Bruce Schneier and Nathan E Sanders argue that American campaigns are using AI in the least interesting way available — as broadcast, the same one-to-many model as internet advertising 30 years ago and television spots before that — while a party in Japan has already turned it into a listening instrument. Team Mirai built an AI interviewer, collected more than 300,000 messages across 16,000 interviews, cited the results in legislative committee sessions, changed parts of its platform in response, and holds 12 seats in the Diet. Their claim is that none of this is out of reach for a candidate running in the US midterms.
The case for pessimism is already well stocked. Voters are anxious about what the technology is doing to the country, politicians are using deepfakes to spread lies, and the White House publishes low-quality material built for propaganda. What the authors add is that candidates are simultaneously missing the other half of the technology: AI can help them hear voter concerns in more depth, bring a wider set of citizens into a discussion, and build platforms that track what people actually want. There are almost no American examples. The examples come from Japan, Scotland, a handful of US universities and a few companies.
Schneier is a security technologist who teaches at the Harvard Kennedy School and at the Munk School of Global Affairs & Public Policy at the University of Toronto. Sanders is a data scientist affiliated with Harvard's Berkman Klein Center. They wrote the book "Rewiring Democracy: How AI Will Transform Our Politics, Government, and Citizenship" together, and this is an argument from it.
Broad listening is the category they care about: tools that gather public opinion in far more detail than a multiple-choice field on a survey form. Team Mirai's AI interviewer walks a citizen through a specific issue in a long conversation and helps them articulate a position; the party then scales the same approach across policy areas through a portal that uses AI to work through pending legislation. It calls itself a "party as tool" and builds instruments any Japanese party can use to reach voters.
The reason Japanese voters participated is an incentive, not a novelty effect. Talking to the interviewer returned more than posting on Twitter/X or shouting into nothing: Team Mirai members quoted the results of those interviews directly in legislative committee sessions, published summaries of what they had collected, and revised the party program based on user proposals. Participants could see that the party was reading and might act. The party is open about its lineage, pointing to Taiwan's civic hackers from the zero-government movement, who built political influence out of a campaign for transparency.
Scotland offers the many-to-many version. CrownShy, a nonprofit funded in part by the Scottish government, is building a platform meant to move the town hall meeting into digital space. Its tool, Comhairle, combines AI interviewing with software for summarizing competing viewpoints, running virtual assemblies and publishing video responses, so legislatures and campaigners can consult a large number of citizens at once. Both Team Mirai's and CrownShy's systems are open source, and although both are funded by political actors — a young party in Japan, the governing party in Scotland — they are built to improve democratic process rather than to serve one side.
American candidates are not short of comparable tools, most of them built in America. deliberation.io, connected to Stanford, runs structured dialogues with thousands of participants using AI and has been tested on public input sessions with Washington residents. Cortico, a project at MIT, mines recorded conversations for opinions that are underrepresented in public debate, and is currently running listening sessions in libraries across the country. Talk to the City, from an American nonprofit, analyzes large bodies of stakeholder feedback. Remesh, an American startup, sells a commercial service that generates recommendations from AI-run dialogues and has been tested in policy development.
There is also a history here, and it did not end well. Two decades ago, organizations like Code for America and Barack Obama's 2008 campaign team built software aimed at better policy and broader civic participation, and Congress eventually funded a new federal agency, the US Digital Service. In 2025 the Trump administration repurposed USDS into the US Doge Service. This spring, Higher Ground Labs, a Democratic-aligned investor in campaign technology, launched a new fund whose focus areas include "AI-Native Campaign Systems" and "community-governed messaging platforms" intended to surface genuine citizen-originated ideas in real conversations.
The Japanese and Scottish cases are strong, and that is exactly why they do not transfer cleanly. Team Mirai's listening worked because Team Mirai legislates: the return a citizen got for a long interview was committee testimony and a changed platform. A challenger for a House seat has nothing equivalent to offer between now and November — no committee, no bill, nothing to point to except a promise to remember what you said. Broad listening in a campaign context asks people to invest more effort than a survey in exchange for less proof that it mattered. That is a product problem, not a technology problem, and none of the tools on the American list solves it.
The money makes it worse. Every US entry the authors name is a university project, a nonprofit or a single startup; the only campaign capital in the picture is a Democratic-aligned fund. If broad listening arrives in American politics through that door, it arrives as one party's infrastructure — the precise thing the authors praise Team Mirai and CrownShy for avoiding by open-sourcing their systems and building for anyone's use.
The question the argument leaves alone is what a campaign does with the transcripts. Sixteen thousand structured interviews are not only raw material for a platform; they are the most detailed voter file anyone has ever assembled, with every stated concern attached to a named person who volunteered it. The authors' case is that the output is better policy. The identical input produces better microtargeting, and nothing in the framework distinguishes a campaign that listens from a campaign that profiles. In Japan the results went to a committee hearing. In a US general election there is no equivalent public venue where the data has to surface.
There is also a fault line running through the authors' own argument. They urge voters and politicians to separate the technology from the companies profiting from it, and their prescriptions go well past campaign tooling: strip big tech money out of politics, hold AI companies liable for the harm their models cause, tax their revenue, and possibly nationalize them if the AI bubble bursts. Those are serious positions. They also sit awkwardly beside an invitation for every congressional candidate to build their voter outreach on top of AI systems, which would make campaigns a new customer base for the industry the same authors want taxed, exposed to liability and held in public hands.
Polarization is already sorting the parties into AI skeptics and AI boosters, and the authors want the technology judged separately from the firms selling it. The trouble is that broad listening will be judged by who shows up with it first. If it enters American politics as a Democratic campaign product funded by a Democratic campaign fund, it will be read as a partisan weapon regardless of what its source code says — and the one genuinely democratic use anyone has demonstrated for AI in politics will have been spent as a tactical advantage before it was ever tried as a public one.