# Democracy has a listening problem. These AI tools could actually help

> Source: <https://www.fastcompany.com/91563889/democracy-has-a-listening-problem-these-ai-tools-could-actually-help>
> Published: 2026-06-25 11:00:00+00:00

When four-year-old Joca, a golden retriever with dark soulful eyes, died on a scorching Brazilian tarmac after an airline’s tragic mistake, the nation’s outrage could have remained just another trending hashtag. But Fernando, a young man in São Paulo, submitted an online legislative proposal for “Joca’s law.” Senators held a hearing, and eventually the Senate and the House passed legislation.

While most online engagement today amounts to clicking, liking, and sharing—what the late political theorist Benjamin Barber called “the caverns of private solitude”—Brazil’s system shows how institutions might use digital tools differently.

In 1999, I saw both the promise and peril of using technology to create that participatory spirit. Working with Barber, I built [Unchat](https://vimeo.com/394225913), the first software designed specifically for democratic deliberation.

With his distinctive baritone voice and youthful cowlick, Barber, who died in 2017, was fond of arguing that democracy, if it is to survive the assaults of hostile modernity, needs to do more than enable free speech. It needs to give citizens the power to decide and act. Unchat [allowed](https://www.bu.edu/law/journals-archive/scitech/volume91/noveck.pdf) groups to come together to debate, discuss, and decide across distances.

Uniquely, the software enabled each participant to take turns moderating the conversation, with control rotating after a set time. This was an intentional construction to encourage fairness and collective decision making. Our aim wasn’t just to enable talk for its own sake, but to create opportunities for practical participation, where citizens actively shape political decisions.

The late 1990s were heady times for those who believed in the transformative potential of the World Wide Web to reimagine democracy. Vint Cerf, one of the internet’s architects—rarely seen without his signature three-piece suit and French cuffs—remarked that, with the Internet’s capacity for two-way communication, “[p]eople could provide feedback. I think a lot of legislators originally thought of the Net as another avenue for communicating with their constituents,” he noted, “except then they discovered the constituents could talk back and talk to each other!”

Since the ’90s, the number and variety of platforms have only grown, and with them the ability to reach out to new audiences; but our institutions have remained largely unchanged. Stories of citizens directly shaping how institutions decide are vanishingly rare, even in most democracies. Despite the proliferation of online engagement tools, two-thirds of people still believe they have little influence over government decisions. Social media platforms, designed to maximize ad revenue rather than dialogue, have transformed citizens from active participants into passive spectators, staring at what comedian Hasan Minhaj calls the “rectangle of sadness,” distracted by national politics we cannot influence.

We thought that the Web would usher in a new Athenian golden age. But when everyone talks, institutions cannot listen.

Even when participation takes place at scale, it rarely penetrates the institutions where power resides or becomes sustained practice. In addition, meaningful participation remains too taxing for most institutions to sustain over time. The UK Cabinet Office estimates that “a consultation attracting 30,000 responses requires a team of around 25 analysts for 3 months to analyze the data and write the report.”

Exceptions highlight the path forward. [Peer-to-Patent](https://obamawhitehouse.archives.gov/open/innovations/Peer-to-Patent”), the online public engagement my students and I launched with the U.S. Patent and Trademark Office in 2005, succeeded by integrating public participation directly into Patent Office processes and creating a structured handoff between public input and examiner decisions.

Similarly, [Challenge.gov](http://challenge.gov), operated by the US General Services Administration, enabled over two thousand prize-backed competitions over the last decade. Federal agencies pose specific problems for public solvers with defined rules and outcomes. The government [reported](https://www.americanprogress.org/article/enhancing-accessibility-u-s-elections/): “Challenges have produced concepts for the next ‘lunar loo’ (space toilet), an improved digital wallet user interface, protecting fish from water infrastructure, opioid detection in international mail, and ‘getting out the count’ for the census. And yes, those self-driving vehicles got their start in federal prize competitions too!” (The platform was decommissioned by the Trump Administration this spring, but government agencies continue to [run challenges](https://www.usa.gov/find-active-challenge) through their own websites.)

These successes remain rare. For most platforms and experiments, participation is staged at the edges of governance rather than embedded in its core. The result is a cycle of citizen engagement that looks impressive from the outside but has little effect on what governments or companies do.

When an eighty-two-year-old retiree from the coastal city of Maceió called Brazil’s Senate hotline to suggest that medication labels needed larger font sizes, he had no idea that a senator from Amazonas would champion his simple suggestion.

From a proposal to end housing subsidies for deputies and judges and terminate perks for former presidents to the idea to legalize marijuana and ban straws, the Brazilian case stands in stark contrast to the United States, where a member of the public has almost no way to participate in what our Congress does.

Central to the success of Brazil’s participatory lawmaking is a significant investment in the technology for public engagement. The Senate Digital Services team builds the tools, while a dedicated fifteen-person e-Citizenship team moderates submissions, coordinates with parliamentary offices, and ensures the smooth operation of its four participatory processes.

The e-Citizenship team also conducts outreach, actively promoting civic engagement and, above all, working to integrate participation into institutional practices. Whereas other systems have buckled under the volume of public input, Brazil’s Senate processes allow millions of citizen interactions through a combination of clear thresholds, dedicated staff, and well-designed digital infrastructure. The public can even “vote” directly on all bills under Senate consideration, from their introduction to final processing. While these votes aren’t binding, they give lawmakers valuable insight into public opinion. In the decade since the system went live, some 15 million registered participants have cast 34 million votes on thirteen thousand bills.

While the Brazilian Senate’s digital democracy initiatives have achieved remarkable reach, the tension between scale and quality of engagement limits progress. The sheer volume of participation—thousands of legislative ideas each month, tens of thousands of hearing questions annually, millions of votes on pending legislation—creates significant operational pressures on Senate staff and has led to design decisions that make the processes less useful for the public and politicians. Input is too often duplicative and of low quality.

Now the e-Citizenship team is turning to [artificial intelligence](https://www.fastcompany.com/section/artificial-intelligence) to address current limitations while expanding reach and impact. AI could help citizens craft more effective proposals while helping staff identify and connect related submissions. It could analyze successful past proposals to guide citizens in creating submissions more likely to gain institutional traction, offering tips as people are drafting.

With AI support, citizens could submit comprehensive proposals that define problems, outline solutions, and provide supporting evidence. The technology could quickly analyze submissions, pointing out areas where additional data or explanation might strengthen the case. Instead of one submitter, members of the public could collaborate to write better proposals in which citizens build on each other’s ideas to create more robust legislative suggestions.

AI analysis could also address the platform’s problem with duplicate submissions. Currently, citizens often submit similar proposals about the same issue, splitting potential support among multiple entries. For instance, there were at least three Joca proposals. AI could identify these overlapping submissions and actively connect citizens working on similar ideas, encouraging them to collaborate rather than compete for attention and votes.

The current system of displaying only the three most popular ideas on the front page of the Senate’s website could evolve into something more dynamic and fairer. Rather than relying solely on vote counts, AI could create an intelligent rotation system that considers engagement patterns, relevance to current legislative discussions and news, geographic representation, and topic diversity. This would bring attention to promising ideas that might otherwise be overlooked while keeping the platform democratic.

For Senate staff reviewing submissions, AI could streamline the evaluation of citizen legislative proposals without replacing human judgment. Natural language processing could identify potential constitutional issues, categorize proposals by topic, and generate initial assessments of feasibility.

There are models for how to batch, cluster, and evaluate inputs at scale. In 2024?, the city of Bogotá deployed the “Chatico” chatbot via WhatsApp and the Web. [Chatico](https://bogota.gov.co/mi-ciudad/gobierno/conoce-mas-de-chatico-el-asistente-virtual-para-acceder-servicios) lets citizens engage in participatory budgeting, propose “citizen causes,” submit votes, and interact about city services. Over one two-week campaign , 56,000 residents cast 132,000 votes on 2,700 local projects. Behind the scenes, Chatico uses AI to classify and organize messages and feed results into live dashboards for city officials. Unlike one-off consultations, it is embedded within Bogotá’s governance structures and overseen by the city’s ICT office.

In the UK, the Department for Work and Pensions has piloted an AI system called “white mail” that processes twenty-five thousand citizen communications a day, including handwritten letters, to flag urgent cases from vulnerable individuals and prioritize them for human review. (Notably, the system has also [raised](https://www.theguardian.com/society/2025/jan/27/dwp-ai-whitemail-benefit-claimants-applicants) concerns about transparency and accountability).

In Derry, Northern Ireland, the [Voice Matters project](https://www.peoplepowered.org/news-content/digital-participation-case-study-ireland”) has piloted AI-supported transcription and categorization to make sense of public deliberation in real time. The system transcribes discussions, clusters related contributions, and visualizes areas of agreement and disagreement as meetings unfold, allowing facilitators to highlight overlooked perspectives without drowning in repetition.

Brazil’s [Public Consultation platform](https://www12.senado.leg.br/ecidadania/principalmateria”), which has already collected over 37 million votes on legislation since it launched in 2013, could evolve far beyond its current thumbs-up, thumbs-down voting system. Rather than simply tallying yes and no clicks, AI could help capture and analyze the reasoning behind citizens’ positions, transforming what is currently a binary approach (approving or disapproving of a bill) into a deliberative dialogue between citizens and their government.

Google’s Jigsaw unit, which focuses on technology and human rights, and which has [developed](https://support.perspectiveapi.com/s/about-the-api-faqs?language=en_US) machine learning systems to [help](https://www.poynter.org/tech-tools/2019/jigsaw-is-fixing-comment-sections-one-language-at-a-time/) publishers like the *New York Times* manage toxic comments, has recognized that controlling harmful content isn’t enough to create meaningful democratic participation.

Christopher Small of Jigsaw explains that he and his colleagues are focusing on the “last mile problem”—how to transform the output of online discussions into something institutions can use. “These platforms produce massive amounts of data,” he says. “The challenge is making sense of it all.” Jigsaw’s goal is to provide decision makers with actionable insights—as Small puts it, telling a mayor, “Here are five key areas of agreement that could inform legislation, and here are five complex points of contention around this infrastructure plan.”

They have experimented with using AI to produce summaries for decision makers in Bowling Green, Kentucky, for example. They [ran an engagement in 2025](https://www.fastcompany.com/91278879/jigsaw-bowling-green-kentucky-civic-engagement) where 8,000 residents submitted 4,000 ideas and cast over a million votes about their community’s future. Their “sensemaking tools” use machine learning to categorize public comments into topics, identify patterns in how residents vote on different ideas, and generate concise summaries that highlight areas of consensus and disagreement. Rather than just facilitating more conversation, Jigsaw’s approach acknowledges that democratic participation requires translating public input into usable information that can influence policy decisions. Without this translation layer, even the most robust public conversations risk becoming performative exercises rather than meaningful inputs to governance.

For decades, many scholars and policymakers have treated public participation as a problem to be managed rather than a resource to be cultivated. This “realist” view, which gained prominence after World War II with the growth of money in politics, treated participation as destabilizing. Many academics even challenged the notion that ordinary Americans have the time, competence, or capacity to participate, writing off the public as incapable.

This skepticism has led institutions to design consultation processes more as exercises in public relations than as genuine attempts to share power. And when we evaluate democratic innovations, it leads us to focus on metrics like participation rates rather than actual policy impact.

Brazil’s experience suggests that participation becomes meaningful only when it is connected to decision-making. What makes this moment different is that artificial intelligence may finally give institutions the capacity to hear, organize, and act on public input at a scale that was previously impossible.

Unlike earlier Web-based platforms that only expanded the volume of talking, we can use AI to make sense of the collective intelligence of our communities and uncover better ways to connect participation to decisions and action.

The challenge is no longer getting people to speak. It is building institutions capable of listening.

*Adapted from *[Reboot: AI and The Race to Save Democracy](https://rebootdemocracy.ai/)*, Yale University Press, 2026. *