Making sure AI is used responsibly at NSF

The work on artificial intelligence conducted at the National Science Foundation is broadly twofold: investing in research and development, and responsibly deploying new and evolving technologies internally. For NSF’s Dorothy Aronson, the division director and chief data officer serving as the assistant chief information officer for AI, her role is focused on the latter.

Although she technically holds three titles within her agency, Aronson discussed how she brings her IT past into her agency’s development of AI solutions and guidance and readying of NSF’s workforce to have a baseline level of comfort and knowledge surrounding AI tools in agency operations. 

Nextgov/FCW: Who do you report to in your organization? 

Dorothy Aronson: That's a very complicated answer, I think in some ways because I have multiple titles all within the Office of the Chief Information Officer. So I report to the chief information officer. As the chief AI officer, what I'm doing is establishing governance infrastructure for AI, and at NSF we have really three components of governance: internal facing, the external facing –– which is the research –– and then there's a combined governance board that oversees both of those organizations, and I'm on all three of those boards. Because the law requires that the chief AI officer reports to the deputy head of the agency, so the third board provides that kind of interface.

Nextgov/FCW: How many people are on your team? 

Aronson: Right now I have five billets. Four of them are filled. So there will be five people reporting directly to me, and right now there are only four. And that's just because I'm in the middle of hiring the fifth person. It’s an exciting team. Will the organization grow? Within the group that I have, I can envision at least three of them having subordinate teams under them. Today that's not the case. So we're set up to scale in that way; we're set up to scale into an organization that can grow if and when the demand for AI solutions grows.

Nextgov/FCW: What is your role in things like AI acquisition and workforce development?

Aronson: Acquisition really is through our contract shop, I would say, and I'm working with them, of course. With respect to the workforce, there's a lot of stuff going on. I don't think that my small staff is going to be the only people in the agency creating solutions. In fact, I would say we do have a new AI community of practice going so people throughout the agency can come together and talk and share their ideas and concerns. We're looking at creating a small test bed so that people can try things. 

There's a lot more education needed for everyone. I think everyone has to understand the capabilities and appropriate use of AI tools. And even though we have some policy, we really have to share that policy more broadly. We have to communicate broadly, and so we need to to keep that going. We also have a very large group called the Enterprise Data, Governance and Education Community of Practice. So the AI Community Practice is new and forming, and it's a subset.

Nextgov/FCW: How do you see the role of chief AI officer evolving over time?

Aronson: My role is not about science. It's an internal-facing responsibility for the responsible use of AI within the National Science Foundation, as opposed to the investment piece. I watched, over all these years, this continuous transformation of technology through various stages of its maturity, and now what I'm seeing is the same kind of pattern of activities happening with data and AI, but in a much smaller timeframe. So it's really thrilling to see, to be part of this explosion and rapid innovation that's going on, and a rapid adoption of new technical solutions. 

It's really important for the AI people and the IT people and the data people to be in coordination. Because as I said: I don't think we're going to be developing the AI solutions. I think that the central IT shop is probably going to be developing the major solutions, but we'll be maybe piloting them. We'll be helping them with governance and with tools, etc. But I think the IT workforce is a very critical part of making this work.

Nextgov/FCW: What is your biggest priority right now?

Aronson: On the administrative side, we're piloting certain capabilities in order to determine the long term cost of maintaining AI capabilities within our business systems. So we're piloting things because the real cost of AI is not in the creation of the AI component itself, it's in the long maintenance cycle, the machine learning and the long term maintenance of the product. So we're evaluating solutions that would allow us to implement tools today and estimate the long term and centrally manage those tools. 

For example, it's not hard to develop a capability, but somebody has to maintain that thing over time. And so what I believe is that the tools that are coming out, end users will be able to create their own solutions. They can do it now. The problem is: if lots of people throughout NSF are creating AI solutions, I need to be able to understand where those things are, what they do, and then ensure that they are managed well over their lifetime. So I'm really focusing on how do we create the policies, you know, like the rules of behavior, the rules of the road, to ensure responsible use, not only in the beginning phases of the AI, but over its life cycle. And so I'm hoping that we're able to implement, through my shop, a machine learning and ML operations infrastructure that will allow us to monitor everything, regardless of who's developing it. So that we can liberate the customers who have the use cases and allow them to do their own modeling, developing tools on other products and on other existing AI capabilities. I'm responsible for managing them or ensuring that they're managed well, so that's really the focus for me right now. It's that whole infrastructure to ensure responsible use.