Can AI Improve Democracy?
Liberal and social democracy are in crisis. Democracy has always been in some crisis over the last century, but this time is different. Fewer people today participate in or trust the institutions of democracy – civic associations, parties, and unions – and more have tuned out politics or turned to strongman fiefdoms and far-right cults. In a global survey in 2023, the Pew Research Center found that almost two-thirds of people in 24 democracies believe their political systems need to be completely reformed. Support for representative democracy is sharply down since 2017 in half the countries surveyed and only higher in three countries (Brazil, Mexico, and Poland). Democracy is a very recent experiment in human governance and may be about to undergo a phase change for better or worse.
The purpose of IEET and its technoprogressive lens is to be as balanced and honest as possible about the risks and costs of technological innovation while proposing democratic and egalitarian techno-social policies. We have often used "democratic" as shorthand for all the rights and institutions we associate with liberal and social democracy, so the crisis of the democratic social order and its relationship to technological change is one of our central concerns. Specifically, we are interested in the risks and benefits of automating the political process and administrative state.
The 2023 Pew survey also found that support for technocracy, a system where experts, not elected officials, make vital decisions, has increased significantly in most countries since 2017. Algorithmic governance can be considered a form of technocratic rule, although it serves a democratically accountable state. Back in 2014, an IEET affiliate scholar, John Danaher, began addressing the issue of rule by algorithms, and in 2016 wrote "The Threat of Algocracy," which identified the "increasing reliance of public decision-making processes on algorithms" as inevitable, and a threat to the legitimacy of the state. In 2017, I also published "Algorithms and Posthuman Governance," which argued for the inevitability and desirability of automated state functions and called for software innovation for democratic deliberation and citizen activism. (See Sara Hinkley's 2022 report on "Technology in the Public Sector and the Future of Government Work" for a review of many examples of automated government functions.)
Since then, "algorithmic governance" studies have snowballed, including research on criminal justice, public sector automation, firm management, civic participation apps, "smart cities," and the dataification of society. There are algorithmic governance books, issues of journals, and an Algorithmic Governance Research Network "dedicated to understanding the social implications of artificial intelligence, big data, platform capitalism, datafied knowledge, and automated decision-making on society."
I will review the risks and benefits of automating the functions of the democratic state later. I'm also looking forward to reviewing García-Marzá and Calvo's new Algorithmic Democracy, which proposes and critiques ways that democracy could be improved with digital tools, including replacing politicians with algorithm-based decision-making, having one's digital twin participate on one's behalf, the creation of digital platforms for citizen participation and training machines to be morally intelligent.
This essay focuses on how automation can improve citizens' access to trustworthy information, connect us with other citizens to organize for political influence, and ensure that the democratic state reflects all citizen voices. Then I propose that outsourcing the time-consuming tasks of citizenship to our digital twins would greatly expand our political capacities, and that making superintelligent AI a public utility or publicly owned infrastructure would greatly expand our capacity for democratic design.
Let's take a look at some of these issues and ideas.
Who Should I Trust To Tell Me What Is Really Going On?
People increasingly rely on social media and search algorithms for political information, and many observers believe these filter bubbles and echo chambers contribute to political polarization and misinformation. Social media companies have been under pressure since 2016 to implement automated filters for misinformation and harmful content, leading to AI content filtering itself becoming politicized. While the Pew Research Center found that Americans are equally divided and confused about the use of AI to detect and eliminate disinformation on social media sites, Republicans are much more suspicious than Democrats of automated moderation and of the news media in general.
What I imagine as a helpful innovation for citizen news is an AI assistant as epistemically sophisticated as the Ground.News website, which aggregates stories flagging which are from Left or Right media sources and which stories each side is ignoring. Our news apps could learn our preferences as the Google News app does and add disinformation flags like the spam warnings on our phones or emails or community notes on Twitter, flagging suspicious content or providing context. It's not that I aspire to be epistemically centrist, but ideally our personal AI agents will be able to understand the information terrain and apply our interests and biases to show us appropriate content. While some people may choose not to filter out misinformation, research suggests that many citizens don't have the media literacy and knowledge of current events to recognize fake news and would appreciate some algorithmic guidance.
How Can I Connect with Like-Minded Citizens for Activism?
In the 1970s, the American conservative activist Richard Viguerie pioneered mailing lists and direct mail to begin building the American Right's army of donors and voters. Since then Americans have seen many iterations of technologies praised as new methods for citizen engagement, from Howard Dean's experiments with viral video and meet-ups in 2003 to widespread activist and party access to tools like Nationbuilder or NGPVan, integrated platforms for political fundraising, compliance, field organizing, and networking. These tools have become more sophisticated and widely used even as party (and union and church) identification and membership have declined in the U.S. and most democracies.
There is little evidence that electronic methods of citizen organizing are adequate substitutes for face-to-face political parties and institutions, at least not yet. For instance, the Italian Five Star party, the Spanish Podemos, and several Pirate Parties have experimented with generating party platforms or legislative votes by having all members participate in online debates and polls. The results have been mainly an incoherent failure. (See Gerbaudo's The Digital Party: Political Organization and Online Democracy for an overview.) It may be that these tools are better at consolidating the supporters of single issues into campaigns than aiding the tedious work of bringing twenty party factions together around a party platform, although Novelli et al. proposes that "intraparty democracy" could also be improved with real-time sentiment analysis.
But new tools with more powerful capabilities are arriving every day. In "A systematic analysis of digital tools for citizen participation," Shin et al. review 116 digital democracy tools, from mapping your neighborhood and collecting supporters and funds to collective decision-making. See DemTools for a real-time directory of such tools and the use of commonly used technologies for political organizing is reviewed in University of Southampton' "Technologies for Democratic Deliberation." The Lorentz Center's white paper on "Enabling the Digital Democratic Revival: A Research Program for Digital Democracy" is also an excellent review of tools, benefits, and concerns.
One tool that has received much attention because of its use in Taiwan to determine gig work policies is the deliberative platform Pol.is. Pol. categorizes comments and commenters into like-minded groups and incentivizes efforts at consensus. (See Duberry's 2022 Artificial Intelligence and Democracy for his review of the risks and promises of AI-mediated citizen-government relations.)
How Can I Know I Am Being Represented?
People want to believe that democratic institutions are listening to them and reflecting their interests, and in theory, information technology can allow citizens to directly interact with the state in new ways. Since the voices of the affluent and corporations are much louder in democracies, however, we need digital tools that can even the playing field and ensure that all voices are heard.
Unfortunately, digital tools for citizen-state communication are still more useful for paid lobbyists and the already powerful than citizens. Sanders and Schneier mused last year about how lobbyists, or the interests that hire them, could use AI to supercharge their lobbying and influence campaigns. AI can monitor the legislative process and make recommendations about when and what pieces of microlegislation would be beneficial. Chatbots can spit out legislative texts that are optimally vague or precise and tailored to be inoffensive. Lobbyists can use AI to generate boutique quantified legislative impact assessments or map out a campaign strategy with recommendations for how to sway each legislator. While Sanders and Schneier recommend more regulation and transparency of lobbying, the real message is that political influence still depends on money, staff, and tools, which corporations and the affluent have more of than ordinary citizens. (Schneier and Sanders are working on a book on AI-assisted politicians, legislators, administration, law, and citizens.)
While democratizing political influence over the state is a tall order, ensuring that governments at least hear a broad and diverse range of voices in response to policy is something AI could be great at. For instance, Sanders and Schneier point to automated sentiment analysis of public opinion to help policymakers gauge citizen concerns and views in real-time. Content analysis of citizen comments could be used to identify the principal positions on a topic and knowledgeable advocates for each side. Algorithmic tools can also be applied to speaking with every citizen quickly and in their language and improving faith in electoral representation by drawing fairer electoral boundaries or flagging potential electoral fraud. Citizens can be engaged in online debates and decision-making around municipal budgets, as has been done in Chicago, Brazil, and Finland. As Marco argues, however, digital democracy applications inevitably require trade-offs between making choices coherent and manageable (A vs B), versus making the process more egalitarian and inclusive (A to Z, depending on fifty factors).
Citizen Apps in ExoCortex and Cognitive Enhancement for Citizenship
In the 1960s, the U.S. Students for a Democratic Society briefly adopted the ironic slogan, "Freedom is an endless meeting," anticipating how prohibitively time-consuming participatory democracy is. But we could participate in more decisions if we could outsource our participation to a trusted digital twin. Such a technology is proposed in Alistair Reynolds' space opera universe, Revelation Space, which includes a civilization of anarchists or radical democrats called the Demarchists. Citizens of the Demarchy all have neural implants with agentic features that participate on their behalf in a constant stream of political debates and votes. Ideally, a digital citizen twin could constantly translate your mind, interests, and values into political communication and action. A few people have begun working on this, such as this paper on using BCIs to vote.
Short of using a BCI for this task, which is probably decades away, we could start training our digital citizen AI agents through glasses and wearables, our "exocortex," now. For instance, Deseriis has a nice review of design decisions to reduce the user burden in (exocortical) digital democracy apps, and a team in Helsinki has an excellent paper proposing building "citizen digital twins" based on streams of government data about us, as a way to reflect on life and goals, interact with state services, and predicting needs.
Can Democracy Be Improved with Corporate Tools?
The regulation of disinformation and social media that began in 2016, the European AI legislation, and U.S. infrastructure spending to re-shore chip manufacturing are all examples of the democratic state placing parameters on corporate technology. Once we have more experience with digital democracy tools, regulatory standards could require that our operating systems include features like fact-checking, secure voting, or citizen activism. But if the most optimistic projections are correct, we may arrive at artificial general intelligence or superintelligence soon, systems that some believe will be so impactful and dangerous that they will be nationalized or regulated by public utilities. (See also Christiaens' "Nationalize AI!" and Verdegem's "Dismantling AI capitalism.”) Since these systems are built partly from the accumulated work of everyone who has written or recorded anything, there is already a case for partial public ownership. A publicly-owned ASI would allow experiments with 21st-century democracy (and authoritarianism) that we currently can't imagine.