Artificial intelligence in government has been saved
Limited functionality available
What do government workers need more of? Time. And that’s exactly what artificial intelligence promises to offer. Our research suggests that cognitive technologies can help agencies free up billions of labor hours per year, to be spent doing real work, not drudge work.
Technology is developing so rapidly, and people are concerned about what is this going to mean for my job and what is our society going to look like 10 or 15 years from now—and we don’t really have very good answers for them.
Visit the AI in Government collection
Read the full article
Read the executive summary
View the analysis
Watch the video
TANYA OTT: The future of artificial intelligence: what we know, what we don’t know, and what it all means for you.
I’m Tanya Ott, and this is the Press Room, Deloitte University Press’s podcast on the issues and ideas that matter to your business today.
When you think of artificial intelligence, you probably don’t think of this: [sound of hiking through the woods].
I’m hiking through the woods at a park in Birmingham, Alabama. It’s the South, so it’s hot—but the tree cover is helping a bit. This park is part of Red Mountain, here in the rolling foothills of the Appalachians. The trail is dotted with the remnants of 19th-century iron-ore mines. Not exactly high-tech, but there’s an interesting intersection between technology and wilderness areas like this.
Back in the 1980s, you could walk into any outdoor store and find these big paper green and pink topographical maps. The US Geological Survey, which is part of the Department of Interior, made them. Hikers and hunters and other outdoor enthusiasts used these topo maps to make sure they didn’t get lost in the wilderness.
But by the 1990s, GPS and other similar technologies started getting broader adoption, and all of a sudden those maps, which had been painstakingly drawn by hand, started being produced on computers using a digital process. Suddenly, lots of cartographers were out of work. It was an early digital disruption.
It’s a story that intrigued Peter Viechnicki. He worked in military mapping for a few years and is now a data scientist with Deloitte’s Center for Government Insights. Peter and the center’s executive director, Bill Eggers, do a lot of thinking about how government can become more efficient. Right now, they’re keyed in on artificial intelligence [AI].
Here’s Bill, and then we’ll hear from Peter.
BILL EGGERS: There are very few things that will have as profound an impact on our lives in business and society over the next decade or so than AI. And we wanted to understand the potential impact of it on government organizations.
PETER VIECHNICKI: There’s a lot of anxiety out there among folks in the country who are worried about what artificial intelligence is going to mean for them. Technology is developing rapidly, and people are concerned about what is this going to mean for my job, and what is our society going to look like 10 or 15 years from now—and we don’t really have very good answers for them. So we thought we would just try to look at this in a data-driven way and take our best guess.
TANYA OTT: Before we get to that, though, let’s walk through exactly what AI includes—because it’s a whole lot of things.
BILL EGGERS: AI is actually a broader way of looking at a set of technologies, what we call cognitive technologies, that are each doing specific things. You have things like speech recognition, which can accurately transcribe human speech. It’s gotten better and better over the years. There is machine translation, which translates text or speech; there are some massive advances in that in terms of translating documents, books and so forth, from one language to another. Google made some really big advances in that over the last year.
Then you have other cognitive technologies like computer vision, when Facebook is able to actually look at the picture that you’re seeing and say that’s so-and-so or that’s so-and-so. Or when you have Google Photos, and you download them, and you can type in, “I would like to see all of the photos I’ve taken in front of monuments,” and it’s able to do so. That’s complete computer vision.
And then you’ve got machine learning: when computers learn without being explicitly programmed. They learn by trial and error and by mining information just for patterns and data that could help predict patterns and new data. So when your email program will flag a message spam, or your credit card company flags a potentially fraudulent post, that’s machine learning. Examples in government would include unmanned aerial vehicle robots used for disaster response and robots used to assist in home health care. So those are a few of the forms of AI and cognitive technologies.
TANYA OTT: Why is this so significant to government?
BILL EGGERS: It’s similar to how it’s significant to every company, every business you know. If you look at the digital revolution that’s occurred in our society over the past 10 to 15 years or so, AI is the next stage of that. Some people actually believe—they will make the case to you—that the Internet, in terms of its impact, looks very small in comparison. I was at the Consumer Electronics Show earlier this year in January, and walking around there, they had a whole room full of exhibits and essentially companies that were embedding artificial intelligence into all sorts of toys and products for babies, of all things! You have toothbrushes with AI embedded in them. You have all sorts of fitness devices. You have autonomous vehicles. Basically, it’s huge. It’s becoming more and more an integral part of our lives. So we need to look at how government can take advantage of that to improve services for citizens to reduce cost and increase speed and so forth.
PETER VIECHNICKI: What we’re doing is we’re looking at how much time can be freed up. Government workers spend a tremendous amount of their time doing tasks that take up a lot of time during their days, and we actually see a lot of potential for AI to come along and free up a lot of that time. So those hours that currently are being devoted to really routine tasks—we can give them back to government workers, and this should allow them to focus on the stuff that’s actually important in their jobs.
Just to give you a concrete example: Government workers spend a tremendous amount of time doing paperwork, and they have to because that’s the way their jobs are currently structured. But a lot of the systems that hold that paperwork can be redesigned to help speed up filling in the paperwork. Once that happens, government workers will get hours back during their workweeks. Take an example like a corrections officer who deals with personal issues in a state government agency. That person, if he or she doesn’t have to spend three hours each day filling out forms and documentation, could take those three hours and spend them working with more parolees, and getting a better understanding of the parolees, and doing a better job at the actual kind of human interaction—all the kinds of things that they feel are important at the core of their job. That’s what I’m excited about: that idea of freeing up time and allowing government workers to actually do what they want to do, what they’re passionate about, what they do best.
TANYA OTT: One of the things that really struck me when you’re talking about this paperwork issue is about one of the agencies that struggle the most with paperwork, which is the child welfare department. In most states, case workers are carrying too many cases, and the paperwork can be overwhelming. You write about the Colorado Department of Human Services, where they studied 1,300 child welfare workers and recorded how much time they spend on different activities for a month. And you found that caseworkers spent 38 percent of their time on documentation and administration, and only, like, 5 percent on parent or family contact and just 4 percent on child contact. That would seem to me to be a really great example of perhaps automation allowing much more time to interact with the actual people affected in these cases.
PETER VIECHNICKI: Yeah, that’s a perfect example, and that’s our hope. Those kinds of situations are found across government right now. Child welfare is a great example, but you can find that in every state agency and every federal agency where people just aren’t able to handle all the cases that are assigned to them to the best of their ability because of all these things that are dragging down their time. We hope that those people will get that time back, and then they’ll be able to interface with those families more effectively and have an impact.
TANYA OTT: What’s the actual AI technology that would help them go through that paperwork?
PETER VIECHNICKI: There’s a set of technologies that are called robotic process automation, and they can do things like taking data from one database or multiple databases, and putting it into a form and either stored in a spreadsheet or held in a documentation system. So the child welfare worker or the parole officer or grants management office (or whatever the government employee is) spends a lot of the time getting information from one place and looking at it, and then storing it in another place. That can all be automated by these robotic process automation solutions. We’re just starting to see government agencies taking advantage of this, and that’s exciting.
TANYA OTT: So that would eliminate the need to be going through multiple different databases trying to find pieces of information, coalesce it all together, and manually automate that entire process.
PETER VIECHNICKI: They could go get right to the endpoint—which is like, let’s look at all this information together, and then let’s make a decision and act on it, because humans are the ones that are good at judgment. We’re good at assessing information and deciding what to do. Artificial intelligence can’t really do that at this point. We know it can do it in certain narrow contexts. But it’s that gathering together all the pieces that you have to do before you can make that decision—that part can really screw it up.
TANYA OTT: What you’re saying, I think, is that AI technology is going to streamline that process for me as a consumer?
BILL EGGERS: Yes, basically some processes will be automated in general. Basically, a lot of the kind of paperwork and those sort of processes that typically have gone through different humans handing it from one to another will essentially be automated—the use of digital signatures and so forth. But there’s all sorts of manual, tedious tasks that government employees have to actually do all the time, and they don’t enjoy doing it, that bots are now able to do for you.
We’ve actually estimated that, potentially, over the next five to seven years, with a big investment into this, you could free up in the federal government 1.2 billion person-hours—potentially up to 40 billion in savings. Think about if you had an extra 25 to 30 percent of your day that you could focus on a more mission-critical path, because a lot of the tedious manual things will be taken away from you. At the same time, cognitive technologies, in terms of machine learning and others, are going to help individuals get connected to vast databases and give them better predictive capabilities. So you’re going to be freeing up some of the time to focus on the mission, and these technologies will help to dramatically increase and extend human abilities to actually deliver on that mission.
TANYA OTT: It’s not just about reducing and streamlining work—it’s also about being able to make new discoveries or new insights.
PETER VIECHNICKI: Exactly. We have been talking to Army Research Labs about that, and we’re talking with a group that designs military radios. One of the things they’ve done recently is automated the testing of the circuits that go into those radios. They sped up the testing process, which used to take one person the entire summer to test one circuit: “Oh, wow, we can only test 10 percent of the connections on that circuit!” Now they’ve brought in this solution that automates the testing, and it only takes two weeks to do it. They test 100 percent of the circuit, so it’s a huge speed-up of the testing.
What they’re telling us is that this allows the engineers to devote their time to the higher-level tasks. This was sort of the drudge work that they had to do, but it was long and very painstaking, and nobody really enjoyed it, and they just had to get through it. They knew it was necessary, but now they’re able to take that time and use it for designing the next set of military radios. They can use the brainpower that they have to apply to a problem that requires creativity.
TANYA OTT: How good is the AI technology at doing that kind of task compared to a human doing it?
BILL EGGERS: In some regards, it’s better, and in some regards, it’s worse. One of the things that they had to figure out how to deal with, in an Army research lab, was the very fine motor manipulation of the testing circuits, because they have this silicon wafer that has the notes or the radio circuit on it, and then they have to take an electrode and connect it to one point on the circuit, and then another point on the circuit. It’s a very fine tolerance, within millimeters, where you have to connect to the electrodes, and so they had to be able to get a robotic system that could move with that much precision. And recently they’ve been able to.
It used to be that only humans could move with that level of precision. But then we would have to spend a lot of attention in order to be successful doing that, and we would get tired, and we would have to take breaks, and we can only do it for a certain amount of time. Now that we’ve got this robot that can do that, they don’t really get tired in the same way that we do, so they can speed through a lot more of that process.
TANYA OTT: So your projection in your research is that it’s going to make government a lot more efficient. But what happens to jobs, then, if we’re able to get something done in 30 percent less time? What’s the impact on employment?
BILL EGGERS: What our analysis showed is, at least in the medium term—again, we’re just looking at five to seven years out—that it’s not going to lead to large job losses in government for a number of reasons. Number one, it’s typically 20 to 30 percent of someone’s time as opposed to replacing an entire position, so that time can then be redeployed. Now, there are massive backlogs in many areas of government—what I hear from government officials all the time is, “We were just drowning under this all this increased demand.” So it’s going to be able to better meet that demand and extend capabilities.
We talk about the state of Colorado, where the average child welfare worker is only able to spend less than 10 percent of their week working with actual families; the rest is spent on all sorts of other tasks and paperwork and so forth. This will free up frontline employees to spend a lot more time focused on really addressing different people’s issues and on their mission. That’s what we see in the near term: Certainly, over time, just like with automation, some jobs will go away. But I think we’ll create new jobs. The one thing we never see in terms of citizen demands of government is that the demands go down, right? They’re always going up. So I think we will have a lot of problems. In the country, there’s a lot of things that need to be fixed. This should free up employees to focus more on those things that really matter, that really make a difference—more mission critical, as you’re saying.
TANYA OTT: Are you concerned about worker morale as automation becomes more prevalent in the workplace?
PETER VIECHNICKI: Yes, very much.
TANYA OTT: What’s your concern?
PETER VIECHNICKI: I think if it’s handled badly, bringing AI into the workplace could lower worker morale. Somehow, we have to make it such that people see a guy as a partner, and they have to realize that they’re collaborating with AI in order to accomplish more than they could otherwise. But it’s not obvious how we’re going to get to that realization. There is a lot of anxiety about what role I will play in the workplace within the government context. The people who are leading our government are going to have to make a bunch of decisions about what they do with the time savings that AI will give to them.
One of the things we talk about in our paper is, if it in fact frees millions of hours from government work or time, then the leaders of those government agencies will have a choice to take those hours as a cost savings. In other words, you reduce the workforce over time because you don’t need as many people to do the same amount of jobs. Or they might decide that they want to allow those same workers to do other things—allow them to innovate and create new value by doing things that they weren’t previously doing. So that’s going to be a choice that's going to be made at the top levels of government organizations.
I think that’s going to play into worker morale, because if workers realize that [automation] is being used just to slowly replace them, that would have a negative effect on them. But if workers perceive that a guy is, like we said at the beginning, freeing them from the drudgery and allowing them to create, innovate, and deliver new value to citizens, then I think the partnership with AI is a really positive thing, and it would have a positive effect on morale.
TANYA OTT: I imagine there are some jobs of which a larger percentage of them—60 to 70, maybe 80, percent of someone’s job—being automated in some way, particularly at the lower levels of government employment.
BILL EGGERS: Again, just like in our broader society, some jobs will eventually go away. Data entry: You don’t need people doing that. A lot of shuffling some paperwork and so forth, and other areas. But a lot of these jobs will also change.
I think the one thing to understand is that we have some time here. This is not going to happen overnight, or this week. The government really takes time to redesign work to figure out how to realize a lot of those efficiencies. So I think with proper workforce planning—where you’re looking at retirements, you’re looking at other issues—I just don’t perceive large job losses over time. Because you move people to other tasks, you reskill individuals, and so forth.
We have a lot of unmet needs right now and jobs that need to be done in government, and there’s just not enough people to do that right now. Think of it very much as a shift: We’re going to be able to shift a lot of people to much more impactful kinds of work that really make a difference for citizens and businesses.
TANYA OTT: Every government agency that I know of has numerous programs and processes. How do they determine which programs or processes to automate?
BILL EGGERS: There’s a number of different ways that we’ve actually looked at on how to do this. We’ve examined dozens and dozens of cases in the private sector and public sector, and we put together a kind of a methodology framework. We call it the three Bs framework: where it’s viable to actually use cognitive technologies, and where it’s absolutely vital that you use them. So there's a lot of tasks right now where it’s viable to use them—whether speech recognition or vision such as for initial telephone customer service, or the processing of handwritten forms.
There’s a lot of things now that are viable that weren’t viable even five years ago. But that doesn’t mean that just because something can be automated, it’s worth automating: Viable is not always valuable. Sometimes you have tasks that low-cost workers perform efficiently and competently, and they don’t make sense to spend a lot of money on automation.
Then there’s the whole set of business problems where putting in place cognitive technologies is absolutely vital. These are areas that require a high level of human attention perception and may be all but unworkable without the support of cognitive technologies. So we did an analysis looking at 25 years ago—a Government Accountability Office report, over 1.5 million pages, with 40,000 recommendations, just to try to understand what that told us about the management issues within government over the course of several decades. Now, for humans to do that, it might take 10 years. But we had machine learning do it, and it took a much shorter period of time—a couple of days. In the end, that is something that you simply wouldn’t be doing without computers.
So that’s where you can extend the capabilities. There are some areas like cybersecurity fraud detection where you absolutely need a learning system like this that can respond to ever-changing threats in unpredictable ways. So I think that’s the way to think about what’s viable, what’s valuable, and what’s absolutely vital, and then that helps you to understand where to prioritize some of your investments.
What are some of the pitfalls that agencies need to be careful to avoid? There’s a number of unintended consequences that are possible if you’re not doing this wisely. Automated tools can be dangerous without maintenance and just common sense. One of the studies have found that automated systems, if they’re not implemented well, could undermine worker motivation—they can cause alienation, reduce satisfaction, and so forth. So you have to pay a lot of attention to the human-machine interface and doing that well. AI programs may also have an embedded bias in them. For example, one AI program that was designed to predict the odds that recidivism could not identify important factors such as prior arrest records made consistently incorrect predictions.
So these must be done very thoughtfully and with care and really using human-centered design processes, where you’re spending a lot of time understanding how a worker can work with these technologies to augment what they do.
TANYA OTT: I think one of the other challenges—and again, this is my observation from working with state agencies as a state employee for many, many years—is that there isn’t always the appetite for failure. In order to innovate, you have to be willing to accept failure, and that hasn’t always been rewarded in a government context.
PETER VIECHNICKI: That is true. And we see that in the way that government approaches software systems in particular. But one of the things that’s happened over the past decade or so is that government has actually become more agile, to use the buzzword. And one of the things that’s part of the agile mind-set is the idea that you try things quickly and fail early. And we do see that at least on the software side, the government has been making its software projects shorter and less costly and smaller and allowing them to be tested and fail earlier than before. In the past, you know, the software systems might have taken years and years to come online and millions and millions of dollars to complete. Nowadays, by and large, government agencies are buying things in smaller chunks and cheaper chunks, and then they’re able to jettison something if it’s not working. So at least on the software side, we’re seeing that government can take that kind of mind-set and innovate.
TANYA OTT: Of course, jobs are what everyone is thinking about. Oftentimes, when we talk about artificial intelligence technology and automation and things like that, have you tried to take a crack at sort of quantifying sort of mid-term and long-term, you know, how many jobs we might lose with AI? What’s your thinking on that front?
PETER VIECHNICKI: Yeah. Well, the way we approached it was: We feel that the thinking about how many jobs are going to be lost or gained tends to oversimplify the question, because the way we see it is that when technology comes into a workplace situation, it doesn’t typically replace a job wholesale. It typically replaces certain tasks within that job, and everyone who’s working in an occupation—they’re performing a number of different tasks every day, so their job really consists of a bunch of different tasks. In fact, many economists have used the phrase that occupations are “baskets of tasks.” And so what it is going to do in our view, is it's going to come in, it’s going to affect some of those tasks but not others. And one of the things we think is likely to happen is that certain tasks will be delegated to AI and then other tasks will be complementary, and they will be the ones that the humans will spend more of their time doing. Those are actually going to grow. It’s a fairly well-known principle in labor market economics that demand for these tasks that are complementary to technology will actually result in increased wages. So there’s going to be positive and negative effects that are happening at the same time. So some of the things that people are doing are going to go away, and then other things that the people are doing—there is going to be actually more demand for those activities. And so there’s probably going to be more people brought into the parts of the occupation that perform those particular tasks.
So it’s going to be this whole set of complex shifts where the labor market is trying to get back to a sort of an equilibrium state where a demand and supply are in balance, and modeling this is very complicated. We haven’t quite gotten there yet, but it’s one of the things that we hope to do in the future. But just to bring you back to your question: So how do we think about job losses? I am going to go out on a limb here and say I don’t know. I think it’s a really difficult question. On the one hand, there’s people who I respect and trust who are world-renowned experts in trends and technology effects on labor markets that are very, very optimistic, and there’s a number of economists who believe that technology will only help with job creation over time. Jobs themselves will change as a result of technology. But ultimately, the whole economy will grow because it will become more productive. New jobs will be created as a result of this technology coming in and more people will be put to work doing those things in the end. So there’s a bunch of people who are very, very optimistic about AI.
But I have to tell you there’s another bunch of people—and they’re equally well known and equally well respected and have done equally rigorous quantitative research—that are very, very pessimistic. And they believe that we’re in a different situation now, and at some point in the not-too-distant future, AI is going to be so powerful that it is actually going to replace a number of human jobs. And then there’s going to be a bunch of redundancy in the labor market.
And I really don’t know from a personal standpoint, but in the research that we’re presenting here, we’re looking at the next five to seven years, and within that timeframe in the government context, we are not thinking there are going to be a lot of job losses. We think that the most likely effect is that worker time is going to be freed up, and then that time will be used to do things like reduce those backlogs that we talked about or do a better job at some of those core services that we talked about, like the child welfare workers or the parole officers and those kind of occupations. So that’s what we think in the near term. Long term, I think it’s still a totally open question. The answer could just vary wildly.
TANYA OTT: When you say long term, what are you talking about, because I remember seeing a statistic in some Deloitte research out of the UK government that suggested that they might lose 861,000 public sector jobs by 2030, and that’s really not that long term. Right? That’s 17 years from now.
PETER VIECHNICKI: And that’s a really powerful report, by the way, and that’s clearly verging on the more pessimistic side of the effect of AI on government workforces. Possibly it’s informed by the political climate in Britain and how that affects the way that they generate those models. But I think one of the things that we just don’t know is the rate of change of technology. It seems to be speeding up, and the development of AI is occurring so rapidly that it seems to surprise everyone who looks at it. So, you know, between now and 2030 that seems like a long time away, but the technology is changing so quickly that it’s going to be upon us before we know it.
TANYA OTT: I’m reminded of a computer that can write news stories, and there was some coverage of this maybe about a year and a half ago that completely freaked out everybody in the journalism business.
PETER VIECHNICKI: Yeah!
TANYA OTT: They set up a human reporter against the computer, covering I think it was a sports game or something like that. And the story wasn’t too bad. It could appear in a newspaper.
PETER VIECHNICKI: I know. And yeah, that’s a great example, and you see other cases where certain things that we would think of as only humans could do that—all of a sudden they’re being done by AI. And one that comes to my mind is composing music. Some researchers have figured out how to train neural networks, I think, to compose songs. And the one I heard, like you said, you know, it wasn’t that bad. But composing music—I mean, we think of that as really sort of one-of-a-kind core parts of our humanity, that artistic expression and creativity.
TANYA OTT: I mean, there’s the science of music–the mathematical intervals between notes and things like that—but then there’s the art of it. There’s the heart of it.
PETER VIECHNICKI: Yeah, exactly. So we haven’t yet seen a bunch of composers laid off. But just to bring it back to the research that we did, because we’re looking at the next five seven years, we sort of looked at AI as it is today. And so when we developed our models of what will happen in the next five to seven years, we said that tasks that require creativity are less likely to be replaced by AI right now. And so when we built our models, we included that factor in the models. Some other factors that [require] having empathy for the person that you’re talking to and changing what you’re doing based on that person’s emotional needs—that’s a task that requires social intelligence. And that’s something that AI is not good at today. Now, it’s possible that it will learn how to do that in the future, and there’s a lot of very exciting developments along those lines. But right now, AI can’t really do that. And so those tasks we also treated differently and assumed that they wouldn’t be replaced as quickly as other tasks.
TANYA OTT: So some jobs will disappear, some jobs will change because people working those jobs are freed up from more mundane tasks, and some new positions may be created. That’s kind of the takeaway, right?
PETER VIECHNICKI: I think I would say few jobs will disappear in government over the next five to seven years. I think almost all government jobs will change, like you said. Some of them will be changing more than others. There’s some where much of the work that they’re doing won’t be affected by AI. But there’s some where much of the work that they’re doing will be affected by AI, and those jobs will change more dramatically. And then, like you said, I believe new jobs will be created, and we’ll see that. But I think one of the main takeaways from our modeling is there’s a choice to be made, and the choice is how much money and effort is government going to invest in AI technology. Is it going to go big in AI and transform itself, or is it going to just kind of watch and wait and eventually buy off-the-shelf AI when it becomes cheap and readily available? And so the level of investment is really going to affect how quickly government transforms, and that’s going to require big funding and political decisions that are going to have to be made. And that was kind of outside of the scope of what we were able to model, but we create some scenarios that show what the different effects would be like if you had, let’s say, a massive investment versus a more modest investment.
TANYA OTT: Peter Viechnicki and Bill Eggers write about how AI can improve speed, cost, and quality in the public sector in an article published by Deloitte University Press. You can read it at dupress.com. While you’re there, be sure to check out our podcast archives, including a recent show on the opioid epidemic that seems to resonate with a lot of people.
I’m Tanya Ott for the Press Room. Have a productive day!
This podcast is provided by Deloitte LLP and its subsidiaries and is intended to provide general information only. This podcast is not intended to constitute advice or services of any kind. For additional information about Deloitte LLP and its subsidiaries, go to Deloitte.com/about.