Former New Jersey official argues AI could strengthen, not weaken, democratic institutions

Few technological advances have caught the attention of state and local governments like artificial intelligence, both for its promise but also its pitfalls.

One state that witnessed AI’s opportunities and wrestled with its challenges was New Jersey, which has been a leader in adopting the technology both among its residents as well as for government employees. The state has, among other initiatives, embraced an AI assistant to help save employees time on various tasks, expanded the technology’s use for public benefits programs and instituted an employee training program for generative AI.

At the heart of the state’s AI embrace was Beth Simone Noveck, New Jersey’s first chief innovation officer, then its first chief AI strategist. She is also a professor at Northeastern University who directs its Governance Lab and Burnes Center for Social Change.

This year, Noveck joined the Australian Resilient Democracy Network as a Global Fellow, and published a book, Reboot: AI and the Race to Save Democracy, about the fragility of democratic institutions, and how AI could help strengthen them, if used correctly.

Noveck sat down with Route Fifty recently to discuss her new book, her vision for the future of AI and its role in promoting democracy.

This interview has been edited for brevity and clarity.

ROUTE FIFTY: Why did you write this book, and what is the overarching message you want people to take away?

BETH SIMONE NOVECK: I wrote it in large part because I think we're having the wrong conversation about AI, where you're talking either about the robot apocalypse and how AI will be the death of humanity and the end of all jobs, or we're having a euphoric conversation about how AI will solve all of our problems and how it will compete with other countries for dominance in the AI marketplace. What we're failing to talk enough about is that messy middle of how we are already making more progress on using AI to repair our institutions, improve governance and focus on our democracy. It was really an effort to shine more light on what's already happened, and what we need to do more of to use AI well, so that's the motivation.

What I also hope people will take away is that we have this singular opportunity, and there's both the crisis that we're facing in our democracy, there is the subversion and fear of subversion of our elections. There are the failures of representation and the lack of voice that people have in their government. There is the sense that government is lacking capacity and doesn't do as well as it needs to solve problems for people. By any measure, our democracy is not doing as well as it needs to, and that creates an opportunity for us to look at how we can use these powerful technologies to address that change.

ROUTE FIFTY: Why are we so glommed onto those two themes that you talked about? Why are we not having the conversation you want us to have about AI?

NOVECK: If it bleeds, it leads, and we love to talk about what's going wrong in the world and it makes for good headlines. Number two, we love to talk about the sport of politics and who's up, who's down, who's winning, who's losing, and this predates AI. We don't talk enough about what happens the day after the election, and the hard work of governing. Getting attention for these topics used to be more difficult, and I feel like now, there is more interest in questions of government functioning, government capacity. There's a recognition that government is an incredibly important set of institutions for solving problems, and they need to work better… The third thing is the messaging that's often pushed at us, and we have to look at where they're coming from. Even [telling us] AI is a dangerous thing is often said by the very same people who make these technologies, and it is done as a distraction from focusing on some of the more urgent and immediate issues to focus on when it comes to AI and how we use it. It's long been a problem that we have not focused enough on. We have not focused enough on the question of actually having fixed the problems of our democracy and our institutions here and now.

ROUTE FIFTY: You were with the New Jersey state government in various roles. What role did that play in shaping your thinking?

NOVECK: For us, I do think that [role is] as the first state to have the new AI use policy to encourage all of its workers to go out and learn how to use these tools, training all our workforce in AI, not because we're so AI lucid, shall we say, because we rightly viewed the fact that this tool, the next generation calculator, the next generation load processor, is something we all need to know how to use, both to understand what these tools are good for and what they're not good at, but more importantly, how do we think about using them to serve the public better?

I tell a lot of stories about the work we did in New Jersey, but also work we've done around the world to focus on how we are using these tools to actually serve the public interest to do the work of government. Stories like using AI to comb through our databases and find 693,000 kids who are entitled to a food benefit who weren't getting it because they got lost between two databases. These powerful data processing tools are really good at trawling through huge amounts of data, and we were able to find those kids who otherwise would have gotten lost.

It's the ability then to use AI to sort through and synthesize and summarize citizen comments that allowed us to do a lot more. For example, public engagement and asking people, both the public and workers for the state, what should be our AI policy? How should we be using these tools? What do we want to do with them? What don't we want to do with them? To do things differently in ways that we might not have been able to do before. It was very much inspired by that work. 

The Chief Forester of New Jersey, we went out and started asking, what should be the future of our state parks? How should we design them? What services do people want? He was able to go out and do that and actually act on people's feedback, because he had AI to make it easy for him to do that. He's trained in forestry, not in computer science, so the tools enable better ways of working with information that were not possible before. Definitely in New Jersey we had a lot of very entrepreneurial public servants doing great stuff, and because we did have policy for them, we gave them access to a platform, we gave people access to AI, and then we also trained them, we were beginning to unlock some exciting public entrepreneurial spirit.

ROUTE FIFTY: Does that speak to this idea of fixing our democracy a little bit, and showing what the technology can do to help make people's lives better?

NOVECK: A lot of the reason why I wanted to write this book was to show what some people are beginning to do in the hope that we can engender more of that. Government trust is at an all-time low, as we know people's concerns about and fears for our democracy. Only 7% of young people in America think democracy is helping and functioning as it ought to. Globally, we have, by some measure, only 29 democracies left in the world, and I think this decline of democratic institutions really has at its root — there's many different complex long historical reasons, but if we can pinpoint one thing — it really is this lack of government capacity, the ability of institutions designed for an age of information scarcity who are now struggling to keep up with and be effective dealing with incredible challenges of the current era. People don't trust our institutions, because, in many cases, our institutions don't deserve people's trust, they're not resourced to do as well as they need to. Where my hope is that, used well, these tools can really help, but we need to have an agenda for doing that nationally.

ROUTE FIFTY: We've got Congress trying desperately to stop states regulating AI. We've got everything coming out of the Department of War with Anthropic. Who's in charge of this technology?

NOVECK: At the moment, some very powerful companies that are extraordinarily wealthy and stand to become much, much wealthier, especially as two of the big players just announced they are going to go public, we're going to create a new class of trillionaires. We've seen the consequences both with our social media and now with AI, what it means to have fully corporate control over platforms that we're using for how we speak, how we have conversations, how we process information. We've seen the consequence of what happens when your AI starts spitting out, for example, white supremacy or some other racist meme or whatnot. We have very little public accountability, public oversight, public insight into these tools and how they're used.

One of the reasons to use the term democratic AI to insist on naming this thing, is really to be able to call attention to the fact that we need much more democratic control over these tools. That doesn't just mean regulating what the training data is that goes into them, or just regulating testing these before we deploy them. There is so much more that has to go into thinking about what it means to create AI for public purpose. You see a lot of jurisdictions now beginning to experiment with creating their own large language models, either as a government project or as a public-private partnership, or using their procurement or regulatory authority to push back on private companies and make demands. You just say you want to have more control over these tools than the last generation of tools, that you recognize the consequence, especially with social media, where now we worry about the addiction it's created, especially for young people, the impacts on mental health, and the consequences of essentially ceding the public square to fully commercial outfits that have very little public oversight.

One of the helpful things is that, maybe, as some of these companies go public, we will at least get more insight into how they operate. But we need much better ways of building these tools in an open way, in a collaborative way. Smaller models that are publicly run, governed and controlled. We really need a public AI infrastructure to complement the also necessary commercial infrastructure that we have.

ROUTE FIFTY: Do you think governments are up to the task?

NOVECK: We're seeing a lot of governments outside the U.S. take strong action, motivated —  my concern is — much more by this debate around so-called AI sovereignty. One of these I'm excited about is the role states can play using their collective action, their purchasing power, but also their collective design and building power. So much of the debate with regulation, the desire for a moratorium on state regulation, is still very focused on a narrow set of issues, which is the safety of the model and who is testing data that was included. That was just a fraction of the set of issues. I think instead about, for example, how Los Angeles has worked with UCLA to design [a tool] and be able to predict people at risk of homelessness, and to use that algorithm to engage in what some people call anticipatory or one-click government, to call people before they become homeless and offer them services identifying the people most at risk and potentially most in need.

I think of the GrantWell project that the AI for Impact Fellowship students built with Massachusetts to help small cities and towns apply for grants, and now taking that tool free and open source and rolling it out to multiple states so that they have access. It's free, it's open source, it doesn't cost the taxpayer anything, but more importantly, it's tools we've learned in open collaboration with citizens that are designed for public purpose. I think there is a huge amount that we are already doing and much more that we can do if we focus much more on this question of AI for public purpose, and we start asking the question, what are the problems that we're actually trying to solve with these tools?

We're having lots of debates about tech in schools and the dangers of mental health and phones and all of this, which is an important conversation, but what we're not doing is having the conversation about the fact that only 40% of kids in this country read at grade level and why we're not racing to make investments in tooling that could help give every kid a personalized reading tutor. Yes, we should not have kids on TikTok in the middle of the school day, but we should be asking how we can augment the teaching staff we have to give more kids access to a reading tutor or a math tutor, the ability now to use AI to give little Jimmy a book about dinosaurs, which interests him, and Jane unicorns or trains, which interest her, so that they're more engaged in reading. Those are the conversations we need to be having, and my concern is that, when we're only talking about a very narrow set of issues around governing AI, we're not talking about using AI to govern.

ROUTE FIFTY: You've drawn this parallel already, but I do want to double click on it. Is the social media phenomenon the worst-case scenario that could happen if we get this wrong?

NOVECK: We're already seeing some of the consequences of putting these tools into our governments, our schools, our communities. Tools over which we have no insight, no oversight, that are run by companies that are not accountable to us. The concern that we raise, for example, to say we're going to train teachers in AI and put AI in schools, is that AI that's designed to maximize corporate profit, or AI that is designed to actually benefit kids while protecting their privacy, safety and security, or are we going to be putting in AI to push advertisements to people to create more addiction and profit and eyeballs. The danger is that we see the same problems with social media are amplified 100-fold. People who bring these large language models into their lives become very dependent on how we use them, so best case it's trying to sell you a product and worst case, it's pushing you a political agenda, trying to control how we think, how we talk, and creating new forms of addiction.

There's a lot of concerns about the future of human agency and human creativity. There are concerns about displacement of workers, obviously, that have existed for a long time, but we have to come back to the fact that this is a choice. These are tools. AI doesn't do anything by itself. AI doesn't fire workers; CEOs fire workers, managers fire workers. AI is not as autonomous as we are led to believe. These are tools for processing information, and, in this day and age of information super abundance, there are huge benefits we can reap from them, depending on the problems that we're trying to solve.

I'm most excited, for example, about those jurisdictions that are saying they have for too long operated behind closed doors, and with frankly some good reason, because when you ask the public to give input, it can be very, very hard to manage more information coming at you, especially in a polarized and highly politicized environment. But I'm excited about those places that have gone out and said they recognize that the public has beneficial insights to share, have collective intelligence and wisdom, know how to share, and can go out and ask people.

Because AI can make it easier to listen, you have places like Hamburg, Germany saying they're going to do all of their city planning decisions, they have AI tools to help make sense of what people are saying, and they've shared those tools free and open source with nine cities, with places like Bogota that have gone out and said they're going to go out and expand their participatory budgeting efforts to go out and hear from 50,000 people in two weeks in their last participatory budgeting effort, and to recognize that more participatory, more robust democracy is actually something that could be possible. But it's not automatic. It really is going to depend on how we choose to use these tools, how we design them, and what kinds of problems we're trying to solve with them.

ROUTE FIFTY: What should a government executive be doing right now, today, on AI?

NOVECK: The two big things we need to think about, one is universal training and learning, and how important it is that we're not only understanding what these tools are, but how we use them in our day-to-day work. There's a big difference between these two things. It's one thing to understand you can use AI to summarize something, to write something, to edit something, and another thing to incorporate them. I can have, for example, summaries of every meeting, whether it's internal meetings or city council meetings or whatnot. What do I do with those? I can collect police body camera footage, and now I can use AI to analyze that footage, but what do I want to do with that? It's the North Carolina police precinct that's using the ability to create summaries to create training videos to learn from these to train officers, as opposed to just collecting this information. The learning and the conversations around how we want to use these things is incredibly important.

The public defender in the state of New Jersey, an office that, until this spring, did not have Wi-Fi — we're talking under-resourced in the extreme — sat down with her team. They didn't just go out and build a chatbot or buy a tool or adopt AI, they sat down and they had a conversation around using the tool in their work, getting everybody hands-on to say, “What is it we want to do with these things? What is the problem we're trying to solve, and where might AI help us?”

We're seeing this in all of our data around the people through the InnovateUS project that I run, where we run this peer-to-peer free learning community around AI and other skills. The number one differentiator for people around adopting and adapting to the use of new technology is whether leadership gets training or not. It's not the staff; the staff will tell you it is whether their leader understands what these tools are, what they're for, and can therefore steward that conversation.