How technology can help save democracy | The TechTank Podcast

[music]

CO-HOST DARRELL WEST: [00:00:00] Thanks for joining our Brookings TechTank podcast. I’m Darrell West, a senior fellow in the Center for Technology Innovation at the Brookings Institution. Technology is accelerating at a rapid rate and altering how governments and businesses make decisions. But there is concern about the way in which our institutions operate and the manner in which AI and other tools are altering communications, decision-making, and governance itself. To discuss these important questions, we are pleased to be joined by a distinguished expert. Beth Simone Noveck is the Northeastern University professor and director of the Burton Center for Social Change and the Governance Lab. She is also the author of a new Yale University Press book entitled “Reboot: AI and the Race to Save Democracy.” In it, she offers a roadmap to how we can handle the current moment and the ways technology can expand participation and improve service delivery. Beth, welcome to our Brookings TechTank podcast.

GUEST BETH SIMONE NOVECK: I am delighted to be here. Thanks for having me.

CO-HOST DARRELL WEST: I know we met many years ago when you worked in the Obama administration, and I was always impressed with how thoughtful you were about the role of technology. And of course, now you have this new Yale book that talks about how governments are using AI, the challenges they face, and the possible paths forward. So I’d like to start just by asking you about the book’s title, “Reboot: AI and the Race to Save Democracy.” Why do we need a reboot?

GUEST BETH SIMONE NOVECK: Well, why, why we need a race is hopefully very clear. We cannot act fast enough, I think, to address the crises that our democracy is facing today. Whether it is the 7% number that is the view of most American young people about the health of our democracy, that is a… it is so low to almost be nonexistent that [00:02:00] young people today do not feel our democracy is healthy. We know that rates of trust in government are low and getting lower. American people don’t feel that government n- does what it needs to do in terms of solving the problems that they face. Whether it’s the political climate in which we live, in which partisanship is the, become the norm and violence is rising, I think we all recognize that democracy is teetering on the brink. And globally, there’s only 29 democracies left in the world. So on the one hand, this is partly about racing to address those challenges and problems. But I specifically use the term reboot because I think it’s not just about fixing and repairing what we’ve done before, but hopefully also doing things qualitatively differently, really trying to have the democracy that we deserve, the government to which we’re entitled and to restore the culture of democracy that we’ve all yearned for so long, not by, again, doing things the same, but by doing things better.

CO-HOST DARRELL WEST: Actually, I like that notion not just of trying to fix things, but trying to do things better. So if you were the czar of this democracy and could make whatever changes you want, which I know is a contradiction in terms right there but what are the things that you would do differently? What are the ways in which we could use technology, or how can we redesign our institutions so that we have a better democratic system?

GUEST BETH SIMONE NOVECK: I think let’s start with the whole notion of democratic institutions in and of themselves. So to the extent to which we are supposed to live in a system in which it– we are at least to some extent self-governed, in which those people who are governed have a voice and a say in the decisions that affect their lives, I think we would all agree that we have really strayed very far from that ideal. So even in a representative system, in a system in which most of our lawmakers are spending two-thirds of their time dialing for dollars instead of what we sent them to Washington for, which is [00:04:00] actually engaging in the work of legislation a system in which both at the federal level, the state, and the local level, we have vanishingly little voice, and those people who have dollars and lobbyists have much greater influence over those processes I think we can all agree that is not working as well as it needs to. And so what I would like to see and what I’m excited about is the opportunities really for new forms of engagement and participation that AI can enable precisely because they are becoming not just more effective to do but also more efficient to do. So let me just be clear what I mean by that. There’s a lot of reasons why we haven’t had voice and why money in politics has been a problem, but I think at the minimum, one of the big reasons we haven’t been able to participate is because we haven’t known how. I just testified recently on Capitol Hill in front of a bipartisan subcommittee asking the question: How can we actually use new technology to engage with and listen to the American people? It has been hard to do that, right? We’ve had 30 years of the web allowing us to talk, and that has made it harder for our institutions to listen. So I’m not just talking about people shouting at a town hall. I’m talking about, we put up websites and give people a way now, anybody to participate. We’ve moved from a world in which you have to mail in your comment to a federal agency to one in where I can go on regulations.gov and submit a comment, but what happens? What happens is, what happened when the FCC did its net neutrality rulemaking, and you get 22 million comments. What happens is, what happened under Obama when we asked the American public to say what the first 100 days of the president’s agenda should be, and we get 125,000 comments. So it is no wonder that we so rarely ask because we do not have the ability to listen. So now what I’m most excited about, if you wanna ask me, what it is that we would do differently, it would be that we would shift to institutions where participation, engagement, where the use of our distributed collective wisdom and collective know-how is not [00:06:00] something we do once a year at the ballot box, it’s not that we do once in a blue moon when we do some, citizen engagement exercise and in front of the cameras, it’s something that we do as a regular part of how we work and how we’re doing business, and that’s what I try to illustrate in the book, is the many places that are moving in that direction already.

CO-HOST DARRELL WEST: So you talk about the need for new forms of engagement and participation. You want people to engage not just once a year, once every other year when they’re voting, but on a more regular basis. What would that look like? What are the vehicles to do that? What are some of the new ideas that might actually enable new forms of engagement and participation?

GUEST BETH SIMONE NOVECK: So let’s take some very concrete examples so that it’s not abstract. Take a place like Brazil, again, plenty of challenges and problems large country, larger legislature, high degree, as we know, of hyper-partisanship. Yet in the Senate, for example in Brazil, they have not one but four different ways that the public can participate in the lawmaking process. Like many places, they have the ability for the public to actually suggest a bill and other people to then ratify, sign on to, and say, “Yes, I’m in favor of that proposal.” They have now, you could say, “Oh, gee, I’ve heard of that idea before, the idea of some kind of petition.” We had something like e-petitions in the UK earlier. We had on whitehouse.gov under the Obama administration the ability to suggest a policy. We had this system. The difference is now that a small team, a small group of civil servants working for the Brazilian Senate uses AI to help make sense of that welter of input that they get. When they have a hearing now in their chamber, all of those hearings are open to the public to write and suggest questions for lawmakers to ask. And that is now not a kind of nice-to-have once in a while sort of thing. It is now the majority, the vast majority of the questions get asked are taken by senators from what the [00:08:00] public has proposed. It’s become the norm in how they do things, giving people more of a voice. There is also an opportunity for schoolchildren to do so they have a special sort of mechanism that involves some lessons and tutorials and kind of teacher plans, but that encourages classrooms to actually get involved around a policy issue, trying to promote civic education and again, giving kids a way to propose legislation. None of that would be possible if, again, there wasn’t the use of AI to allow, for example the small staff to be able to more quickly group, synthesize, collect those suggested questions, and then match them to the topic of the hearing to ensure, for example, that we can filter out those things that are off topic, find things that are relevant and then be able to very rapidly manage that sort of mass of information. So you’re seeing lots of great examples like that, and not just in small places. We’re all very used to the idea of, “Oh, Estonia does it better. Singapore does it better.” But I wanted to pick specifically an example from a big, complex, hyper-partisan country like Brazil, although I can give you many others besides, in other places and cultures where they are moving towards this more engaged way of working. It’s still extremely early days And extremely early days in terms of really thinking about how do we bring that kind of collective intelligence into our work.

But I’m excited by the fact that California launched Engage California after the wildfires, and then because they used AI making it cheaper and easier for them to manage the process, they’ve now repeated it three times on other topics with different audiences. So again, that participation thing becomes not just a nice to have wave at the cameras thing, but becomes much more about how do we create learning institutions, learning organizations that know how to better take on input and manage information.

CO-HOST DARRELL WEST: No, I love that idea, and the Brazilian [00:10:00] example of ways people can participate during the lawmaking process I think is a great example of how technology can be part of the solution. We’re obviously used to thinking about technology screwing things up in a lot of different respects of our political process, but there may actually be ways it can be part of the solution as well. In your book, you also talk about ways that AI or other technologies might be able to reshape voting and elections in constructive ways. What do you mean by that? What are specific ways in which that might take place?

GUEST BETH SIMONE NOVECK: So I know it probably doesn’t feel like this at the moment. Every headline is about the corrosive impact of AI in elections. We’ve seen the use of really racist and misogynistic deepfakes, for example, in recent contests. You can pick your video du jour, whether it’s in the Los Angeles mayor’s race or whether it’s in the Michigan gubernatorial race. Internationally, we’ve seen the use of deepfakes, for example, in the Hungarian elections rather recently. So I think a lot of the conversation is around the polluting of our information ecosystem with people using AI to rapidly generate hyperrealistic ads to attack their opponent, for example, or to self-aggrandize while, claiming your opponent said something they didn’t say or to make up some, you know– pretend that your opponent said something, again u-using AI to fake something that they didn’t say. So I, I would just be remiss if I didn’t point to the fact that we’re all very worried about the information and disinformation that’s polluting our political discourse. It’s why 49 states have some kind of legislation or regulation against deepfakes and we have at least in, in some context, some regulation at the federal level of the deepfake problem. But there is also empowering uses of AI, especially in the electoral space, and I wanna see us talk more about, invest more, pay more attention to those things, not just to [00:12:00] play defense. So for example, Eric, the compact between 25 states that is designed to clean the voter rolls and to put, as Eric has done, 10 million people back on the voter rolls who are entitled to vote, Eric wouldn’t work without the use of AI to essentially sift through large quantities of data. These are just, AI is just really a data processing tool. And places like Eric can use it to sift through those voter rolls, find somebody who has moved between two states and who might have gotten lost along the way, and ensure that they are added back onto the voter rolls and re-enfranchised. It’s groups like Vote ER, part of a project called Healthier Democracy, a nonprofit that goes into emergency rooms around the country and helps to register people to vote and also sign up people for government services. They use AI to help them create better websites create materials that they can then use cheaply and easily to enfranchise people, give them their right to vote back, or give them access to government benefits. It’s places though, again, we’ll come back to Brazil since we mentioned it, which use AI to read you your ballot. So we all know that we can talk to AI now, not just type with it. So a place like Brazil is actually ensures that when you go into the voting booth, if you are vision impaired or maybe just like me, you tend to forget your glasses and can’t see anything, your ballot can be read to you using an electronic voice so that you do not miss out on the ability to vote in an informed manner.

So that’s just a few examples. There are many more besides. But again, it’s about shifting the narrative, shifting the conversation to asking about what we should be doing with AI to make our elections better, not just what we shouldn’t be doing and how do we play defense against these deepfakes that frankly haven’t been borne out to play that much effect or have that much impact on how people do and don’t vote.

CO-HOST DARRELL WEST: So you mentioned service delivery here, and I’ve always been interested when we think about democracy, some [00:14:00] people are really focused on the process angles and ways we can use technology to improve elections. But it’s also important that we think about ways technology might actually help to solve problems, because if people see democracies failing to solve problems, they are going to lose faith. You mentioned that in your opening comments. We know trust in American government is very low right now and that’s true for a lot of governments ar-around the world. So I’m just curious how governments might be able to use AI or other technologies to improve service delivery.

GUEST BETH SIMONE NOVECK: Nobody knows more about this than you. You are the author, probably the f- if not the first, then the most important book on digital government that and have been writing about these issues for a long time.

CO-HOST DARRELL WEST: Flattery will get you nothing on this podcast, Beth.

GUEST BETH SIMONE NOVECK: Yeah for the mas- it’ll get me double what you’re paying me to be here. So the, I think the attention to the ways in which technology can help us deliver better services, implement our policies, has been a space we have occupied for a long time. And I think as much as we wanna be critical about, as we tend to be about the functioning of government, we have made a lot of progress. We live in a world in which I can not only get information 24/7 from government, I can transact with government online. We’ve come a long way from every service having to wait online nine to five in an office. The sloths in the DMV metaphor of Zootopia, we’ve, we have gotten away from that, but not far enough away from that. So the exciting thing, I think, is the ability to do just much, much more of that faster and cheaper now with AI because again, these are very powerful information processing tools. So for example, the notion that every service could be delivered with language that is accessible to people, that now I can take legalese and government speak and translate it into something that anybody can understand, both in plain English. We’ve been fighting for the– we know the folks who have been [00:16:00] advocating for years and years to implement our legislation on plain English, on trying to ensure that we actually just communicate in a way that we’re telling the American people, “This is the benefit to which you’re entitled, and this is how you apply for it.” We don’t do a very good job of that right now. We make it really complicated with complex forms and complex directions. We can get better at that now because AI can just help us to do that translation, obviously into multiple languages, but also just into simpler terms. So let me give you a quick example. I work with a team of student fellows that we call our AI for Impact Fellows. They take a sabbatical from their studies. They get paid to build AI for civic and government use. They’ve worked with a group called Innovate Public Schools out of California to build a tool that helps parents of kids with disabilities understand, summarize, translate their child’s IEP, the Individualized Education Program. That right now is a 100-page document you get from your school district that says these are the services to which your kid is entitled, extra time on the test or extra counseling or whatever it may be. If you are low literacy, if you are not a native speaker or just you’re a busy person that 100-page PDF can be really hard to make sense of. What are the rights that, you know, and services my kid is entitled to? How do I advocate for my child? So together with a network of parents, none of whom I might add are native English speakers, we designed and built a free tool to help people manage that process. Now, would it be better if that process were simpler and we didn’t give people a 100-page PDF? Absolutely. But it’s gonna take us, for everything like that, we don’t have enough hours in the day to change all of those policies, to legislate all of those things. So in the meantime, we can do a lot just to make government more accessible, to make it more conversational by ensuring that you can, for example, interact with a chatbot to get answers to some of your questions. In New Jersey, when a business has a question about starting, running, or growing their business, they call a hotline [00:18:00] called the Business Action Center. The great part is the Business Action Center is staffed by very knowledgeable professionals who know all of the rules, regs, grants tax breaks, opportunities available to businesses, but those people have to go home and eat sometimes.

So there is a tool where you can now ask your question and get an answer 24/7. The tool doesn’t make up any answers. It only gives you an answer that has been vetted by a human. So if you ask a question that’s never been asked before, it will tell you, “We’ll get back to you and email you tomorrow,” allowing a human to check it in the meantime, and then that question and answer are loaded in the database, so the next time someone asks it, they will get an answer within seconds. So it’s that kind of thing with humans in the lead, AI as the helpmeet that allows us, I think, to do more to make convers- to make government both more accessible, more conversational, more responsive to the needs of citizens.

CO-HOST DARRELL WEST: I think all those are terrific examples of how technology can help improve service delivery. Certainly the language translation aspect that you mentioned the vision help for those who are visually impaired, the chatbot answers available 24/7. Those are all technologies that are here now, can be implemented. But I’m just curious, what challenges do government agencies face in implementing AI solutions?

GUEST BETH SIMONE NOVECK: It is a problem both of technology, but I think even more of process. So as we know, most governments are not set up today. We have hollowed out much of the capacity of government to be able to build and run its own technology. So you have places like Colorado and New Jersey, for example, that have a digital team or an innovation team. They go by different names. California has something similar. But not every… even in places where you have a kind of, let me call it an innovation shop that is used to designing and building new products whether you have that or not, it can be very difficult for government [00:20:00] to have the wherewithal to be able to adopt an innovative or new product. Number one they just don’t have the people trained in AI. They don’t know how to test the tool. They don’t know how to incorporate the tool. Secondly, AI tools are very special in the sense that they do not behave the same way over time, and they change how they act depending sometimes on the underlying AI model that’s used in them. So I can have a tool that spits out an answer today in one way and then spits out a different answer tomorrow. So you really need AI to test these things over time and make sure that they continue to work over time. But lastly, and perhaps more important, it’s not so much about the tech as about how we use the tech incorporated into our own work.

So when I start to add the use of a chatbot or add the use of an algorithm or add the use of an LLM to be able to do things differently, I have to change how I work. Let me give you a quick example here. This one came across my desk not quite a government one, more of a healthcare one. But in the world of healthcare, Kaiser Permanente developed an algorithm to help predict when a patient is likely to regress in the hospital. So there’s certain red flags and data indicators that tells you this person is likely to go backwards rather than forwards. It’s great data on the effectiveness of this tool. We’ve seen lots of predictive AI that doesn’t work very well. This one is very promising. What they’re struggling with is not the tech. What they’re struggling with is the question of whom do we give it to? Do we give it to a nurse? Do we give it to a doctor? When do we use it in the course of patient care in order to have the maximum effectiveness of it? And that’s the same thing we’re wrestling with in government is how do these things change the way we work, and how do we incorporate them into the way that we work? That requires some thoughtful consideration, not just buying a tool.

CO-HOST DARRELL WEST: One of the things I liked about your book is you embed your [00:22:00] analysis of lawmaking, elections, and service delivery the broader context of economic forces the rise of market power in the tech area. Of course, it’s clear we have a small number of very large tech firms that have tremendous influence over the options we have and the kind of information that is available to us. So what do you see as the risk of concentrated AI power among the major tech companies for all these questions that you’re talking about, democracy, governance, and policymaking?

GUEST BETH SIMONE NOVECK: So I think we do have to be significantly concerned about the tools we are putting into our governments, we’re putting into our schools, we’re bringing into our homes. We’ve seen the consequences of fully privatized social media and companies that are seeking to maximize clicks, maximize our engagement and what that’s meant for the health of our political and cultural discourse, right? The sort of manufactured outrage that we are persuaded towards on social media because it engages us, keeps us online, and keeps us looking at advertising. We’ve seen what the consequences of that have been. So we’re rightly asking the question now about who owns these tools, how we regulate them what the values are of these tools. There’s obviously a lot of debate around regulation who has the right to regulate, what we should be regulating, AI safety how we’re testing and training these things.

There is not enough conversation about the much if you ask me, easier and faster route, which is procurement. We’re buying these tools. You have governments paying to give every employee access to these tools, and while that may still dwarf kind of use in the private sector, government is and always has been a very large purchaser of technology and has the opportunity to push back, not just through legislation and through the [00:24:00] legislative branch, but also through the executive branch in the procurement power process of saying, “We want these tools to work differently. We want them to work better.” And I think it’s important that we use that power more effectively. Number two, I think it’s extraordinarily important that we not just worry about the inputs, not just worry about the training data going into these tools, but that we have an ongoing process of monitoring of the effectiveness of these tools as we go forward, especially the tools that do more than just, translate a sentence here or there, the things that are, the algorithms that are actually involved in decision-making, the agents that will actually take decisions for us, especially if they are involved in a decision about depriving somebody of a benefit or rewarding somebody with a benefit, deciding on whether somebody goes to prison or how they get sentenced, something that affects somebody’s life meaningfully we want to be very careful about the tools that we’re using. So I’m excited about the movement towards what some people call public AI, or I would call public purpose AI. I don’t think it means government has to go out and rebuild ChatGPT or Claude for itself. We want a commercial marketplace to thrive. We want the innovation and the jobs that come from commercial investments. But I do think that on top of these large language models, we can build a whole lot of free and open source tooling that is more transparent. I think we can use our procurement power even where we’re buying commercial things to ensure that those things are more explainable, more accountable, more transparent and that we’re doing a better job and in demanding the companies do a better job of supporting the testing that we need, not just before the fact, but after the fact, to ensure that whatever the tool does today still works tomorrow, right?

My tool might be really good right now at using data. For example, City of LA has implemented the use of an algorithm to predict when somebody’s at risk of being homeless. Works very well now. They’re showing results like 71% of people, and I don’t know quite how they’re measuring this, [00:26:00] 71% of people, they are getting to them engaging in early interventions that they feel are preventing them from going homeless. Whether those are the right data points over time, does that algorithm continue to work? Is its predictive power as good as it needs to be? That is not a one and done. We need an ongoing testing process, and that starts with knowing what these tools are and how they work.

CO-HOST DARRELL WEST: Yeah, we definitely need more testing and more assessment of these things before things are released into the wild, which seems to be the pattern today. So my last question for you is whether you are an optimist or a pessimist on the long-run future of technology. Do you think AI ultimately is going to strengthen democracy or harm it?

GUEST BETH SIMONE NOVECK: I can take no credit for this term, but I am neither an optimist nor a pessimist. I am what Hans Rosling called a possibilist. I think one would not be wise to be either at the moment, but except to be… What I wanna be is to- hopeful, to paint a picture of what’s possible, to talk about the actions and investments we need to take and to make in order to realize that future without being Pollyanna-ish about the inevitability of achieving it. I don’t think it’s inevitable at all, and we’ve seen what’s happened with the internet, what started with a great degree of euphoria that this, we were somehow inventing the new public square, that this would reinvigorate our democracy, that suddenly everybody would have the right to speak, and all things would go right with the world. We saw what happened with that dream. Realized to some extent in many good ways, lots and lots of good things. Many things have gone wrong that we failed to anticipate. So I think we have to be hopeful possibilists. We have to fight for what’s possible, and I think that’s why it’s so important to have this conversation around democratic AI, ’cause the danger is if we’re only having the pessimistic conversation, if we’re only talking about the robots eating all the jobs, if we’re [00:28:00] only talking about the end of humanity, the loss of human agency the, robot overlords we are distracting attention from the here and now from talking about the policies we need to ensure that we have good jobs with AI, to ensure that we are fixing our institutions, that we are rebooting and improving our, how our institutions work. And that’s also the s- by the same token why we can’t just talk about American competitiveness and why it’s so great that the stock market is going up and we’re beating China at the AI race. Not clear that we are. But we, the wealth that’s gonna be generated, the jobs that are gonna be created, that conversation too, which has led to saying, “Oh, we can’t talk about regulation because we can only talk about the great economic benefits.” Both of those are wrongheaded distractions from the much messier, more nuanced middle, but one that is, I think, just as possible, if not more which is the world in which we are investing in universal education in AI, the one in which we are using our procurement power and our regulation to push back on the companies that do exist, and building alongside them public and open complements to the private marketplace, one in which we are engaging with communities and developing and building these tools. And most importantly, that we are centering, that we are focusing on the questions that matter, not just asking, “How can I make, more pictures of fill in the blank AI slopaganda?” I do like to make pictures, I confess, of, our cat dressed up in funny costumes, but there are many more important questions that we need to be focused on.

If we’re gonna be setting up data centers, investing in energy, investing in compute, we should be using them then for the, to the end of making government work better and making democracy more effective.

CO-HOST DARRELL WEST: So I want to thank Beth for sharing her thoughts with us today. Her new book is “Reboot: AI and the Race to Save Democracy.” It’s available from Yale University [00:30:00] Press, as well as your favorite local bookstores. So Beth, thank you very much for joining us and sharing your thoughts on technology and democracy.

GUEST BETH SIMONE NOVECK: Thank you. It’s been a pleasure to be here, and I welcome the reactions to the book and to the idea of democratic AI. You can read it, you can even listen to it. And I look forward to talking to you, Darrell, above all further, about what we can do to realize a future vision for democratic AI.

CO-HOST DARRELL WEST: So at Brookings, we write regularly about technology and democracy. You can find more information on our Brookings TechTank blog located at brookings.edu. Thank you very much for tuning in.