in the IT shop, with the data, you need somebody that’s a topic expert as well. At my agency, I’m happy to see that we’re starting to have data scientists incorporated into our business lines. And we need more of those to exist throughout because there needs to be an expert or an owner of that data. For example, when I’m dealing with the National Transit Database, it helps that I have somebody from that business line that can look at the patterns that we’re recognizing in the data and they can say whether or not they think there truly is causality there, or if it’s really false positives or inconclusive things.

AI is something that augments the human, but the human still needs to make the decision. And the human needs to give this an eye test. So, we need some sort of expert that can take the contexts or the indicators that the machine learning algorithms are giving us, and interpret it – determine whether they think it’s accurate, or point to other things that we should be looking at to explain why we are seeing correlations or causality.

This article is an excerpt from GovLoop’s recent report, “How Data Drives Innovation in Government.” Download the full report here.