Reboot: AI and the Race to Save Democracy, with Beth Noveck (TRN5-V78)

Transcript: Reboot: AI and the Race to Save Democracy, with Beth Noveck

[00:00:00 The video opens with the CSPS logo. Vanessa Vermette appears full screen.]

Vanessa Vermette: Welcome, everyone, to today's session: Reboot AI and the Race to Save Democracy. Thank you all so much for making the time to be here today.

[00:00:16 Overlaid text on screen: Vanessa Vermette, Vice-President, Canada School of Public Service]

Vanessa Vermette: My name is Vanessa Vermette. I'm a Vice President here at the Canada School of Public Service. It's my pleasure to be here today and welcome you to today's event.

We are really at a pivotal moment for the public service and for public institutions globally. Certainly here in Canada, you will have all seen the release of the government's AI strategy very recently. We're navigating a landscape that's defined by declining public trust, rapid and overwhelming technological change, and really an undeniable strain on democratic systems. For public servants, like ourselves, the pressure to react, to regulate, and to manage risk is really huge.

Today, we are here to shift our gaze, however, from what we fear about technology to what might be possible to build with it. Our guest today is uniquely equipped to help us do exactly that. It is my absolute privilege to introduce Beth Simone Noveck. Beth is the director of the Burnes Centre for Social Change, and the Governance Lab. She's a professor at Northeastern University and the force behind Innovate US, an initiative that has pioneered AI literacy and skills training for hundreds of thousands of public sector workers.

Most recently, she's the author of the vital new book, Reboot: AI and the Race to Save Democracy. What sets Beth apart in her work and why her voice is so crucial for us in this room is that she bypasses the standard techno-utopian hype, the doom and gloom narratives. Instead, she anchors her insights in very practical operational reality of the public service and the public sector. Her work asks a really important question: How can governments use artificial intelligence to listen better, to deliver smarter, and to strengthen state capacity?

For a Canadian public service that's focused on reinforcing tech governance, transforming service delivery, and protecting democratic institutions, this playbook couldn't be more timely. She reminds us that democracy is not a spectator sport, and its survival depends on how effectively we can innovate from within as an institution. So, please join me in giving a very warm Canadian public service welcome to Beth Novak.

[00:02:18 Vanessa gives the stage over to Beth Noveck.]

Vanessa Vermette: The floor is yours, Beth.

Beth Simone Noveck: Thank you very much and thank you for coming in from outside on such a beautiful day. I feel slightly guilty, but we'll try to make this quick and energizing and get you out of here into the sunshine is my hope.

[00:02:38 Overlaid text on screen: Beth Simone Noveck, Professor & Director, Northeastern University, Author; Lead, The Governance Lab; InnovateUS]

Beth Simone Noveck: So, it is, as you point out, a very, very timely conversation with what's happening in Canada, and I just want to thank, of course, the Canada School of Government, which has been such an incredible inspiration in my own work.

[00:03:05 Slide: An image of Beth's book, REBOOT: AI and the Race to Save Democracy, by Beth Simone Noveck.]

Beth Simone Noveck: The Innovate US project you mentioned, and I just want to say this before I get started, which provides peer-to-peer learning, free peer-to-peer learning for public servants in the US, which we don't have because we don't have a Canada School of Government, has been entirely inspired from, copied from, plagiarized from what you've done in Canada, especially some of the new modalities of learning that you've developed using new technology have been an inspiration, and I've written about a lot in my last book, about public sector learning.

[00:03:25 Split screen: Beth Noveck and slide, as described.]

Beth Simone Noveck: But here today, we're going to talk about something else, where learning is very central, but is the thing that is front of mind for all of us, which is the twin crises, I think, that we're facing, one around our democracy and the other around the advent and rise of AI.

It will come as no surprise that whereas we enjoy the pleasure of democracy here, that is not a widely shared phenomenon. By some estimates, the Swedish V-Dem Institute calculates only 29 liberal democracies left in the world.

[00:04:02 Split screen: Beth Noveck and slide.

Text on slide: The Crises of Democracy.]

Beth Simone Noveck: 3/4 of the world's population living in autocracy, and that is a declining number.

[00:04:10 Split screen: Beth Noveck and slide.

Text on slide: "We have frequently printed the word Democracy. Yet I cannot too often repeat that it is a word the real gist of which still sleeps, quite unawaken'd.

It is a great word, whose history, I suppose, remains inwritten, because that history is yet to be enacted." – Walt Whitman, 1871.]

Beth Simone Noveck: So, this crisis of democracy and what that means, of course, is a big term that we would need to spend a lot of time unpacking. Over 100 years ago, Walt Whitman already said, that democracy is not a very well-defined term. It "sleeps quite unawaken'd", this term. It is something that history has yet to sort of define.

[00:04:34 Split screen: Beth Noveck and slide. The slide shows a drawing of a multi-headed snake, titled Democracy. Each head is labelled, as follows:

Institutional Effectiveness: dysfunction, gridlock, and the inability to solve problems;

Representation: exclusion, gerrymandering, and the failure to reflect the people;

Elections: subversion, disinformation, and voter suppression;

Truth: disinformation, propaganda, and the erosion of shared facts;

Civility: polarization, hate, and the breakdown of respect;

Participation: apathy, barriers, and the disenfranchisement of citizens.]

Beth Simone Noveck: So, I do want to call out the fact that it is a many-headed hydra. It has a lot of different dimensions to it. And I'll say a little bit about the crisis, but I want to spend most of our time on what those solutions might look like, what the positive story is of how we could use AI to respond. But I did want to call out that we have this question, this phenomenon. For many people, democracy is, of course, about what happens in the voting booth, whereas for other people, it's an issue of civility, of truthfulness, of how we talk to one another.

[00:05:10 Split screen: Beth Noveck and slide. The slide shows a variety of news headlines about the dangers of AI, as described.]

Beth Simone Noveck: But it is, I think, for most of us who are thinking about and working in government, it is a question of institutional capacity, our ability to solve the increasingly complex problems that we're facing.

On top of, in parallel to this phenomenon, the crisis of democracy that we're facing, of course, along comes AI. And it is giving us a lot of pause. And you will see a lot of headlines like these about the robot apocalypse; the end of humanity; artificial general intelligence taking over the world; et cetera, et cetera. And it is again giving us a lot of pause.

[00:05:44 Split screen: Beth Noveck and slide. Text on slide: The Crises of Elections.]

Beth Simone Noveck: So, let's talk for a minute just about these crises. I want to bring you a little bit into the depths of despair before lunch, only to hopefully bring you back up before we leave today, although we'll see where we end up in the Q&A. Vanessa, let's hopefully end on a positive note, or at least a hungry note to get people off on their way today.

[00:06:14 Split screen: Beth Noveck and slide. The slide shows a variety of news clips about the prevalence of deep fakes in the 2025 Canadian election.]

Beth Simone Noveck: So, the electoral crisis, and especially as it interacts with AI, we are quite familiar with. The deepfakes, disinformation, misinformation debate is very, very well understood. You saw lots of examples of the use of deepfakes, of fake videos, of fake audio, both in recent Canadian elections, also in US elections. You remember the story about the Biden robocall, the fake Biden voice calling and telling everybody not to go out and vote.

[00:06:43 Split screen: Beth Noveck and slide, as described.]

Beth Simone Noveck: We see examples of this from all over the world, most recently in the Hungarian election. There is undoubtedly a proliferation of disinformation and misinformation, and I don't want to make light of this phenomenon. It's why we've seen extensive regulation. It's probably the area of AI where we have seen the most widespread regulatory response really globally.

[00:07:08 Split screen: Beth Noveck and slide. Text on slide:

These are the 3 biggest emerging risks the worlds is facing;

"Misinformation and disinformation is the most severe short-term risk the world faces." – World Economic Forum.]

Beth Simone Noveck: Because there is this concern about, especially around elections, the fact that the proliferation of disinformation and misinformation are really soiling and polluting our political discourse.

[00:07:22 Split screen: Beth Noveck and slide. Text on slide:

The Crises of Participation.]

Beth Simone Noveck: Now there isn't yet, I might add, conclusive evidence that disinformation, misinformation, deepfakes, the fake video of the candidate caught doing XYZ or saying XYZ, that it has impacted any elections, that it has changed the outcome of any elections, but we are surely very worried about both the integrity of our elections and more importantly the perception of the integrity of our elections. And this is no more so the case than in the United States. As you know right now, where the topic of electoral integrity persists even since the 19 – sorry, 1920, 2020, feels like 100 years ago, it's only 5 years – the 2020 electoral cycle and the so-called Big Lie.

Part of this phenomenon though, this democratic crisis, is also one of engagement and of participation. We live in democracies.

[00:08:26 Split screen: Beth Noveck and slide showing two graphs titled:

Few people see opportunities to participate in policymaking.

Few feel heard by policymakers.]

Beth Simone Noveck: Democracy is supposed to be about self-governance and self-rule, our ability to participate in making the decisions that most affect us. Globally, no matter what survey you look at or ask – and I know this is a little small here – most people feel that they have no say, no sway, and no influence over the political decisions that affect their lives.

[00:08:39 Split screen: Beth Noveck and slide, as described above.]

Beth Simone Noveck: They do not have a voice in governance and, by the same token, those people with a great deal of money, influence, and more lobbyists have more influence.

[00:09:00 Split screen: Beth Noveck and slide. Text on slide: Silicon Samples]

Beth Simone Noveck: The concern is that AI is only going to make this worse. That the proliferation of what sometimes are called silicone samples, that is to say, fake people, AI people, AI personas, will increasingly become our way of interacting.

[00:09:23 Split screen: Beth Noveck and slide, as described.]

Beth Simone Noveck: So, just last week we put out a little newsletter about AI and democracy. We put this up on Thursday. This is a new tool called Habermolt, named after Jürgen Habermas, and it proposes the fantastic idea that instead of having actual humans debating and deliberating with one another, we can just have AI debate and deliberate with itself. What do you need actual humans for? We can all just take a break from democracy and let the AI do its own work.

[00:09:51 Split screen: Beth Noveck and slide. Text on slide:

The Crises of Institutional Effectiveness]

Beth Simone Noveck: I'm slightly making fun, but it's a serious concern. It's one thing when a company goes out and says, we're going to test our product with an AI sample to say, "Do you like Coke or Pepsi better, or this toothpaste or that toothpaste?" It's very concerning when you think about governments thinking, "Hmm, we're going to save the work of talking to actual humans, humans who yell at you, and instead we're going to go out and just do focus groups with AI." But back to this bigger concern about institutional effectiveness, which for me is the real serious root of our questions of democratic crisis.

[00:10:30 Split screen: Beth Noveck and a series of images, as described.]

Beth Simone Noveck: We have increasingly complex and difficult challenges from wildfires, to flooding, pollution, pandemics, and we are not doing as well as we need to, and the citizenry does not feel we are doing as well as we ought to in solving these very complex challenges.

[00:10:42 Split screen: Beth Noveck and images of books:

Re-coding America, by Jennifer Pahkla;

Why Governments Fail so Often, by Peter H. Schuck;

A Cascade of Failures: Why Government Fails, and How to Stop It, by Paul C. Light;

Beyond Politics, by Randy T. Simmons; Results, by Charlie Baker & Steve Kadish.

Additional text on screen:

Declining Institutional Capacity: The Executive is doing less with less.

Agencies cannot deliver services efficiently enough or cheaply enough. We are failing to give people the benefits to which they are entitled.]

Beth Simone Noveck: And that's why you see a whole lot of books that are written about the topic of declining institutional effectiveness, declining institutional capacity. There is a concern that no matter how well we do, the world gets more and more complex, and we have to do a better job of responding to increasingly complex challenges with fewer and fewer resources.

[00:11:08 Split screen: Beth Noveck and slide. Text on slide:

We can upgrade our Democracy. Will we?]

Beth Simone Noveck: So, the question is, what are we going to do about all of this? And again, AI making it potentially more complicated and harder, not easier, I want to make the case – and the attempt in this book is to make the case – that we could use AI to respond to these challenges if we choose to do so.

[00:11:36 Split screen: Beth Noveck and slide. Text on slide:

Why AI matters more than just a calculator.

AI synthesizes; AI talks; AI is everywhere; AI is for all of us.]

Beth Simone Noveck: And let me be very clear at the outset, AI by itself is a tool. AI by itself is not going to do anything. We have nuclear technology, and we use it to build bombs, and we use it to power our electrical grid. The choice ultimately is up to us. The notion that AI somehow displaces workers is a distraction from the fact that it is humans, it is managers and CEOs and humans making the decision to fire workers as a result of AI, so it is going to be very much up to us to decide where we make our investments, what actions we take.

Again, I don't want to discount the particular challenges of this technology, which we'll get into, but we have a powerful set of tools if we point them in the right direction. And I think what makes them incredibly powerful is the fact that: number one, you do not need to be a technologist to use them. You do not need a computer science degree to use them. They are driven by plain language. Number two, they're increasingly ubiquitous. And number three, they're incredibly good at processing huge amounts of information.

[00:12:52 Split screen: Beth Noveck and slide. Text on slide:

From Scarcity to Abundance]

Beth Simone Noveck: And, in government in particular, we are living in a world in which we are living in institutions previously designed for a world of scarce information, where we put experts at the centre of our government who would monopolize, who would have the best knowledge that we needed to solve hard problems, to a world in which we now have a democratization of information and information abundance that, frankly, is ill-suited to the way our institutions are designed, but [an] incredibly useful opportunity for AI.

[00:13:23 Split screen: Beth Noveck and two slides. Text on slides: Tetizador Voz: Leticia Voice for Visually Impaired Brazilian Voters; Responding to the Crises of Elections: Elections are one of the most information-intense thing democracies do. In a compressed window of time.]

Beth Simone Noveck: So, we come back to this crisis of elections, an area in which information and how we process information is incredibly central. How we get out information to people who are going to run for office; to vote for office; how we manage the information of counting the votes. This is an information processing problem. And there are some really exciting uses of AI.

For example, in Brazil, they are choosing to use AI at the ballot box in order to translate and speak your ballot so that if you are visually impaired, your ballot will talk to you, not just be something that you have to read. In India, where they're using AI to translate your ballot into 22 languages to create opportunities for access.

[00:14:17 Split screen: Beth Noveck and slide titled: Elections should be free, fair, and frequent. They are often none of these. Additional text as described.]

Beth Simone Noveck: So, there is a wide range – and sorry, that is extremely small print small, so small I don't even know what I've said up here but I'm going to tell you, I'm going to make some stuff up, and maybe you can read it. Here we go. Some more examples, and I'll switch to the French in a moment.

So, Brazil over here, also New York, the Ballotpedia example out of California. Again, you may be familiar with the fact that California tends to have these referenda and give you this 150-page information booklet, which can be incredibly overwhelming and difficult for citizens to understand, let alone read, so the opportunity to use AI again to synthesize that information, to shorten it, to make it more accessible.

ERIC, the compact among 25 states now uses AI to help comb through the voter rolls to ensure that people who move between two jurisdictions are not lost and have the opportunity then to get added back into the voter rolls. ERIC is very specifically designed not to disenfranchise voters, but to re-enfranchise, to ensure the maximum right to vote. And they use AI to do this work of sifting through databases of information to be able to better manage the electoral process.

[00:15:39 Split screen: Beth Noveck and slide titled: Making elections accessible, understandable, and equitable. Additional text as described.]

Beth Simone Noveck: You see examples now, or the opportunity for examples, of using AI to create better training for poll workers; to create better training for voters; to create the kind of voting information systems common in Europe where you can get, again, answer a set of questions about your own preferences and get information about which candidates are best matched to your preferences. Again, using AI to speed up that process.

Meedan, a nonprofit that operates across 53 countries, uses AI. They have built open-source AI to spot misinformation campaigns and fight against them. And then they use AI together with news organizations to essentially not only spot patterns of misinformation and disinformation, but to write and circulate campaigns via social media, so to send out information via WhatsApp. If you're on a WhatsApp group with your family and they're all talking about this candidate or that candidate and spreading rumours, as people do on Facebook and WhatsApp, they are helping to inject truthfulness into those conversations in partnership with national electoral organizations and media organizations.

[00:16:56 Split screen: Beth Noveck and slide. The slide has an image of a robot introducing itself as Gov Virtual Assistant. It is assisting people with questions about voting including where the closest polling station is; if they are eligible to vote; and voting procedures and upcoming dates.]

Beth Simone Noveck: So, you can imagine lots of applications in the voting space, but again, we tend to be only having the defensive conversation about deepfakes and misinformation and, in my view, not enough conversation about "What is our proactive agenda, our positive agenda around making elections free, fair, and frequent?"

[00:17:14 Split screen: Beth Noveck and slide. The slide has images of "Joca" and his owner. Text on screen Joca's Law.]

Beth Simone Noveck: Now let me shift for a moment to the crisis of participation and engagement. This is Joca. Joca was, I'm sad to say, a golden retriever living in Brazil. This is his owner. His owner had to move for work from one side of Brazil to the other. Joca was checked in under the hold in the airplane. You can fill in the rest about what happens, but it doesn't go well for Joca. He doesn't end up at his destination. He's left on the tarmac, and he dies. So, you can imagine a newspaper headline lamenting what happened to this poor dog, people screaming on social media about this terrible airline and what they did.

[00:18:04 Split screen: Beth Noveck and slide, as described.]

Beth Simone Noveck: Well, Joca's owner is able to go onto the eCitizenship platform, the eCidadania platform of the Brazilian Senate, and write a petition for a new law around animal transit and animal transport by Brazilian airlines. Fast forward, thousands of people sign on to this, and new legislation is actually enacted. In Brazil, they have not one, but four different ways that the Senate takes public input, not just in formulating legislation, but almost all the questions that senators ask of witnesses in committees, those are written by the public.

[00:18:50 Split screen: Beth Noveck and an article from the UK Government titled "When everyone talks, no one listens." Additional text includes: Across the 500 consultations the government runs annually, the tool could help save officials from around 75,000 days of analysis every year, which costs the government £20 million in staffing costs.]

Beth Simone Noveck: The eCitizenship team, if you're thinking about a large country like Brazil and how this poor, beleaguered group of civil servants that's about 15 people manages this process, the answer is they use AI to help them sort out questions that are relevant for a given hearing; to match a question to what the topic is of the hearing; to take proposed legislation and help to redraft it in a form that the Senate can actually introduce. They even have a system whereby schoolchildren can participate, so it becomes an educational exercise in civics.

The exciting opportunity here, I would say, is that whereas the web has given us 30 years where we can talk, 30 years where anyone can express themselves – hooray – on YouTube, on TikTok, on fill-in-the-blank platform, it has made it increasingly difficult for institutions to listen. That's why the UK government did this study where they calculated that the amount of money that it costs them to analyze, to make sense of the citizen engagement that they do.

I was in Singapore not too long ago. They were at that point experimenting with doing little 50-person citizen juries. We will have 50 people deliberate on an issue. And they set up a whole operation to do this, and then did it 3 times and then stopped. Why? Because even the input from 50 people writing that up, summarizing that, making sense of it, translating that into policy was just way too much time and effort.

[00:20:39 Split screen: Beth Noveck and slide titled "Responding to the Crises of Participation", as described.]

Beth Simone Noveck: And so, that's why the UK government now has, through its AI.gov effort, this consultation analyzer platform that it's developed to enable it to do more and more public engagement. It's not limited to online work. In Mexico, the Supreme Court ran a set of public engagements about the future of its criminal justice system. Very high degree of distrust, very great degree of unhappiness, so they brought 4,000 citizens together across 14 different national conversations with victims of crime, the families of those who had committed crime, prosecutors, defence attorneys, all these people talking in person and they recorded it and, again, used AI to help them rapidly make sense of it and begin to translate that input into better outcomes and outputs for government.

[00:21:25 Split screen: Beth Noveck and previous slide titled "Responding to the Crises of Participation" with images of government home pages from California and Bogotá.]

Beth Simone Noveck: So, I am excited about these new experiments with uses of AI to help us, as institutional players, be able to listen. In California, the Engage California's platform, created after the wildfires went out, asked citizens, residents of LA, "How could we do better at responding to wildfires?" It worked so well that they said, "Oh, now we're going to go out and use it on public servants." And they just ran a big engagement with thousands of public servants about improving services. They have another one running now on another topic.

In Bogotá, they did this kind of work via WhatsApp. Getting 50,000 responses from people in a matter of 2 weeks at, again, a very, very low cost to government. And decreasing cost, allowing government not just to view citizen engagement as a nice-to-have, once-off single experiment, but as something that you can repeatedly do.

[00:22:14 Split screen: Beth Noveck and slide. Text on slide: Finland: New Ways of Talking.]

Beth Simone Noveck: In places like Finland, they are using open-source technology to not just allow people to respond via text, the typical sort of citizen engagement or citizen comment process, but to actually use a visual image generation we can do with AI now, to allow people to participate in urban planning in new ways. So, you can say, "I want the park bench here," or "I want a bike lane there," and it allows the public to visualize their street or their park in a way they couldn't have done before.

[00:22:58 Split screen: Beth Noveck and slide. Text on slide: Hamburg, Germany: Improving Listening.]

Beth Simone Noveck: 2258In Hamburg, they've spent the last 3 years building a toolkit – not just to allow people to give input, they've had that for a long time – called DIPAS, which is their citizen engagement system. Sorry, it auto-translates into English here. I'm hoping the French is – Oh, I don't have the French, sorry, of their website here. Thinking digital participation further. It's a very funny – it sounds better in German, let's put it that way. But the further in this is that now they've built this toolkit which allows them to analyze what citizens are saying. The reason I wanted to show this to you is because now there isn't a single urban planning or environmental decision that is made without citizen engagement. And because the tools are free and open source, 9 cities now use this platform.

[00:23:56 Split screen: Beth Noveck and slide. Text on slide: Responding to the Crises of Capacity, as described.]

Beth Simone Noveck: I have been in an endless discussion trying to bring this to the city of Boston, which is one of my next projects so that we can get more people using this backend toolkit that they've invested 3 years in developing that is designed for public sector by public sector to make sense of citizen comment.

In the work that I used to do in New Jersey, where I was the head of AI for a number of years and before that the chief innovation officer, we used AI to troll through our databases to find 693,000 children, whom we had lost, who were entitled to a free school lunch, a free breakfast, a food benefit. But because the kid is Jack in this database and John in that database; they've moved in foster care between this jurisdiction and that one, we had simply lost a lot of people who were entitled to a benefit that we were then able to give out.

[00:23:56 Split screen: Beth Noveck and slide. Text on slide: AI and the Crises of Capacity; What good AI in government really looks like.]

Beth Simone Noveck: One last example in this capacity space here, which is some work that my students are doing that I'm really excited about. This is a project called GrantWell. GrantWell is a tool that we built for the state of Massachusetts to help smaller communities find, write, [and] apply for state and federal grant money.

So, this is something that every community has to do. If you're in a big city, fairly easy. You have a team of grant writers. If you're in a small town or you're a civic group and you want to replace your old school buses with energy-efficient vehicles, you want to take the lead out of your pipes, for which there is public money, you do not necessarily have the resources to know how to navigate that process.

[00:25:45 Split screen: Beth Noveck and slide, as described.]

Beth Simone Noveck: So using AI, they've built a tool custom-built around the public grant making process, but what's more exciting about that is that now the same tool that we built for Massachusetts, we are giving away for free to 4 other states. Sitting down with those states to co-design version 2 to improve the tool and do things like integrate relevant sources of data so communities have access to that when writing their grant application.

There are many, many good stories of this. I see I have yet another one. This one is from Taoyuan, Taiwan, second largest city in Taiwan. They've stuck a sensor on literally everything so that they're throwing off data and then using AI to optimize the delivery of services from lung cancer screening, to speeding up the time it takes an ambulance to pick up a patient and get to the hospital, cutting that time in half and saving lives simply by using AI to analyze large quantities of data.

[00:26:42 Split screen: Beth Noveck and slide. Text on slide: Yes, but…]

Beth Simone Noveck: Okay, let me get towards the end here. So, great. I've shown you a million examples of all the things we could be doing that are so great, but are we actually going to do this?

[00:26:57 Split screen: Beth Noveck and slide. Text on slide: Corporate AI in the Democratic Hen House.]

Beth Simone Noveck: Let's talk about some of the challenges before we wrap up with how we actually realize this opportunity. Challenge number one that we all are worried about is, "Yes, but you want me to do all these great civic things with a bunch of tools run by Elon Musk and Mark Zuckerberg and".. fill in the blank. You know that there are debates now. Europe is very busy talking about AI sovereignty equals don't buy American technology. We have lots of discussion about the concerns, whether it's American or something else, about putting purely commercial tools into public sector context like schools and governments.

[00:27:44 Split screen: Beth Noveck and slide. Text on slide: AI Slop Will Drown Out Truth, Civility, and Deliberation.]

Beth Simone Noveck: Do we have to worry about platforms that are designed to maximize profit over public purpose? We're seeing, of course, lots of ways in which these tools are being misused, not for just disinformation and misinformation, but essentially just for wasteful and silly purposes.

[00:28:04 Split screen: Beth Noveck and a slide showing two images. One is of a muscle-bound baby; the other is of a dog serving water in a restaurant.]

Beth Simone Noveck: I just wanted to be able to show you these pictures, but there's a lot of stupid AI slop out there.

[00:28:04 Split screen: Beth Noveck and a slide titled "The Era of AI Slop-aganda" showing images of Donald Trump depicted as Jesus, the Pope, and a muscle-bound Jedi.]

Beth Simone Noveck: And it's coming from – this is the White House's own account. We have a government that is – this is not a political comment. This is a factual comment – we love to make this AI so-called slop, these pictures.

[00:28:32 Split screen: Beth Noveck and a slide showing more "AI Slop" involving Elon Musk and DOGE.]

Beth Simone Noveck: And we have a lot of concerns, obviously, about how people are going to use these tools for things other than the laudable and noble purposes that I'm talking about. We know that we can use AI in government and data in government to make government more efficient. But we also know that we have seen a lot of those uses recently in ways that I would argue are not very helpful.

[00:28:57 Split screen: Beth Noveck and a slide: a DOGE document titled "Updated data on DEI related contract cancellations with full detail; Protesters holding signs saying "Funding Saves Lives"; a quote from Elon Musk: "USAID is a criminal organization. Time for it to die."]

Beth Simone Noveck: For example, it was announced last week we're going to wholesale – this policy is not so new – but we are wholesaling, going through every government grant looking for words that offend and then using that to cut grants.

[00:29:15 Split screen: Beth Noveck and a slide, as described.]

Beth Simone Noveck: Or this story from yesterday or 2 days ago. We're just wholesale going through and finding any contract that the Forest Service is using. I just heard this on the news, I guess a day or two ago. And we're just cutting all of those contracts without looking carefully at what those are for. So, the story was one about how it's going to cost the government so much more money, so much more taxpayer dollars because we're essentially just using AI to just cut grants, cut contracts, cut workers without really careful regard for how we're doing this.

[00:29:49 Split screen: Beth Noveck and slide. Text on slide: From Crisis to Renewal: Democratic AI.]

Beth Simone Noveck: Okay, now here's where we wrap up with how we actually try to do the right thing, as opposed to doing the wrong thing. And I think there are, again, lots of wrong uses as we think about how we're using AI.

[00:30:07 Split screen: Beth Noveck and a slide titled "What is Democratic AI?", as described.]

Beth Simone Noveck: So, this is why I use the term, and why I wrote the book to talk about the idea of Democratic AI. And here I mean democratic with a small d, as in democracy, not as in Democrats versus Republicans. This is, again, about what I still think is a good word that we need to rescue. And I think it's important to have the word Democratic AI to talk about AI, and uses of AI, that are centred around public purpose; that are centred around participatory ways of working; and AI that we are developing with and for the people who are most affected; and where we are thinking about the democratic outcomes. That is to say, the benefits for the public rather than the alternative.

[00:31:05 Split screen: Beth Noveck and an image of a book "If Anyone Builds It, Everyone Dies" by Eliezer Yudkowsky & Nate Soares.]

Beth Simone Noveck: We need a term because – oops, let me see if I have a picture here. Nope. Yes, that's the picture I wanted – because this is what a lot of the headlines are about. And I mentioned this at the outset, but I want to come back to the fact that we know if it bleeds, it leads. That's what sells papers; it's what creates engagement. This is a book that came out a couple of months ago. This got a huge amount of press.

But it takes up a huge amount of oxygen when this is the only thing that we're talking about.

If we only talk about the robot apocalypse, if we only then talk about our defensive posture, we're not talking about what does it mean to do things well and to do things right. And it's not, frankly, a conversation we're having enough of.

I was at an event just last week in Mexico together with a foundation called the Scott-Morgan

Foundation I had not heard of before. They came to show off the work that they were doing, using AI, to help victims of ALS regain their voice. So, with a few minutes of a recorded voice from what the person sounded like before they had ALS, they're able to recreate and give that person a voice.

They put on a panel at this event entirely of people with ALS who can barely articulate words. You and I would not be able to communicate. They can understand us perfectly fine, but we can't understand them. With AI, it was able to synthesize and give them back a voice so they could speak publicly at this event that I attended. It's those stories, it's the things that we could be doing right that I think get drowned out when we're only talking about the robot apocalypse.

[00:32:44 Split screen: Beth Noveck and a slide titled "How do we get there?", as described.]

Beth Simone Noveck: And frankly, even if we're only talking about AI for good, which is a topic people like, we want to talk about, again, medical uses and climate uses and other things, but we need to be talking fundamentally about this crisis of our democracy and what we can do. So, what can we do and what do we need to do?

[00:33:03 Split screen: Beth Noveck and a series of slides titled "How do we get there?: Public AI", as described.]

Beth Simone Noveck: Four parts to this. The first is the AI that we use. There is now a robust but very young conversation around the topic of what some people call public AI. So, public in the same way that we think about public parks or public libraries. We have bookstores, but we also have libraries. Nobody's talking about getting rid of bookstores. We love bookstores. But we also want to have public alternatives and options that exist alongside commercial alternatives.

So, there are explorations now of what is the regulation we need, what is the investment we need, to ensure that our AI is affordable, to ensure that it's explainable, and again, that it is built for public use and public purpose, that it is not just built for commercial uses. We want the cool tools for making the stupid cat pictures or the AI slop. I love making those images. I should have put in [that] we have a pastime of making AI pictures of my cat in my house and thereby potentially polluting the planet.

[00:34:15 Split screen: Beth Noveck and a series of images of eLearning homepages, and articles about various global governments invested in Public AI, as described.]

Beth Simone Noveck: But when it comes to government use in particular, we want to be thinking about what are the tools we're putting into and using, especially for public purpose. So, there are governments, like Sweden, like Estonia, of course. Everybody hates it when you mention Estonia. Nobody wants to be told you should be more like Estonia.

But last week, France announced they were going to put over $100 billion into investments in public AI. Again, part of that is to develop a commercial marketplace, but part of it is also to develop tools for government in the way that they are thinking about doing in these places. And again, it's still early. Spain has invested a huge amount in public compute, but the government is still using Microsoft.

Now, is that a terrible thing? We've been using Microsoft products for a long time. It's a conversation we need to be having. They're developing their own Large Language Models. Should the government in its chatbots, in its services, are there additional products and tools that can be built for and with the public in ways that offer more democratic control? And that's a conversation, again, we need to have.

Sweden has said, "Look, we're going to do both. The chatbot we're going to use in government is going to use our government-developed, native LLM built on Swedish data from our National Archives as a public-private partnership."

[00:35:48 Split screen: Beth Noveck and a series of slides titled "How do we get there?: Upskilling", as described.]

Beth Simone Noveck: So, again, early days in this policy space, but looking at government as the largest purchaser of tech, what can we be doing differently? We need to be building AI with those who are most affected. Community-centred; participatory; human-centred. How many events have been had from this stage about human-centred design? I am sure many, many, many, many. Those lessons don't get lost when we're talking about – sorry, let me see what picture is coming. Oops, go back – when we're talking about AI.

So, the picture I'm looking for, which maybe is later. Quick example, again, team of AI fellows. I call them my students, but they're not students. They're fellows who take a sabbatical to build open-source AI with government partners. Building a tool to help families of kids with disabilities understand and advocate for their rights. Tooling built entirely with, for, and by the families who are most affected. Families, I might add, who are low literacy and whose native language is not English.

[00:36:59 Split screen: Beth Noveck shuffles through a series of slides briefly and comes back to "How do we get there?: Upskilling", as described.]

Beth Simone Noveck: So, last piece. Sorry, let me come back here because I think I missed a slide. It's here out of order, so I'll come back here.

Third point. The point about training and learning. And this is Coles to Newcastle at the Canada School of Government. But really thinking about how we spread AI for all, not just AI for some, to use the Canadian phrase, not just for the public, but for the public sector. It's only by learning what these tools do, what they do well, and what they don't do well, that we're going to be able to make decisions about how we want to use them.

[00:37:40 Split screen: Beth Noveck and a series of slides, as described.]

Beth Simone Noveck: So, to show you these now for the third time, this is the website for Innovate US, We also have Innovate EU, which is launching today in Brussels. But we set this up in order to promote peer-to-peer learning around public sector uses, not originally of AI, but now AI has eaten everything that we do because it's what everybody wants to learn about. And so, we are making the very strong case that investments in learning for public sector, investments in time spent learning about AI is going to lead to changes in behaviour among public sector, which will ultimately lead to better outcomes for residents.

[00:38:22 Split screen: Beth Noveck and a series of slides, as described.]

Beth Simone Noveck: So, we are, for example, developing and launching in 2 weeks a new program on how to use AI for public engagement: telling that story from Hamburg; telling that story from Bogotá. There are examples and advisors from 42 countries involved in developing this program in the hope that we can actually change the way government works; that we can actually get more people, [inaudible,] to practice more and more public engagement in the work that they're doing. Not just thinking about using AI, again, to be faster, to be more efficient, but to really change the way that we do things.

[00:39:06 Split screen: Beth Noveck and a series of slides depicting warnings against AI, as described.]

Beth Simone Noveck: There are lots of warnings, of course, about doing AI wrong. Like the notion that we're going to just cede all of our authority and decision-making power to an algorithm; that we're going to just let an algorithm make decisions about who gets a benefit and who doesn't. That's what's led to protests in places like the Netherlands. You might have seen the Netflix movie about the post office story in the UK, "Mr. Bates versus the Post Office", good movie if you haven't seen it, about ceding government authority to some technology.

The learning is incredibly important if we are not going to fall victim to that. We are going to do it wrong if we don't learn. And more importantly, we need the learning to understand even where we get what these tools are, how we're going to incorporate them into our work.

[00:39:58 Split screen: Beth Noveck and slide, as described.]

Beth Simone Noveck: So, this is an example from a hospital system. The head of AI at Kaiser Permanente in California was telling me they developed this new algorithm that predicts when patients are likely to regress in the hospital, to give you the red flag warning signs of patients that are at risk. The issue in this case, there's a lot of bad predictive AI. The algorithm actually apparently is pretty good and pretty effective. That wasn't the issue. Where they were struggling with is, "What do we do with that algorithm now that we have it?

[00:40:33 Split screen: Beth Noveck and slide titled "How do we get there?: Public Participation", as described.]

Beth Simone Noveck: Do we give it to nurses? Do we give it to doctors? What do we do with this to incorporate it into our work?" Oh, here's all my out-of-order slides. Sorry. I don't know what I did. Sorry. Let me go back here.

[00:40:58 Split screen: Beth Noveck shuffles through slides, pauses on one, as described, and resumes with a slide titled "How do we get there?: Research", as described.]

Beth Simone Noveck: Here's the picture of the cool tool we built for the families with disabilities. It has very pretty graphic design, so I will show you that.

Last two points here. That means if we're going to realize these benefits, we have to put front and centre the kinds of questions that we are studying, the kinds of questions that we're asking. I would say our universities are not yet doing this right, and I spend a lot of my time now talking to universities trying to get them interested in studying questions of public sector, and to ask questions of how we can use AI to incorporate – come back to that algorithm comment – how we can start to incorporate AI effectively into our work in public sector; what should we be building; how we should be using it; what works and what doesn't work?

[00:41:39 Split screen: Beth Noveck and slide, as described.]

Beth Simone Noveck: We have a massive new global initiative to get companies together and universities together to ask the question about how we can use AI to respond to new challenges around cybersecurity. That was Project Glasswing that was announced a few weeks ago when Anthropic came out and said, "Our AI is so powerful it can defeat any cybersecurity system." Massive panic, big challenge. We have to get everybody together to start to address cybersecurity with AI.

[00:42:16 Split screen: Beth Noveck and slide titled "AI For Democracy".]

Beth Simone Noveck: Where is our massive panic around democracy and how we use AI to fix things? Where is our moonshot for AI, using it to address the challenges of our institutions and the challenges of our democracy? We don't have those career paths, those journals, those conferences. We don't have that focus yet, and my hope and conclusion is to put some of this on the agenda.

[00:42:37 Split screen: Beth Noveck and a slide showing a FIFA World Cup football.]

Beth Simone Noveck: The World Cup launches Thursday? This week. Shall I ask how many people are going to be sick, going to find themselves sick, have a cold coming on right now and will suddenly not be available?

[00:42:54 Split screen: Beth Noveck and a slide showing a FIFA World Cup official consulting VAR to make a call.]

Beth Simone Noveck: One of the things that is very controversial, as with the World Cup – you may have seen this, this is not new – is technology. Who knows what VAR stands for? I've already forgotten. Visual Assisted – where are my footie fans? –Replay. So now in football, again, this started about 10 years ago, we're going to use cameras at the lines and then use cameras plus off-pitch humans to review and make the call instead of the on-pitch ref. We've now integrated AI into this process, so that AI is going to be making the call about offsides this week. In a recent survey, 91% of football fans – 91% – agree that this is a terrible idea. They hate this, and they want us to go back to a world of human refs making the call.

So, I end on this note to say that I think the lessons of VAR are ones that we need to take to heart here. The problem with VAR, it was solving for a problem we didn't have. It was designing a system that did not involve the fans, did not involve the players, and imposed the use of a technology, again, for the benefit of commercial gain rather than the enjoyment of the sport. A sport which, for cultural; for social; for religious reasons, we put the ref at the centre of. We want humans to be at the centre of this all-important cultural phenomenon. I think VAR is a sort of reflection of how we need to think about AI for democracy.

[00:44:44 Split screen: Beth Noveck and slide, as described.]

Beth Simone Noveck: We need to be building these tools, as the Pope said, not as Jerusalem, not as Babel, not as the saviour of all things, or the end of all things, but we need to focus on this messy middle of the very practical things we can do with communities, for people, with people, together with public servants to address the challenges that we face.

Thank you.

[00:45:16 The audience applauds, Beth takes her seat, and Vanessa comes back on stage to join her.]

Vanessa Vermette: Thank you so much, Beth. You really flew through that, so there's a lot, I think, to chew on in the slides you just presented. But I want to start first by congratulating you on your book. That's really a huge achievement to get this out into the world at such an important time.

Beth Simone Noveck: You have no idea.

Vanessa Vermette: A lot of labour pains, I'm sure. But can I start by asking you what inspired you to write this book? Was there a particular moment, an interaction, something you saw where you said I have to synthesize this and put it out into the world?

Beth Simone Noveck: Well, I think there's two inspirations to this, one historical that is not unique to this book. I did my studies, my schooling, I actually studied history and was focused on the 1920s in Germany and Austria. So, I had the great fortune, which I recommend after university graduation, to go spend a year in the coffeehouses of Vienna researching what caused Austria's institutions to break and to succumb to fascism.

So, the original interest that pervades all of my work, and motivated this one as well, was this question of, "What [are] the right decisions for us to make?" There's such a panic about now what we're doing with social media and the ill effects it's had for kids, the way we're putting technology into schools, and we're having conversations about banning the use of phones in schools, for example.

We've all had the sense that we've done something wrong that has led to the internet not being this Housian space for free speech but instead being what Anne Applebaum calls "Democracy's dumpster fire", which is how a lot of us feel about it if you spend any time on social media now. So, the question is, "We're at the advent of the era of generative AI. How do we do it right?" was motivated for me very much by these historical questions.

And then fast forward, I have to say I owe a debt – in inverted commas, inverted quotes here – to Elon Musk. Yes, very inverted, because we saw what it means to use AI wrong. But the debt of gratitude we owe to Elon Musk is that it has never caused more attention to the topic of government capacity and government functioning.

When I worked for Obama, and helped to start the Open Government Initiative, we were passionate about what we did around open data and government modernization. I was told Obama needs to be in another continent before you are allowed to make an announcement. Because at that point in time, government capacity, government modernization, the functioning of government was considered the third rail of boredom.

They can't talk about government. Americans don't want to talk about government. Like, ooh, boring. Quite literally. I'm not making that line up when I tell you I was told he has to be in Asia. That's when we made our big open government announcements, when it was so sort of beneath the radar screen. It brought a lot of attention to these topics, not in a good way, but it has caused a lot of people now to ask, "How do we do things right?"

I don't have to tell you here how much rethinking of core assumptions, for example, the trade relationship between the US and Canada, has upended a generation-long – I just have to say one word to you, which is rupture – that was the term invented here and shared globally. So, it was that rupture that I think – never waste a good crisis. Might as well write a book about it.

Vanessa Vermette: That's a good approach. I want to ask about some of the examples that you gave in your talk about where AI is already improving the democratic process in certain locations. Is there one example that really stands out to you that you can take us a little bit deeper into? How people trusted the process, what really happened on the ground, what were the results?

Beth Simone Noveck: So I think, let me say, we're in early days right now where a lot of – and I think you've seen this here too – a lot of our early uses, precisely because we're very worried about the public not trusting AI, public servants not trusting AI, that a lot of early uses are very mundane, shall we say. They are internal uses to simplify some language; to take something in complex language and make it easier to understand; to do translation in some way; to synthesize some documents.

And I especially don't want to dismiss those because I think those are incredibly important. And they also help to decrease some of the concern and fear when we realize this is just a next generation of word processor. And my hope is that we all calm down in a few years and sort of go, "This is like Microsoft Word 2.0." where we go, "Oh yes, now I can upload a couple of documents and have it give me an answer."

So, in Massachusetts, for example, we had two AI fellows sit down with the Department of Transportation and take all of their policy documents and all of their training manuals and upload them to essentially create an internal chatbot to train new highway engineers. The idea was that a new highway engineer could get an answer sourced and cited based on their own internal documents and save the time of senior engineers for important questions instead of for just some basic, "Where do I find X or where do I find Y?"

We did the same thing around health insurance, Massachusetts again, and we piloted this fellowship model there, but there's a number of jurisdictions that are doing this. Massachusetts had the brilliant idea that people were waiting too long to get health insurance, so they created 6 call centres instead of 1. Well, as you may know, when you widen the road and add more lanes, you actually get more traffic, not less. The engineers here will explain to me why this is exactly, but it turns out it doesn't work very well, and the same thing happens with contact centres.

So, what we did was we essentially did something similar and created a tool for the people answering the phone. We also did this in New Jersey when I worked for New Jersey. We had a playbook that involved sitting down with the people who answer the phone, sitting down with the people who write the documents, again, in a very human-centred way, to understand what is their behaviour; what is their need; what do they need the tools to do, and then create tools to help them get answers so that when they answer the phone, A, the public is still talking to someone human, but B, they're getting an accurate answer faster, spending less time on the phone, which is better for everybody in terms of the answers they're getting.

At the risk of going on too long, I want to give you one more quick example: Public defender in the state of New Jersey. An office that represents the indigent in their civil legal cases. An office that, until April of this year, did not have Wi-Fi. That is how under-resourced they are.

What did they do, though? They took the free tool, NotebookLM, they uploaded 20 or 50 of their past legal briefs to NotebookLM to essentially create a sourced and cited chatbot. They didn't use that. What they did was use that as a demo to sit down and have lunch with their staff. And say, "What problem do we need to solve? What would actually help us to make our jobs easier?"

And then they went out and sat and partnered with Princeton University to build what they call their Brief Bank tool, which they now have rolled out, that helps them in their drafting process. Again, it doesn't write their briefs for them. It just draws from their past briefs to find sample language. In a legal brief, you need like, "Where is our standard paragraph on subject matter jurisdiction? Where is our standard paragraph on XYZ?" to essentially just pull from those and help them do their job better.

But the reason to go on at such length was simply to say that it was about the problem that you pick, and it was about the conversation that you have about how you're trying to work that I think are really the transformative and exciting uses. And none of them are that difficult.

Vanessa Vermette: Yes, it reminds me of the example you gave at the end about the VAR on the soccer pitch solving a problem that we didn't have. Do you see examples today across the world where governments are using AI to solve problems we don't have?

Beth Simone Noveck: I am so grateful that as the chief innovation – we had a CTO in New Jersey who was the guy in charge of buying tech. They originally asked me; do I want that job? And I was like, no, no, no, no, no, I do not want that job because then that job is getting pitched by all manner of people wanting to sell you a solution to a problem that you don't have.

And this problem, this phenomenon, that's not new to AI, but it's getting worse where everybody has an AI tool to do X, Y, and Z. Frankly, an AI tool, if you have Claude or ChatGPT or Copilot, it does most of these things. You do not need a specialized platform increasingly to do many of these things.

But the bad examples, let me say the dangerous examples we're seeing, are the ones about predictive AI. So, we should be clear that there's different kinds of AI. It's a set of tools for processing data. And we can process and find patterns in data that are retrospective, that are past looking. Lots of good uses of that, and those are fairly uncontroversial. The stuff which is future-looking, that is predictive, that is where we see the most risk and danger.

So, I gave you the example of the predictive AI in the healthcare context that was built by Kaiser Permanente, which seems to have pretty good results. There was a predictive tool built for addressing sepsis. Again, a serious problem, an important thing we want to be able to predict that was built by Epic, by the people who do the electronic healthcare record systems, but it was built badly. They used national data for a phenomenon that you really need to use localized data for, and the tool doesn't work very well. The predictions are off, and they're bad.

And so, you start to incorporate those kinds of bad tools, especially in contexts like depriving people of a benefit; accusing people of fraud; predicting when somebody is going to commit a crime; when they're going to reoffend. So, anything that is designed to deprive someone of liberty, or something especially that has those kinds of serious consequences, that's ones where we're seeing not just wasteful things, but actually really dangerous examples. And those will get worse if we do not educate ourselves around how not to buy and build stupid AI.

Vanessa Vermette: Words of wisdom right there. I want to touch on public AI. You said there's a burgeoning emergent conversation around public AI across the world. You mentioned the principles of democratic purpose, democratic process, and democratic outcomes. Maybe talk to us about some of the examples you've seen of jurisdictions that have built public AI and how it's being used. You mentioned Sweden had a public-private partnership model. Are there others? Or could you dive a little bit deeper into that?

Beth Simone Noveck: Yes. So, Sweden has a – so many places have – Sweden has a national research lab called AI Sweden, where they set out, largely for linguistic purposes, because as we know, generative AI is essentially just AI trained on large numbers of words and images from the internet. And because the internet has dominant languages, of which English is the most dominant, the places like Iceland and Sweden and Estonia have tried to build their own language models in order to have a tool that works better in their own languages.

Sweden said, "We're going to use training data from our national archives", until they got sued for doing so. Interestingly, not by the AI companies, but by content companies, saying, "If you cannot prove to us that there isn't copyrighted material in the training data, you can't do this." We have a whole series about public AI and a lot of people from these countries coming to talk, which are all recorded and available online. We'll be turning that into a short course available this fall. The poor guy who was in charge of this has now left for the private sector because he was so despairing over how hard it was to do this work. But, that said, they did roll out a native chatbot for use in government.

And what I'm excited about is that for doing things like a simple informational chatbot, you do not need the computational power of one of these enormous Large Language Models. So, for environmental reasons; for linguistic reasons; for public control reasons; the idea of a publicly developed chatbot – like what they're doing in Spain where they have built a model called ALIA, A-L-I-A, is their native publicly built model, which is a consortium of their public universities that have built their own, native Large Language Model – could be a tool that is done in a public-private way where there might be private investment.

Or there's lots of places, like a lot of the examples I've given you, that the underlying Large Language Model is a ChatGPT, or a Claude, or a Microsoft Copilot, or a Gemini, but it's an open-source tool that's built on top of it. So, I don't think the fact that we might not go out and build LLM Canada or LLM US – that is not going to happen, at least in the US – you're not going to see a federal government alternative to private companies. It's just a non-starter, politically, but it doesn't mean that on top of these we cannot build open-source tools for public purpose that we share.

So, come back to this grants tool, GrantWell. That was a tool, again, that's built on top of a commercial Large Language Model, but it was built for, and by, and with public sector in one state, through a human-centred process, with hundreds of grant writers and community activists participating in the design of the tools. The tools are free, and again, being given away now to other states.

It's not just an issue of who owns the underlying model and who owns the tech, but again, how do we define the problem that we're solving so that we're not doing the VAR problem, but also so that we're thinking over time about how to measure the effectiveness of the tool.

And that is the most important thing. We have so much discussion about what data goes into developing these things. For public sector, what we have to care about is what comes out of it. Does it do a good job of predicting sepsis? Does it do a good job of measuring the patient who is going to, whatever the medical word is for when you get worse, not do well in the hospital? There's a medical word for reoffending, but it's not that. I'm blanking on the word for when you're at risk in the hospital. What does it mean to write an effective grant application?

That is an ongoing process that we need to continue these public and open conversations around. So, sorry, that was too long an answer for something for which there isn't one way of doing things. I don't think it's only, "We must, as government, go out and build our own model."

We're trying to balance competing goods, which include supporting a burgeoning private sector and the job growth that that creates but still thinking about what's the tooling that we put in; how do we develop it; and more publicly manage and govern tools, at least at the top layer, even if the bottom layer is still a commercial model.

Vanessa Vermette: So, I'm going to go to the audience in just a moment. Get your questions ready for Beth. We have mics in the room.

Beth Simone Noveck: Comments, disagreements, challenges. Yell at me.

Vanessa Vermette: Whatever you like. There's mics here on both sides as well as members of our staff that can run a mic to you if you prefer not to get up. Before we go to the audience, though, I did want to ask you about the public consultation example specifically. And I wanted to ask about how we use the tool for this purpose in a way that makes sure that it's not just citizens or organizations that then have the best bots to engage in the process, that get the loudest voices? How can we make sure we hear all the voices in a true democratic fashion?

Beth Simone Noveck: So, short answer there is that because AI is bringing down the cost of making sense of public engagement, it can allow us to do more public engagement, because ideally what you want to do – the problem up until now is because public engagement has been so expensive in terms of money, time, effort, and often doesn't go well.

See, e.g., the town hall system. I don't know if you followed the news in the US, and you have town halls here as well, but the Republican Party issued a moratorium on town halls because their politicians were getting yelled at so much in these things that the party leadership said, "We don't want you doing these anymore because they're not going so well." I don't think the Democrats have done something similar, but it's not because either party does them at all, because they're so costly in all the senses of that word.

So, short story long, when we can bring down the cost, make it easier, more effective to do public engagement, we can do more of them. Which means we can combine a representative sample, the idea of a citizen jury or a mini public or a deliberative assembly, in other words, one of these sortition-based representative samples has advantages that you get all voices are included. And again, sortition, if I'm only doing it based on demographics, that's only one way of sorting things, I can then combine it with maybe my issue is about healthcare, then I don't care about representation from a demographic perspective. I want to make sure I'm hearing from doctors and nurses and healthcare professionals and patients and whatnot.

I can also go out and pick the people, and I can also have self-selected, which also has benefits. You hear from passionate people, people who have something to say, who have a commitment to the cause. I can only afford to do all 3 of these things when I bring down the cost. So, the exciting thing is when it gets cheaper, I then can think about public engagement as something I don't do once, but I do a lot of.

Vanessa Vermette: That's great. I see we have a question ready to go right here. Please go ahead.

Lindsay McRae: Thank you. My name is Lindsay McRae and thank you so much for being here. So, I have a 3-part question.

Beth Simone Noveck: If you're willing, tell us who you are. So, what part of government do you work for?

Lindsay McRae: I work for the Canada School. This is my first day.

Beth Simone Noveck: Oh wow, welcome!

Lindsay McRae: Thank you. Yes, this is my boss.

Beth Simone Noveck: Uh-oh, you already have the job, you don't have to ask anything now.

Lindsay McRae: I'm curious. So, you talked a lot about the use of AI and the different ways that it has positively impacted a plethora of countries. So, my first part of my first question is, how do we balance that use of AI in government with the potential for the government employees to become more dependent on those tools?

And then going from there, we also talked about how a lot of private companies own some of these tools unless you're creating and like reinventing the wheel, but then again, the base code might still be similar so, second question is, a lot of these systems are owned by private companies, and so how do you balance the privacy versus security versus efficiency aspects? And my third and final – sorry – is, do the citizens then have a right to know when AI is being used? Thank you.

Vanessa Vermette: You can choose to answer part of that question, a combination of parts of that question.

Beth Simone Noveck: I'll answer all three really badly. So, let me work backwards. So, when we did our first generation of AI policies, and New Jersey was the first state to have an AI policy, Boston was the first city. Of course, the first-gen policy said you must disclose when you use AI. The American Bar Association – this is on my mind; it came up last week – has a policy which says that lawyers have to disclose when they use AI.

Second-generation policy doesn't require this. Why? Because the world has moved on and now AI is in everything and everywhere. And how do I think about drawing that line? And you know when you sit down to write an email, it finishes your sentence for you, sometimes helpfully, sometimes unhelpfully.

So, deciding where those lines are, I think definitely when AI is involved in decision-making – which is why Canada has a policy around AI algorithmic inventorying. Why a lot of countries have this concept of when you have an algorithm, that is to say, we're using AI to make a decision – then we definitely want to have disclosure.

But AI use, that's a harder line to police. Because, again, Microsoft Word, Google Docs, Gmail, whatever, all these basic programs now are helpful or unhelpful. When I buy bananas, it now suggests to me whether I don't want to buy dish soap because it's using AI unhelpfully. And I'm like, "No, I really wanted bananas, thank you." So, AI's been under the hood for a long time, so I think it's thinking about AI in the decision-making context.

Number 2, ownership, privacy, security, efficiency. Education is really important here because it helps us to understand that number one, in many cases, we are contracting for our privacy. It is buying AI that gives us the right to turn off the use of AI as training data, for example. And it's why governments are negotiating these contracts.

But two things here. One is, if you go in the settings, even of the free tools, you can click a button that gives you more privacy. We know this is a phenomenon with all of our tech. It's not new to AI of we are giving away the store when it comes to our data and our privacy all over the place. And I would like to see us, as government, push back more on these companies to say, "Look, we're not going to pay for our privacy. You need to give us our privacy."

I'm concerned about – here's the part I'm going to say something now. You're going to cut this in the video because I'm going to make a statement that you will disapprove of – I am unhappy that last week, at the State AI Leaders meeting in the US, Sam Altman was invited to keynote the conference.

We should be as states – that was the meaning of State AI, I wasn't there which is why I can criticize it, ha ha ha – as states, we should be getting together to make demands of these companies. And they make plenty of money, and they make plenty of money off of taxpayers, and we should be saying, "No, privacy needs to be central to what you do." And that's not just limited to government use. It should be part of how we think about these things publicly. It's our taxpayer dollars that created the open data that train these things.

When I was in the federal government, and Google stepped up and said, "We will host all of the patent data." We said, "Oh, that's so generous of you. Thank you so much." We didn't realize at the time, and I'm glad we did it because it allowed us to make all patent data open and available, which we, as the government, couldn't do because it would have crashed all of our servers and broken the government. But the reason they volunteered to do it was because it gave them training data to build their translation models. We didn't realize at the time we were powering the beginnings of generative AI.

Last thing about dependency – sorry, I've gone on too long because it's a good question. One of the things I'm experimenting with now is we have some AI fellows building, in connection with one of the courses we're rolling out, a coaching tool. An AI-based tool that helps coach you through thinking about how you do public engagement. Just one example, it could be transit or healthcare or something else. The AI is trained not to tell you what to do. It's trained to ask you questions so that you come up with answers for yourself.

But it's really hard, and that's a design decision to say, I want to use AI to help me have more agency. This is going to be the challenge of our lifetimes, not just for government, but for our kids. We're seeing for school children, for education, how do we use AI to ask us questions instead of to give us answers? It's very, very good at that, but we have to direct it to do that. So, how we think about learning, how is the Canada School of Government – you rethink training public sector in a way that is using AI as a coach instead of as a crutch? That is the question. So, we're doing a whole series, an education series coming up on AI and engagement with workers. What is the conversation we want to have with our own workforces around AI where we're worried about this problem; we're worried about losing our jobs and getting fired; we're worrying about what it means for our own agency. It's a conversation we have to have and make affirmative decisions. The AI doesn't do anything by itself. It does what we tell it to do, so we need to tell it the right things.

Vanessa Vermette: OK, please go ahead.

Audience member: Hi, thank you so much. And thank you so much for the presentation and the work that you do. It's very appreciated.

It seems like Democratic AI requires the people in decision-making powers to hold democratic values, to hold it appropriately, and to wield it appropriately. We do know that the people who do the procurement, who are getting lobbied, they're speaking – there's lots of economic incentives and not a lot of guardrails in place right now. And we know that decisions are made by people for various reasons. They can be political, they can be philosophical, it can be based on evidence, but those are all varying motivations.

So, to that point, as public servants, we're not deciding, in what programs are you doing it, we're not doing the procurement, we're not figuring out best-case use scenarios in a lot of ways. How can institutions build a framework or supports or guardrails to ensure that the decision makers and the leaders are ethically making the right choices but also looking to – like the VAR problem – we're not solving for problems we don't have. Because someone at the top has a very different idea of what problems we're facing than an average public servant. So, what frameworks can we do? Thank you.

Beth Simone Noveck: Yes. Hmm. So, I think there's some big – I don't want to sound too naive or Pollyannish. There's also corruption that isn't right in the procurement space, in particular – "My brother-in-law has a thing that I want you to buy" – That has been a problem, it's always been a problem. That's not unique to AI. There will be bad decisions that are made because somebody's brother-in-law, whatever. You know, we're –

[01:15:02 A RCAF jet roars overhead.]

Beth Simone Noveck: Woo!

Audience member: We're getting a new Governor General today.

Vanessa Vermette: Oh, that's right.

Beth Simone Noveck: Oh, that's what that is. Okay, as long as that's a good thing.

Vanessa Vermette: Celebration. Celebration.

Beth Simone Noveck: Celebratory – whatever that was – flyover.

I'm thinking of a few weeks ago when Kristi Noem was fired. It was in the wake of the realization or the revelation that they had spent – how much? Somebody may remember the number – a very, very large amount on shooting videos of her on horseback, done by a company in a non-competitive process that had actually never shot a video before. That will not get investigated now and you're not going to hear anything about it. But anyway, there was obviously some kind of malfeasance that was going on there, or so it seemed to be.

But I think the exciting thing is, number one, lots of good examples now of people using AI in the procurement space to make the process more efficient, more transparent, free up time for asking the harder questions. So, again, this is putting to one side, you have to have the intent to do things well in the first place. But when I can free up time, as jurisdictions like Utah and Washington DC and Boston are doing, with really measurable results, lots of great efficiency numbers – Boston is doing a study now with Harvard on how much time they've saved on their procurements, on just the routine and [inaudible] part of it, on the paperwork part of it – my hope is that that opens up more time for changing how the process works.

And this comes back to not just how we do what we're doing now more efficiently, but how we do what we're doing now better, so that we're spending time on asking the question, what is the problem that we are solving? Does this tool actually solve the problem? Are we setting the right metrics, in terms of ensuring effectiveness, not just measuring on the basis of cost? We can forgive a lot of failures to do that in a time in which we had so much paperwork to do.

And so, I'm optimistic that there are better ways of doing things right and that we can then start to say, "Hey, let's make sure that one of the questions we are asking is, is this thing effective?'

We're about to roll out a new course together with a group called Partners for Public Good, which is a spin-out out of Harvard that focuses on procurement, that is specifically about how to use AI to enhance the procurement, the values of the procurement process, by saving time on some of the routine stuff. There's a second thing we will do later this year, which is focused on the "what we're buying" and specifically around these questions of how to measure the effectiveness of AI.

But even if we're buying a pencil or we're buying something else, we want to make these processes both more competitive and more inclusive and more effective for the taxpayer. And that's the hope that, again, AI will allow us to rethink how we do our job.

Vanessa Vermette: Thank you for your question. I see Melanie's already at the microphone, and then I'll come here, and that'll probably be – OK, we've got 3 people in the queue, and I think that'll bring us to –

Beth Simone Noveck: Should I ask all 3, and then I'll—

Vanessa Vermette: Yes, why don't we do that. If you could keep your questions short and we'll get to all of them.

Beth Simone Noveck: I'll try to keep the answers shorter, sorry.

Vanessa Vermette: Yes, maybe you're the problem, actually.

Beth Simone Noveck: I am the problem.

Vanessa Vermette: Please go ahead.

Melanie: Hi, I'm Melanie, I'm executive faculty here at the School, but I have a past in open government. We worked together 10 years ago, I remember, so good to see you.

Question, as [a] faculty member, what kinds of skills do we need to get now for our new employees and how do we actually retrain employees? Like reading, writing, summarizing used to be so important and now it can be assisted. And so, what do we focus on with AI?

Vanessa Vermette: So, question about skills and where to focus training.

Beth Simone Noveck: You ready to answer your own question? I'm guessing you have a good answer, probably a better one.

Audience member: Yes. Can I move this up a bit without hurting anybody's ears? I just wanted to go back to what you were talking about with the conference with Sam Altman. And you mentioned you think it's important that we get him, and people like him, to agree that privacy can be something that is maybe baked into AI in order to protect all of us. I'm just wondering, how possible is that to do without the support of federal government? Because I'm guessing in the States you might have that problem, and what happens over there could influence and affect what happens over here. Thanks.

Vanessa Vermette: Okay, great. Thank you. And then one last question over in the back.

Erin: Hi, my name is Erin. Hello. I really actually support that last question, so if we run out of time and you have to skip one, you can skip mine because I could see how my question could be a bit of a segue, but I just was hoping you could either share your thoughts or concerns, like pick your top concern, on how AI is now fighting wars, as opposed to troops being – not that troops aren't out there but I mean, AI is now involved in fighting wars – and how defence of our sovereignty and our rights, our democracy, just what your thoughts and feelings are about that. So, it's a little bit of a segue, I get it.

Vanessa Vermette: Yes. Thank you. Okay, and we'll squeeze you in as well. Yes, just keep talking and I'll kick in.

Audience member: First of all, thank you so much. I'm currently a fourth-year student at the University of British Columbia, and I would say that I'm on the brink between two different generations regarding the use of AI. So, there's one that's more we faced that the AI has been developed and we're trying to take the reins on it, versus the younger generation that's coming into university – I'm really seeing amongst my peers – back to the question of the dependence and the reliance on AI.

So, one of my concerns is so much dependence to the point that we're unable to even recognize some of that AI use and even the recognition of the inefficient or lack of quality outputs that it's putting out to the public. And therefore, creating mistakes that incur higher costs. So, I know now if we implement those frameworks, we can try to take the reins so that in the future we can avoid that complete dependence on AI. But in a pessimistic kind of approach, could we be potentially running out of time now because AI is being so vastly implemented? So yes, that is my question. Thank you.

Vanessa Vermette: Yes. So, maybe start there so that we can end on a positive note.

Beth Simone Noveck: Which one do you think is positive in that list? Not the war one.

Vanessa Vermette: Not the war one. And then not the, "Is it already too late?" one. Unless you're going to say it's not too late.

Beth Simone Noveck: I promise to be positive and to be brief. So, let me work backwards here. On the universities point, we've seen lots of this booing of students at graduation now, I'm worried about this. I think there's two parts to this.

One is we need to be closing our laptops in a lot of contexts. This is again not just an AI thing, but we need to be doing a lot of things without technology to reintroduce the humanity into how we operate. But I am deeply excited about this – and I do this, so I'm totally biased – about the opportunities to get kids using AI around social impact. And again, it's not about the AI, it's about the social impact.

I think the opportunity to rethink think and retool universities to focus on impact. We have spent 20 years now focusing on how to help kids start a business and not enough entrepreneurship has always been the hallmark. And I think the opportunity to now talk about public entrepreneurship, to talk about public service, kids want to address climate change, want to do good in the world. It's what all the data shows, and I think it's important for our universities to be thinking about what are the impact opportunities. AI can be a big help for that. But the fundaments, it's time to close the laptops. So, maybe not the answer you were expecting from me, but I have a teenager, so I'm sensitive to this topic.

Fighting wars. Again, AI is a data tool. These are data processing tools where it helps with precision and accuracy so that we can reduce collateral damage. Great stuff, but we're deeply worried about humans not being, not just in the loop, but not in the lead in how we do things. So, this is a much longer topic about the risks in warfighting in particular, the unmanned uses of AI for surveillance, uses of AI for warfighting, there are so many very major risks. I don't want anything I said to minimize those dangers.

And I'm very grateful that there are lots of people talking about and focusing on these things. I just don't want the exclusively – if we're only talking about those, then we're not talking about how are we feeding the hungry and how are we delivering a better benefit and how are we doing those things because we're so focused on the wars problem.

You know, the war problem is, again, not just an AI problem. Leaders engaging in misguided and fruitless wars and putting people's lives at risk is a very age-old phenomenon. So, at some level, I would rather have robots fighting those wars than humans dying for them. So, a poor answer to a really, really hard question.

Reading and writing, and how we rethink how we do things. Oy, oy, I'd be interested in like whether we think collectively that this is a repeat of what we faced when we introduced the calculator, where we had a lot of these same debates, like would we stop being able to add? Is the calculator going to render us innumerate? When we introduced the word processor, are we not going to be able to write anymore?

We have word processors and people still write novels. Is this qualitatively different because, yes, is Amazon now full of AI slop books that are written by AI that a human has not looked at? Is it going to get rid of poetry and literature as we know it? I hope not, and I think we're seeing a big swing back towards an interest in these things, but I think it'll get worse before it gets better, and we will need to have conversations about how do we continue to teach how to read, and how do we do it with these tools?

So, where I'm excited is, if I want to help a kid learn how to read and that kid likes dinosaurs, I can now give them a book about dinosaurs. And then I can create a book – this kid likes unicorns – I can give this kid a book about unicorns and maybe it's going to actually improve reading. One of the things we did last year is actually spend a lot of time on working on helping policymakers understand uses of AI in the classroom for literacy.

Again, used well, we are addressing things like reading and writing and innumeracy and things that we have – frankly, it's not like we're living in a golden age that we're going to ruin. In the U.S. at least, we've only got 40% of kids reading at grade level. So, we have nowhere to go but up, frankly. But yes, I'm not trying to make light of the problem. It's that we worry about the dependency issue.

States and the federal government, federal government's not going to do anything. They are retrenching a little bit and saying, "Okay, maybe we're going to do some testing of these models", but 50 states banding together – our 50-state AI leaders meet every 2 weeks. It's a group that said, "We want to meet more often, not less often." We can and will do things collectively, I think, especially at the state level and especially outside the US to be thinking about using AI for public purpose. It's where we're seeing a huge amount of innovation.

I am deeply excited, and I'm excited because those 50 states, [in] the complete absence of the federal government, all of those 50 states have now committed to an agenda for learning.

Innovate US has not 50 but 60 partners because we have all 50 states, including every Republican state, every Democratic state, and more cities besides now signing up to say, "It is time we have an agenda around public sector learning. Canada's doing it, so can we."

So, I am hopeful.

Vanessa Vermette: Thank you, Beth. I think that's a great place to leave it. I'm going to quote you as we close to say that AI does what we tell it to do, so let's tell it to do the right things. Please join me in thanking Beth Noveck. Thank you, Beth. And thank you, thank you to all of you for joining us today, and thanks to our audience online as well. Beth, any last words as we close?

Beth Simone Noveck: Enjoy the sunshine.

Vanessa Vermette: Yes, it's a beautiful day out there.

Beth Simone Noveck: Thank you for coming, much appreciated, and spread the word.

And yes, one last thing, which is we do publish the news every week on democracy, governance, and AI, so if you're doing something interesting in your agency or department, please email me, let me know. If you can spell my last name, it will reach me. Noveck at the GovLab, Noveck at Innovate US, Noveck at Northeastern, you name it.

Let us know because everything we're doing is about peer-to-peer learning. We charge nothing to our learners, but we do pay our faculty. You're in government, so you probably can't get paid, but if there's some good story to share, we do lots of this and we'd love to share it in the cross-border as well, not just in Canada, which is why we have learners in 80 countries and I'm about to set up Innovate Bhutan, because people are just showing up to teach and learn from one another.

So, all that to say, please come and share with us and even if we can't pay you, we will be very, very nice to you at the very least. And provide a grateful audience who is eager for the learning, so come visit us. Thank you.

Vanessa Vermette: Awesome. Thank you.

[01:30:42 The CSPS animated logo appears onscreen. Text on screen: canada.ca/school-ecole.]

[01:30:47 The Government of Canada wordmark appears.]