Infusing government with AI, one pilot at a time

When the General Services Administration launched the Artificial Intelligence for Citizen Services pilot in October 2016, its goal was to study systems like Amazon Alexa or Microsoft Cortana to see what these AI-driven tools could offer to government.

Justin Herman, the leader of GSA's Emerging Citizen Technologies Program, has guided federal  pilots to see what AI can offer agencies. GCN spoke with him about what these technologies mean for the future of government.

The interview has been edited for clarity and length.

GCN: What has the past year been like for the AI for Citizen Services pilot?

Justin Herman: The launch of the pilot proved the level of both interest and dedication agencies had across government to pursuing, exploring and implementing AI services for IT modernization. Originally, we were hoping to get maybe six agencies involved with the pilot. Eventually we had to cut it off at 30 agencies.

Our entire role with this program is to develop interagency services that can grow into shared services, shared resources -- really whatever is needed when there is a proven business need from our customers, which are federal agencies. And the success of [AI for Citizen Services] led to the launching of the Emerging Citizen Technology Program itself. We now support and coordinate emerging technology efforts across government in artificial intelligence, robotic process automation and now blockchain, among other things. It continues to grow and increase its impact for agencies.

GCN: What was the goal of the pilot?

Herman: The main goal of the AI personal assistant pilot wasn’t to ensure that every participating agency went to market with a specific Amazon Alexa skill or Google Assistant feature. The purpose was to start identifying a road map forward.

There are some agencies that have developed assistants and they’re available on Github through the Emerging Technology Atlas. One that’s already public is the National Park Service Ranger skill.

Right now is we’ve started the process for negotiating the federal terms of service for Amazon Alexa, which again is one of those steps that we only start taking once there is a demonstrated clear and compelling business need for our customer agencies.

Now that we have the proven use case, we can start looking at terms of service for a whole new generation of AI-supported platforms.

It’s exciting that the maturity cycle of this development has ramped up so quickly. It’s one of the things that people often overlook because they think, "Oh wow, there’s some great tool out there. Why isn’t federal government using it?" Well, it's because we do have higher standards --whether it comes from security, privacy or looking at these terms of service --  and so that’s what we have to do.

GCN: What have you learned about AI platforms over the past year? What are they good at?

Herman: The tools themselves, the individual services, have completely changed and evolved just in the span of one year. So our approach to this has always been completely agnostic. Like a lot of the lessons we’re learning, it starts with not pinning ourselves to any one particular solution, but gaining a greater understanding of the abilities we want -- whether it's 24-hour services, multilingual responses to people, or just how to do more with less.

There is a checklist of the services we want, but we also want greater understanding of the role intelligent personal assistants can play. There are times we've looked back at our checklist and realized, "OK, maybe the technology isn’t where we want it to be for that right now, but we can still look at ways robotic process automation itself can help us streamline."

These technologies have not just unlocked more tools for agencies, but there has also been a halo effect that has empowered federal managers to be able to think in new ways and approach their old problems from new perspectives. It has allowed them to discover that the solution they need isn’t an AI assistant, it could be repurposing technology they already have. As long as they are able to reach their goal and solve their problem, we don’t care how they do it.

GCN: What's the process for how an agency would develop a digital assistant?

Herman: The first thing we want to look at is the problem an agency is trying to solve and the barriers to solving that problem. Then we look at potential solutions. We like to compare three different solutions. One of them could be an automated assistant or a chatbot or something like that. Then we ask and analyze: What’s the achievability? What’s the cost? What impact would it have? Then we'd test and evaluate.

Sometimes during this process, agencies might find another avenue that works. Other times they definitely want to be early adopters in what is an inevitable space. Each time those decisions are made, we gain a better understanding. The insights that went into analysis for one agency give us a better understanding of the problems and how people are approaching them.

One of the things in the field that has absolutely exploded in the last year is robotic process automation. The thinking process that went into what would customer service or outreach look like with an intelligent personal assistant is the same process or approach that agencies have been able to use to look internally and think about what automation do within their own processes. This is all connected.

GCN:  Can you give me an example of robotic process automation?

Herman: I don’t have to go far. GSA's Federal Acquisition Services put out a successful pilot that used a blend of robotic process automation and blockchain to speed up the FASt Lane acquisitions process.

One of the things we see every day is that people think AI is some genie out of a lamp or some fake person you talk to.

The reality is, the practical and tangible ways AI supports IT modernization and shows clear, compelling impact on the mission are ways people might not even know it's being used -- they just know that their day went a hell of a lot better.

The ultimate goal is making the services that we have for citizens faster, more effective and with less wasted time and resources.

Every pilot we support and every initiative we talk about are all woven together. It’s an integrated approach that looks at the role of cloud services, data services, legacy technology and emerging technologies. These are not islands unto themselves, it has to be an integrated approach.

GCN: What are your plans for the AI for Citizen Services initiatives in the coming months?

Herman: There is a lot. Not only do we have to Atlas out to help map things out, but we have created a section where we’re sharing the meeting notes and action items from our interagency meetings, so anyone can see every month the action items on the AI, blockchain or robotic process automation projects.

One of the things we learned early on -- and it’s been a bedrock of this program -- is finding and institutionalizing better ways of working with U.S. business and top research facilities. We cannot do this alone if we want to get it right. That’s why we’ve opened up the meeting notes to demystify where we’re going.

One of the things that is coming out in the next couple weeks is our emerging technologies Pathways to Acquisition resource. It will be a one-stop shop where both federal agencies and U.S. business can see the full scope of ways they can acquire and pilot emerging technologies.

GCN: Looking five years down the road, how do you see virtual assistants being used in the federal government?

Herman: The technology we’re looking at, the AI personal assistants, is new. The problems that agencies face and their need to connect citizens with services more easily, that never changes. If there is one thing I know based on our experiences and working with agencies and businesses, is that the conditions and the reality that we have today are going to be completely different in five years.

Citizens won't be digging through 100 pages of a website to access the services they need. They'll be using voice recognition, smart web forms, chatbots or some solutions that we haven’t thought of yet. I don’t know specifically what will happen in five years. But I know it will be faster, it’ll be better and it won't require sorting through 100 pages of a website.

The point is there are two things that separate the American people from the services they need. There is a bureaucratic divide, services that come from some sub-agency citizens don’t know exists so they have to hunt for it, they have to search for it. The other side of it is the technology itself: wondering what platform can best get people a response.

Right now the burden is on the user to search and find what’s already theirs; in five years, those services will come to them.