How AI could strengthen citizens’ confidence in government

Tom Temin And you’ve documented some cases in the United States federal government where AI has worked out fairly well.

Valerie Wirtschafter Yeah. So I think that, and recently as a couple of years ago, there have been inventories in the federal government of AI use cases over the past years. That’s more than doubled in terms of adoption. There are some very prominent ones that we see every day, things like at the TSA with facial recognition technology to help speed up getting through security screenings. And so that has been a part of what has improved perceptions, just based on survey responses of the airport security process, which is a place where people interact with government very regularly.

Tom Temin And there’s a couple of cases in your survey here of European applications that were really bad that resulted in real tragedy.

Valerie Wirtschafter Yeah. And I think that’s sort of the flipside of when these things are not, either rolled out responsibly with recognition of where they have limitations, where they don’t have proper human oversight, where it’s sort of bias in bias out. There were a couple involving recovery systems, child care benefits scandals, visa applications, where they caused a lot of damage in terms of being able to roll out public benefits. And I think definitely created a dent in trust, a huge deficit in terms of people feeling like the government was there to serve them.

Tom Temin Right. So then it’s kind of an old story that the deployment of any technology that’s more efficient or faster that the government or an agency tries to deploy to improve service backfires, and then you get this loss of trust.

Valerie Wirtschafter Completely. Exactly.

Tom Temin All right. Well, then before we get to some of the cases that can help further, there’s a large body of survey evidence about how low trust in government is. What’s the general scene looking like here?

Valerie Wirtschafter It’s not great. And we hear about this a lot and it’s across administrations. It’s across advanced economies replicated in democracies, just sort of this perception that institutions are not responsive, that they’re not transparent, that they’re not delivering. And I think some of that has really led to sort of sweeping questions around, is this the the governance structure that’s right for me or what what benefits does democracy bring to me anyways? And so, I think there’s potential, especially for technology to be leveraged appropriately and properly to improve that perception.

Tom Temin And the technology industry likes to proceed, as you call it, a move fast and break things manner.

Valerie Wirtschafter They call it that, actually.

Tom Temin Right. I’m saying your reference to them. And so they try things and if they don’t work, they pull them or they get more venture capital in the next round, it works even better. In government, that approach can really be disastrous, even though we hear that term entrepreneurial government, and let’s get things going faster and there’s all these innovation units and grants going on. So maybe discuss the danger of being unlike government and too much like Silicon Valley.

Valerie Wirtschafter Yeah. The sort of move fast and break things approach is this techno optimist perspective that, to unleash human potential all we need is to unburden technology. And that is a perspective that maybe gains traction, especially in a market based economy where people are buying products if a product doesn’t work, get rid of it, move on to the next one. This is a place where especially for advanced technologies, for things that people don’t understand, where regulation, where governance, where oversight care in how these systems run is actually really important to that buy in, because otherwise people won’t adopt it. We’ve seen this, of course, and stemming to the recent tragedy at DCA, with oversight of airlines and air traffic. The idea of flying through the sky is sort of still mind blowing. But there’s a huge regulatory process involved in that, and that helped facilitate some of the adoption there. And so I think it’s kind of this maybe counterproductive or seemingly counterproductive perspective where actually oversight and regulation and transparency about what’s happening does improve adoption. And then you can really reap some of those benefits.

Tom Temin We’re speaking with Valerie Wirtschafter. She’s a fellow in AI and emerging technology at Brookings. From the case histories that you give for the United States and also overseas, that also seems to point to a pattern here, and that is AI is great when it augments human judgment, but terrible when you turn over judgment to an AI powered system. And it’s kind of on autonomy.

Valerie Wirtschafter Yeah. I think one of the things especially, and so what I’ve looked at quite a bit is private sector workforce adoption and where workforce adoptions and what the benefits are that have accrued, especially in kind of experimental settings where somebody gets an AI assistant and somebody doesn’t. And there are huge benefits around productivity, around the quality of outputs, decreasing disparities in performance, job satisfaction, all of these things. And those are definitely portable to the public sector workforce. There’s already conversations in the UK’s National Health Service about what kinds of productivity gains can we see. But it’s not about replacing humans. It’s about taking away the mundane tasks, the tasks that take up time, that don’t involve creative thinking, that don’t involve sort of the human element. And then, of course, where these tools do come into play, that human oversight is still very critical.

Tom Temin And maybe the other danger is if you use it in place of traditional data processing, which is one plus one is always two. In that logic system, you cite a case in Australia some years ago where the AI system was trying to automate debt recovery, and it incorrectly calculated the debts of social welfare recipients. That’s just a mathematical mistake. But in those cases, it had disastrous consequences for those recipients.

Valerie Wirtschafter Yeah, and that’s something that would have been with proper oversight. And there’s been a whole sort of postmortem on that process. But with proper oversight, that’s something that could have been caught fairly straightforwardly.

Tom Temin All right. So then what’s the best way in your studies view, that government should proceed?

Valerie Wirtschafter There’s undoubtedly inefficiencies and undoubtedly ways to better roll out these tools. Especially within the federal government. But I think this probably extends everywhere. It’s a culture change rate to get people in a workforce to think about creative ways to adopt some of these tools. I think in the government context especially, but also elsewhere, talent is really important. And I know that the U.S. government has made some changes to its hiring processes and things like direct hiring authorities, sort of certification efforts to get pools of talent in. But those are new processes. A lot of people in those are still in their one year probationary period. And so now with questions around the federal workforce, right, all that talent that just came in, that has the kind of capabilities here, hopefully that talent stays in place because it’s really important in order to kind of modernize some of these processes. And then, of course, there’s kind of more typical bureaucratic challenges. I know that some federal employees have trouble accessing certain data analysis packages because of perceived security concerns. So figuring out ways to streamline those processes, make them faster, more efficient, less contradictory. All of those types of things I think are really important, and could help facilitate this process. But of course, the human is still really important in it.

Tom Temin Right. So the more humanlike these large language models become, the more you need real humans riding herd on it.

Valerie Wirtschafter Yeah. And it’s not replacing the human, it’s augmenting them in some sense.