Civic Information Architecture
In the six years since the Our Common Purpose report was released, dramatic shifts to the information ecosystem have transformed the relationship between citizens and the media, providing serious challenges to an informed and participatory democracy. Journalism faces business model challenges aggravated by a shift away from search and toward AI, as well as governmental attacks on journalistic independence and on public media. Influential technology platforms have retreated from the difficult, controversial work of fact-checking, seeking to reduce their political exposure, while platforms like Truth Social and X are now explicitly associated with political actors. We believe even larger changes are brewing, from promising experiments in community ownership and management of social networks to the emerging idea of AI sovereignty, which may see governments around the world challenge American and Chinese companies over the ownership of powerful AI tools.
The six years since the Commission on the Practice of Democratic Citizenship offered analysis, strategies, and recommendations in its report, Our Common Purpose: Reinventing American Democracy for the 21st Century, have been eventful ones for the civic information space.1 Some developments are the outgrowth of long-term trends, most notably the weakening of accountability journalism through the shrinkage of advertising revenues. Others are more dramatic, including two abrupt shifts in how social media platforms see their responsibilities to their users and the broader public.
The resulting picture from these slow and fast shifts is not an encouraging one for democracy. Journalism, and particularly local journalism, continues to be in crisis, with difficulties exacerbated by the defunding of public media, leaving some of America’s least resourced communities with even less visibility into local politics and civics. The promise of social media as a broadly accessible public sphere is challenged by the overt politicization of some online spaces (including a social media network owned by President Donald Trump and his family) and the increasing concentration of tech platform power.2
Yet within a complex landscape, there are glimmers of promise. Frustrations with the direction X (formerly Twitter) has taken under the leadership of Elon Musk has sparked a wave of interest in decentralized social media, drawing users to platforms like Bluesky and Mastodon and sparking a broad conversation about what would constitute a decentralized social network. The donor community has focused intensely on the problem of supporting local journalism through rethinking ownership and business models. And some promising experiments are connecting online conversations and information resources, attempting to build new participatory public spaces controlled by local communities instead of multinational corporations.
Understanding these ongoing shifts requires an understanding of both historical events of the past years and significant changes to the technological landscape. The evolving nature of social media offers perhaps the best illustration of the intersections of history, technology, and politics.
Moderation is an uncomfortable necessity for social media platforms. Decisions to remove some types of content and permit others, or to give content higher or lower priority on users’ feeds, are inevitably controversial. They also require resources: both machine-learning algorithms and human labor to train those algorithms and intervene to make close calls. Yet unmoderated social media is financially unsustainable—advertisers concerned with “brand safety” will flee platforms where the conversation turns obscene or hateful—and purely unmoderated spaces like 4chan tend to have sharply limited audiences.
Platforms like Meta’s Facebook have long asserted an unwillingness to “censor” user content, while maintaining complex and widely debated rules about what content is allowed (see the #freethenipple campaign aimed at Meta’s prohibitions on images of “female-appearing” breasts).3 As legal scholar Evelyn Douek has observed, developments in 2020—the same year the Academy’s Commission issued Our Common Purpose—fundamentally changed internet moderation.4 The public health emergency around the COVID-19 pandemic led platforms to take down claims about the infectious disease that were not consistent with advice from public health organizations, and to give more weight to information from trusted sources like the World Health Organization. This shift was generally welcomed, and technology companies were cheered for taking responsibility during a global health emergency.5
This increased willingness to block dangerous content shifted from public health to the political realm in May 2020 when, as Donald Trump began claiming that mail-in voting—a pandemic safety measure—would “rig” the 2020 presidential election, Twitter labeled Trump’s tweet as “potentially misleading.”6 After initial statements by Facebook CEO Mark Zuckerberg that the platform did not want to be arbiters of political speech, many of Trump’s statements on his public Facebook page were also marked with warning labels.
Two dramatic steps at the end of 2020 marked the apex of the shift from reluctant to aggressive moderation of political content. Weeks before the 2020 election, Facebook and Twitter took steps to block a New York Post article about Hunter Biden’s laptop, in part out of fear that the leaked information that led to the story might be foreign disinformation designed to sway the presidential election. Both platforms changed course under public and political pressure, but many Republicans felt like a line had been crossed when a story that met the editorial standards of a major newspaper was blocked by social media platform moderators.7
After the riots at the U.S. Capitol on January 6, 2021, Donald Trump’s accounts on most major social media platforms were temporarily or permanently removed, a practice called “deplatforming,” which had previously been directed primarily at serial conspiracy theorists like Alex Jones, founder of the Infowars website.8 The logic behind Trump’s deplatforming rested on concerns about public safety: platforms worried that his ambiguous statements about the Capitol attack, in which he told rioters “We love you, you’re very special,” might inflame further political violence.9
The deplatforming of a sitting president triggered a complex backlash and a second swing in social media policy. For some international leaders, including then–German Chancellor Angela Merkel, the removal of Trump from social media was “problematic,” raising questions about freedom of opinion and expression.10 Trump has sued several of the platforms that removed his accounts, and several have settled. The decision to settle may be based less on the legal merits of specific cases than on an unwillingness to be in conflict with a sitting president.11
For Republican leaders in the United States, the deplatforming was a casus belli to be addressed through state-level legislation. In 2021, Florida passed Senate Bill (S.B.) 7072, also known as the “Stop Social Media Censorship Act,” and Texas passed House Bill (H.B.) 20, which sought to prohibit viewpoint discrimination on social media platforms. Both laws were challenged via lawsuits, leading to a circuit split in the U.S. Courts of Appeals—the Fifth Circuit upheld Texas’s law while the Eleventh Circuit struck down Florida’s.12 The cases were jointly considered by the Supreme Court as Moody v. Netchoice, LLC and Netchoice, LLC v. Paxton and argued in February 2024. The Court remanded the cases to the lower courts in July 2024, instructing them to conduct a more thorough analysis of the conflicts between the laws and the First Amendment.13
While the judicial review process is lengthy and complicated, market forces react more quickly. In February 2022, the Trump Media and Technology Group launched Truth Social, a social network similar to Twitter. The president began posting exclusively on Truth Social, though automated accounts repost his messages to other social networks, increasing his reach. Similarweb, a company that uses AI to estimate web traffic and offer traffic and audience analytics, sees only 24 million monthly visitors to Truth Social, compared to 4.5 billion for X or 148 million for Bluesky.14
As Trump was launching Truth Social, Elon Musk—an increasingly vocal supporter of the president—began making significant investments in Twitter, before announcing on April 14, 2022, that he would buy the company for $43 billion, a significant premium over its stock price.15 Over the next six months, Musk sought to undo the deal over concerns about automated accounts on the platform and Twitter subsequently sued him in Delaware’s Court of Chancery to force him to complete the sale. Shortly before the suit against Musk went to trial, he closed the deal on October 28 and promptly fired most of Twitter’s executive team and roughly half its staff.16
The following spring, Musk renamed the platform “X,” in a shift that paralleled significant changes in content moderation.17 X stopped moderating “false or misleading information that could bring harm to crisis-affected populations” and ended its “civic integrity policy,” which prohibited misleading speech about election results.18 Musk also welcomed back controversial figures, including Kanye West, Andrew Tate, and President Trump, though Trump’s account is relatively quiet today, largely just posting screenshots of his Truth Social account or reposting X content supportive of himself and his allies.19
In reaction to policy changes and the return of controversial figures, many left-leaning users fled Twitter during and following Musk’s takeover. Another wave left after Trump won the 2024 presidential election and invited Musk to the White House as a “special government employee” heading the United States Digital Service (renamed the “United States DOGE Service” by executive order in 2025).20 X alternative Bluesky briefly topped mobile application download charts in November 2024 as users sought Musk-free social networking, but X itself has maintained a steady presence near the top of app download charts while growth on Bluesky has recently slowed.21
The rise of decentralized social networking as a response to the Musk takeover of Twitter will be examined more fully when we consider signs of hope in the civic information environment. But the launch of Truth Social and the transformation of X are best understood as the next act in the ongoing saga of content moderation. Before 2020, platforms positioned themselves as politically neutral and generally avoided taking overt stances on free speech. During the COVID-19 crisis and the 2020 presidential election, they became increasingly willing to act, citing either public health or civic health as a motivation.
Between 2022 and Trump’s 2024 reelection, some platforms swung back to a “free speech” position, but now without a pretense of political neutrality. In August 2024, Zuckerberg expressed regrets at caving to government pressure under President Joe Biden’s administration: “I believe the government pressure was wrong, and I regret that we were not more outspoken about it.”22 And in a powerful display of symbolism, the CEOs of Meta, X, TikTok, and Google (which owns YouTube) appeared as featured guests at Trump’s 2025 inauguration, suggesting, if not explicit alignment with Trump’s agenda, a realization that failing to cooperate with the White House could be bad for business.
Perhaps the most surprising development around social media and its political alignment has been the strange saga of TikTok, the social media platform based on short video posts launched internationally by the Chinese technology firm ByteDance in 2017. In August 2020, President Trump issued an executive order that prohibited transactions between TikTok and U.S. citizens, an act designed to remove the app from the Google and Apple app stores and cut it off from U.S. advertising networks.23 TikTok sued, and the service continued to operate during the remainder of Trump’s first term. In April 2024, citing the danger of giving Chinese businesses and thereby the Chinese government access to Americans’ user data, President Biden signed legislation requiring a ban of TikTok from U.S. app stores and hosting services unless the company sold its U.S. operations to American investors.24
The Supreme Court unanimously upheld the ban on grounds of national security following Trump’s second election in November 2024; for a year, his administration elected not to enforce the ban, instead seeking a deal with ByteDance to transfer control of the company to U.S. entities.25
In September 2025, the Trump administration indicated a forthcoming deal involving conservative media leaders and the Oracle Corporation, which is controlled by Larry Ellison, who is reported to be a “close confidant” of Trump.26 In January 2026, ByteDance announced that it had approved a sale that includes Oracle and an Emirati investment fund as buyers, as well as tech mogul—and Trump supporter—Michael Dell.27 Since TikTok’s U.S. division began operations under a U.S.-majority board and ownership in late January 2026, watchdogs have begun trying to determine whether TikTok’s vaunted recommendation algorithms are shifting toward favoring right-leaning content. While TikTok is understood as an entertainment platform, it also delivers news and information to 43 percent of Americans under thirty.28
By virtually every measure, the health of news media has declined sharply since the early 2000s: as print news circulation has shrunk, news audiences have moved online and technology companies like Google and Meta have captured huge percentages of the online advertising business. Newspaper employment dropped from over three hundred and sixty-five thousand jobs to under one hundred thousand between 2005 and 2023. During the same time period, over four hundred daily newspapers closed, leaving the nation with just over one thousand daily papers.29
The closure of newspapers is civically important because, although broadcast and social media are more popular places for most Americans to seek news, newspapers provide almost 50 percent of originally reported stories and nearly 60 percent of local news stories, according to a 2019 Duke University study.30 While online-only news outlets are beginning to fill some gaps in news coverage, the most successful in this space are national or international in scope, covering specific beats. Television executive Jeff Zucker observed that online ads yield “digital pennies” compared to “analog dollars,” making it particularly difficult to support online-only local news outlets through conventional advertising.31
As news audiences move online, intermediaries like search engines and social networks have captured an increasing share of news ad revenue. A user sharing a story on local elections on their Facebook account might well drive views to the local newspaper’s website. But it’s also possible that their friends might read the headline and brief summary embedded within their Facebook feed and not explore further, yielding an ad view on Facebook but not on the newspaper website. In 2020, Australia passed the News Media Bargaining Code, which required platforms like Facebook and Google to negotiate a revenue share with news organizations whose content they post.32 Despite initial resistance, and a brief period during which Facebook blocked news content to Australian users, scholars now estimate that the code will transfer $200 million AUD (about $142 million USD) annually to funding public-interest journalism.33 But when Canada passed the similar Online News Act in 2023, Meta simply blocked news from their platforms rather than agree to a mandated revenue sharing arrangement.34 Traffic to Canadian news sites has dropped precipitously, and other nations have been reluctant to follow in Australia’s footsteps for fear of facing similar blockages.
The integration of generative AI into search engines offers an even more significant threat to online news revenues. Search engines continue experimenting with AI-generated summaries that push links to news content down on the results page, making it increasingly likely that users will end their search at the search engine rather than visiting a news website.35 One study suggests that a site previously ranked first by Google for a particular query might expect to lose 79 percent of its traffic when AI summaries are added to the page.36 A 2025 study from the Pew Research Center examined web browsing data from nine hundred volunteers and discovered that only 8 percent of users who encountered an AI summary as a result of a search went on to click one of the links returned in the search, while three times as many ended their browsing session, likely because they found the answer they were seeking.37 Search engines are aware that news publishers see automated summaries as affecting their traffic and have responded by making links to content hosted elsewhere more prominent in their AI summaries, though the Pew study found that only 1 percent of users clicked on those links.
In Ernest Hemingway’s debut novel The Sun Also Rises, a character is asked how he went bankrupt and famously responds, “gradually, and then suddenly.”38 There are legitimate concerns that the replacement of search results with AI summaries will turn the gradual financial calamity American news organizations are facing into sudden collapse.
It is possible that some news organizations may benefit from AI licensing fees: The New York Times is licensing content to Amazon’s AI projects for reportedly as much as $25 million a year.39 Simultaneously, the Times is suing both OpenAI and Microsoft for copyright infringement, claiming the chatbots released by the companies have been trained on millions of news stories without permission or compensation.40
Courts may decide that the use of copyrighted material constitutes fair use because the training of an AI model is a transformative use of copyrighted material. This would damage the possibility of revenue through licensing, leading some lawyers to suggest that a compulsory license, in which AI providers would be required to pay a licensing fee to the news industry, might be necessary to save professional newsgathering.41
Without a solution like compulsory licensing, both news organizations and, eventually, AI companies face a looming crisis: if professional news organizations can no longer make money from online advertising because their traffic has been throttled by AI bots, the supply of reporting that AI systems rely on will dry up, eventually making it impossible to answer recent queries about what’s happening in the world. (While it’s possible to speculate that generative AI systems may eventually be able to produce original reporting, that is far beyond the capabilities of existing systems.)
Our Common Purpose suggests that the ongoing financial crisis in journalism would require consideration of public media models as part of the solution to ensure that quality reporting continues to be carried out.42 While the rise of AI makes the threats to journalistic business models more acute, the defunding of public media by the Trump administration indicates that the cavalry is not coming to the rescue.
In July 2025, the Trump administration rescinded over $1 billion in funding to National Public Radio (NPR) and Public Broadcasting Service (PBS) via the Corporation for Public Broadcasting.43 NPR and PBS will be able to continue operating through donor and sponsor support, but the Corporation for Public Broadcasting, a nonprofit that provided funding for American public broadcasters from 1967 to 2025 using congressional appropriations, shut down in January 2026. In addition to raising complex constitutional questions about who has authority over federal spending, the withdrawal of funding for public media by the Trump administration suggests we will need to find other solutions to the financial shortfalls currently affecting commercial journalism, which will need to pick up the slack resulting from defunded public media.
The defunding of public media should be understood as a broader attack on independent journalism that is critical of political power. President Trump has used a variety of tactics to pressure broadcasters and newspapers that he sees as providing politically biased coverage. He sued CBS over their editing of an interview with then–presidential candidate Kamala Harris. While most legal experts believe the suit had no merit, CBS’s parent company, Paramount Global, settled the case for $16 million, possibly because Federal Communications Commission (FCC) consent was required to approve a merger between Paramount and Skydance Media, which closed in August 2025.44 When comedian and host of CBS’s Late Show franchise Stephen Colbert referred to the settlement as “a big fat bribe,” Paramount announced that his show would be canceled at the end of its season in May 2026 due to financial reasons, leading to speculation that programming decisions were coming under political pressure.45
Humor in particular has been a site of conflict between broadcasters and the Trump administration. When comedian and television host Jimmy Kimmel complained that Trump supporters were rushing to pin blame for the 2025 assassination of Charlie Kirk on their political rivals, FCC Chairman Brendan Carr used a podcast appearance to insinuate that ABC stations should stop carrying Kimmel’s late night show or risk losing their broadcast licenses, noting, “We can do this the easy way or the hard way.”46 Carr’s threat was strengthened by the fact that several broadcast networks are already being investigated by the FCC for possible political bias. Hundreds of protesters marched on Disney’s headquarters (ABC’s parent company) in Burbank, California, and more canceled or threatened to cancel their streaming services unless Kimmel’s show was reinstated.47 ABC changed policy and Kimmel returned to the air less than a week later.
While the United States has rigorous freedom of expression protections under the First Amendment, there are no constitutional guarantees regarding the fiscal viability of any particular media outlets. While U.S. courts would likely prevent the White House from ordering a newspaper to suppress coverage of a particular story, Trump’s $15 billion defamation suit against The New York Times could bankrupt the newspaper should a court decide to hear it and find in the president’s favor.48
When Elon Musk took over Twitter in 2022, millions of users concerned about his proposed directions for the platform began to seriously consider the question of “platform power,” many for the first time. Platform power is the idea that the owner of a social media platform, search engine, or AI chatbot has significant control over the information a user sees or shares. Faced with the possibility that Twitter under Musk might be more tolerant of extreme speech, many users left the platform for alternatives like Bluesky or Mastodon.49
The move from Twitter/X to Bluesky and Mastodon is not just a move from one set of platform rules to another. These new social platforms are designed around the principle of decentralization. Different instances of Mastodon—independently run servers using the Mastodon software—have different rules on what content is allowable and different financial models to pay for the service. Some nodes (social.coop, for example) are governed and fiscally supported by communities, while others are commercial or personal projects controlled by a single person or team. While rules for what content is allowable can differ from one server to another, users of one server can still view content posted on other servers, if their rulesets are compatible.
Decentralization is appealing in theory but has proven confusing for some users in practice. Defederation—in which the owner of a server decides not to forward content from another specific server—is essential to the governance of social networks but can be hard for users to understand. When Gab, a white nationalist social network, began using Mastodon, many servers defederated, making it so that their users could not see posts on the Gab server and Gab users could not see their posts.50 Other defederations have involved sexually suggestive anime and controversy over reporting of trans issues.51 Defederation gives users vastly more control over what ruleset social media operates under—you can choose the ruleset you want and move to another server or open your own if you change your mind. Federated social networks require more thought and expertise than joining a centralized network like Instagram or X.
Users of Bluesky have faced another issue with federation: Is a network actually decentralized if there’s only one viable server? Bluesky has become a popular destination for refugees from X because the user experience is very similar to pre-Musk Twitter and because the service gives users a great deal of control over what content appears in their feeds. But some users have complained that so long as Bluesky primarily has a single set of servers run by the company that has written the software, it remains only theoretically decentralized. A grassroots effort called Free Our Feeds has proposed raising $30 million to build meaningful parallel infrastructure, allowing users concerned with governance of Bluesky to move to an alternative server. While they have fallen short of their original fundraising goals, there is widespread enthusiasm for the idea, including from the Bluesky team.52 In the meantime, Blacksky, a community on Bluesky focused on communities of color, has developed its own implementation of the protocol that Bluesky uses—the Authenticated Transfer (AT) Protocol—which may make it possible for knowledgeable users to run their own Bluesky compatible servers, much like users who start Mastodon servers.
While there has been more conversation about decentralized social networks than actual widespread adoption of these technologies, an increased awareness of how social networks operate and who has power and control over them is a step toward reshaping the information environment in more civically inclusive directions. In addition to conversations about the “fediverse”—a set of decentralized services that operate on the protocol used by Mastodon, and a parallel environment based around Bluesky’s AT protocol—real estate magnate Frank McCourt has launched Project Liberty, an effort to build decentralized social networks around a blockchain-based protocol that he believes will empower a landscape of competitive social networks.53
Our recommendation in Our Common Purpose to consider interoperable protocols as the basis for new, pro-civic social networks is where reality has steered closest to our aspirations, though the growth of alternative social networks parallels our suggestion for public-supported social networks that follow the logic of public broadcast media. And while we did not explicitly address AI in our 2020 recommendations, the rise of powerful AI chatbots is leading to the idea of sovereign AI, in which countries build their own AI systems on servers and data they control. It is possible that sovereign AI may open a discussion of technology in the public interest that might expand to include social media, search engines, and other key technologies.54
Encouragingly, we’ve seen creative experimentation at the intersection of local news and participatory media, suggesting that what might work is a blend of existing social networks and local media, cooperating to bring more local voices into conversations about local events. Vermont Public—the state’s public broadcaster—partnered with Front Porch Forum, an email-based hyperlocal social network that links citizens in each Vermont town, to shape their 2024 election coverage. Vermont Public and Front Porch Forum collected several hundred comments from each county in Vermont and launched a Citizens Agenda initiative, steering their campaign coverage to focus on issues the public identified as most important.55
It’s difficult to build explicitly civic-focused social networks—most people’s lives do not center on local or national politics. Front Porch Forum, which is largely used to help neighbors advertise tag sales and lost pets, has shown that it can be activated as a powerful civic tool at moments of political importance or natural disaster like the 2023 riverine flooding that devastated many Vermont communities.56
These “latent civic networks” may present an opportunity for local news organizations to build local audiences around interests and “activate” them at moments of civic need. The Arizona Daily Star is building a discussion platform—Wildcaster LIVE—around University of Arizona sports, connecting readers with sports reporters in real-time chats.57 These two-way relationships between news consumers and reporters are already shaping sports coverage and hold the potential to reshape coverage of civic issues, through a model like Vermont Public’s Citizens Agenda, or be activated in moments of crisis. And despite the challenges of developing local networks, New_ Public, one of the most respected research organizations focused on online communities, has launched Roundabout, a locally focused community product likely to test theories about the power of latent civic capacity.58
Digital efforts to make local news coverage more participatory won’t be enough to address the core financial challenges local newspapers are facing. The National Trust for Local News is a foundation-financed effort to acquire, combine, and operate local newspapers, creating efficiencies of scale while operating within an explicit public-interest framework designed to prevent the creation of news deserts. The Colorado News Conservancy, for example, was established by the National Trust to bring together the online news outlet The Colorado Sun with Colorado Community Media (CCM), an independent, family-owned chain of twenty-four newspapers and websites that serve Denver and its suburbs.59 The Conservancy recently bought a press from a failing Gannett operation and now uses it to publish its Colorado papers.60
The National Trust has followed similar approaches in Maine and Georgia and has established a promising model, but it has not been without controversy. As leadership of the nonprofit has moved beyond its founding team, it has sold several of its Colorado papers to a for-profit news company in Arizona, raising questions about whether these divestitures are consistent with its mission. Defenders of the National Trust argue that a combination of nonprofit and for-profit solutions will be essential to provide local journalism over the long haul.61
Perhaps the greatest surprise in the information landscape since issuing Our Common Purpose has been the rise of artificial intelligence as a central feature of the information environment. The introduction of ChatGPT as a web application in late 2022 demonstrated the power and promise of large language models—systems that create convincing text by extrapolating from patterns in huge corpora of existing texts—to answer questions and deliver information. ChatGPT effectively created a new, fast-growing category of web applications: chatbots that seem likely to supplant existing search engines.
The visibility of chat services like ChatGPT and Claude, as well as other “generative AI” systems that can produce images and video from natural-language prompts, is provoking society-wide conversation about the social impacts of AI systems. Amid promises and warnings that AI will end work as we know it, lead to waves of political misinformation and propaganda, entrench existing discrimination and biases into code, and possibly threaten all human life—institutions of all sorts are holding dialogues about AI ethics and governance. Given the massive investment in both AI companies and the data center and computer hardware companies that enable them, it’s unclear whether any institutions have the power to halt a societal adoption of powerful AI systems, but it does seem likely that we will have a more thorough and engaged conversation about the promise and problems of AI.
It is helpful to contrast this approach to the shift from broadcast to participatory civic information architectures, where conversation only moved from expert dialogue into the mainstream a decade into a social media revolution. Social media had a long honeymoon, with enthusiasm over a “Facebook revolution” in Egypt counterbalancing emerging concerns about pervasive surveillance, political polarization, and harms to individual users. As legislators consider bills like the Kids Online Safety Act (KOSA), and Australia and France restrict social media to users over sixteen years old, there is a sense that the horse has already left the barn.62 Facebook is more than two decades old. An entire generation has grown up in an information environment shaped by social media. There may still be time for a debate about whether we want an information environment in which tools like OpenAI’s recently shuttered Sora—a text-to-video model and social media app based on AI-generated video—are an acceptable part of the civic information ecosystem.63
We can see the early signs of civic-participation efforts to ensure that AI adoption considers social impacts. For instance, the State of New Jersey employed the online platform All Our Ideas to collect residents’ concerns and hopes about AI systems in the workplace.64 The output was a rank-ordered list of concerns that citizens wanted to see addressed as the state set AI policy.
There is also the intriguing possibility that AI may open new channels for participatory politics. Media scholar Eric Gordon has pointed out that governments collect an enormous amount of data from citizens about their views and preferences, but that this input is rarely central in shaping policy decisions. Gordon believes AIs may help governments through “generative listening,” turning huge datasets and submissions in multiple languages into insights that could make government more responsive to public input.65
For AI to have a positive role to play in a civic information environment, it needs to be trusted. So far, with AI chatbots that have been trained on pirated materials, that are prone to hallucinations, and that often harvest user data, trust has been hard to find. Building conversations around AI and the information environment proactively, not reactively, will be critical in realizing an ecosystem in which AI features important human voices, rather than drowning them out.
In writing Our Common Purpose, we knew we were addressing America at a moment of profound political upheaval. We took advantage of that moment to introduce a range of interventions, many of which had already been discussed and debated by academics and policymakers over the course of years or decades. Most of the proposals we offered for the information environment had a much shorter lifespan, in part because problems like the politicization of the Supreme Court have been understood for much longer than we’ve understood how the rise of social networks would damage the advertising business of local news organizations.
As a result, our strategies and recommendations may have been less forceful than the times demanded. It is not clear that we could have successfully advocated for a media bargaining code, as in Canada or Australia, nor could we have known that funding for local social networks that operated like public media would have been cut under the second Trump administration. But the ambition of proposals may signal the seriousness of the problems we face, and those problems are profound.
We are fortunate that the rise of AI, with both real economic implications and overheated hype, is spurring public dialogue about how we, as a society, should govern our technologies. Alongside conversations about existential risk or universal basic income to compensate those who lose their jobs to chatbots, we need a robust conversation about the future of our information environment. The transformation a company like Google has had over our understanding of the world, through controlling what we can discover via search and by damaging local journalism by claiming its primary revenue stream, may well be exceeded by the transformations OpenAI or Anthropic bring about.
The idea of sovereign AI, in which nations assert the need for AI as a government-funded public good, offers a suggestion. Democracies need a robust information environment, and AI threatens to concentrate that power in the hands of a small number of venture-backed for-profit companies. Before AI transforms news reporting, information gathering, and political debate, we need to take seriously the idea that we must either ensure these new companies act in the interests of a robust and competitive democracy, or we must aggressively find ways to create our own citizen-led information infrastructures.