Tomorrow’s Democracy Is Open Source


Nils Gilman is a deputy editor of Noema Magazine and senior vice president of programs at the Berggruen Institute.

Ben Cerveny is co-founder and president of the Foundation for Public Code and a senior fellow at the Berggruen Institute.

Much time today is spent discussing the threats technology poses to democracy rather than the opportunities. It’s not hard to discern why: Social media has created new vectors for bad actors to spread disinformation that rattles around uncriticized within echo chambers; algorithms increasingly grab our attention by focusing on the most extreme messages; and AI threatens to supercharge both of these phenomena with a dash of embedded bias. These problems have provoked numerous, yet so far largely fruitless, calls for regulation of these technologies.

To be sure, technology needs more regulation. But instead of treating technology only as degrading democracy and civic participation, we can also cultivate it as a positive tool that can enhance democracy. Imagine, for example, that instead of just voting every two or four years, or even participating in the occasional citizens’ assembly or serving in government via sortition — effectively a form of jury duty for policy-making — that every constituent was able to express their political opinions on every public policy topic, continuously, and could do so based on good information? Such a system wouldn’t be just a tweak to the existing democratic practices or a return to time-honored methods; rather, it would represent a revolution in the very nature of democracy itself.

That such a renewal is needed is scarcely in doubt. The anger that suffuses so much politics today is directly connected to the sense among constituents that their governments are not responsive to their concerns. Many democratic theorists and reformers today talk about finding new ways to encourage political “participation without populism.”

The need to improve the responsiveness of our democratic systems of government could scarcely be more urgent given the massive policy-making challenges —  from greening our energy systems to managing the integration of AI into our economies — that we will face in the coming decades. A technologically enabled form of continuous democratic engagement offers the promise of a government that is simultaneously more effective, more efficient and more directly responsive to the will of the public.

The Wilde Conundrum

“The trouble with socialism,” Oscar Wilde is often said to have remarked, “is that it takes too many evenings.” What Wilde meant was that any form of participatory governance requires people to invest free time they might use for more personally urgent or entertaining purposes than digging into the details of local zoning laws, public school administration or utility regulation. For many citizens, the prospect of being expected to be continuously well-informed and engaged in government decision-making sounds a bit like a mandate to exercise more and eat less. Sure, it might be good for you, but it’s not much fun.

Since the dawn of mass-franchise democracy, policymakers and politicians have struggled with how to engage constituents who don’t find politics a worthwhile use of their time. When the political scientist Samuel Popkin popularized the phrase “low information” voter in 1991, he underscored that for most voters, making “low information” decisions about who to vote for was ultimately “rational,” since the amount of personal effort required to understand the nuances between different candidates or policy choices, multiplied by the low likelihood that this effort would make a significant difference in one’s opinion, was simply not a good use of one’s time. It was more worthwhile, Popkin argued, for voters to rely on “heuristics,” or their own more instinctive judgments.

For example, rather than invest time and energy trying to differentiate politicians’ policies, voters could simply observe Bill Clinton’s penchant for McDonald’s and know he was a man of the people compared to the patrician George H. W. Bush, who reportedly did not know what a barcode was. At the limit of this logic, economist Anthony Downs noted in 1957 that it is “irrational” to vote at all, given the vanishingly low likelihood that your vote will decisively impact the outcome. In other words, what these social scientists are observing is that the amount of effort a voter needs to put in, in order to make a difference in democratic practice, is often much larger than the likelihood that any such effort will bear fruit.

Poor democratic engagement isn’t just about low-energy or low-information voters. It’s also a result of low-information political representatives: politicians who, even when they’re in good faith trying to represent the interests of their constituents, simply don’t know and have limited ways of finding out what their constituents really want.

“The need to improve the responsiveness of our democratic systems of government is especially urgent given the massive policy-making challenges — from greening our energy systems to managing the integration of AI into our economies — we will face in the coming decades.”

In particular, politicians of mesoscale jurisdictions — for example, cities with 50,000-500,000 people — often represent more people than they can possibly meet and know in person, but typically lack the resources to do formal polling.In Santa Monica, California, for example, each city councilmember represents around 13,000 constituents and makes roughly $21,000 a year, about the median for city councilmembers across Los Angeles County’s 88 jurisdictionally distinct cities.

In cities of this size, there are rarely resources for polling the public policy questions that come before the city council. So councilmembers rely on heuristics — like the clamoring of people at city council or school board meetings — to determine the preferences of their constituents, even though everyone knows how unrepresentative these meetings usually are. As a result, politicians in these mesoscale jurisdictions are typically almost as ignorant of their constituents’ true aggregate preferences as the constituents are of the stakes (let alone the details) of the policy questions under debate.

Here is where technology can help.

From Policy Code To (Iterated) Software Code

Perhaps the biggest hurdle to evolving the relationship between constituents and active government is the sheer magnitude of continuous information it generates. Overwrought legal language, byzantine proceduralism, and frustratingly inaccessible documentation all render participation nearly impossible for most working people.

This very complexity, however, represents an opportunity for technology to provide tools and platforms to transform the nature of the democratic process. Through summary synthesis, data visualization and simulation of legislative consequences, we can build a system of political engagement that provides real-time interrogable, explorable models of the inputs and outputs that make up the legislative decision-making process. We can use software code to enable new forms of participation in the creation of our legal code. 

In many ways, the process of generating civil code is similar to that of generating software code. Some of the tools that software engineering has evolved to help developers understand and contribute to giant fast-evolving codebases might now be adapted for the domain of legislative ordinances and the generation of civil code. For instance, a technique called continuous integration provides an orchestrated set of roles and intuitive participatory interfaces that enable multiple software developers to contribute simultaneously to a fast-evolving code base.

Such a technique can quickly resolve conflicts and merge different versions of a functional document together quickly and seamlessly. An analogous type of legislative adaptivity will be necessary as society confronts rapid discontinuities in the coming years, from climate instability to epidemics to the use of disruptive technologies themselves. To steer between the Scylla of authoritarianism and the Charybdis of decision-making sclerosis we will need to remake our political processes so that they can dynamically integrate public opinions in a rapid decision-making context.

Some practices in the legislative system could benefit from simple techniques included in the software development process. In municipal bodies, “codification” involves translating bills and ordinances voted on by the city council into changes in the civil code. Such changes to the law can be seen as, in effect, updates to civil society’s “operating system,” but are frequently opaque and difficult to track down.

In a continuously integrated code system, however, you could trace any passage to its origin “change order” — or vote, in this case — and discover the circumstances around how the piece of code came to be. Imagine a process for rich hypertext referentiality and graph visualization of the areas within the corpus of civil code that each ordinance or bill passage impacts, that allows constituents to understand how, for instance, a section of the building code came to be and the circumstances that led up to any changes.

A crucial component of a democratic process of civil code development is that citizens get an opportunity to weigh in before a vote. In the United States, this general idea is so important that it is enshrined in the First Amendment to the U.S. Constitution under the right of citizens to petition the government.

“A technologically-enabled form of continuous democratic engagement offers the promise of government that is simultaneously more effective, more efficient and more directly responsive to the will of the public.”

In state and local government, this same principle is embodied in analog form when there are legally required “public comment” periods for proposed legislation. To date, however, there have been no serious efforts to automate the integration of constituent opinions into the civil code production process. But with large language models (LLMs), or artificial intelligence able to process mass amounts of data, this may be about to change radically.

Introducing ‘Open Insight’

To enable experimentation with dynamic representation of constituent preferences in a more real-time, legislative decision-making process, the Berggruen Institute is prototyping a set of open digital tools to help legislators and constituents understand civil issues and communicate more effectively. We’re calling this proposed platform “Open Insight,” and are designing it especially for use by the sorts of mesoscale municipal governments perhaps most in immediate need of such software-enabled constituent participation mechanisms.

The first part of this application toolset we will prototype is about aggregating and making sense of the history and future of decisions that a legislative body in a municipality has made or will make. City councils frequently publish meeting notes online. The tools we are developing create graph databases of this deliberative and legislative timeline by using machine learning to review such “meeting minutes” and to construct an ontology of the activities and subjects across the past and future of any changes to the civil code, allowing us to create an interconnected map of the decisions over the course of the municipality’s history.

The concept is to create a richly explorable, interconnected map of political action committee activities, the debates and decisions of policy-makers, the press around these decisions, as well as the public’s response in various fora prior to and after any changes. Additionally, and perhaps most importantly, the tools will make it easy to see each ordinance or law that passed, how it changed the civil code, and thus to understand how it has operationally changed the way the city is run. For each paragraph of the civil code, there will be a history of how it came to be when it was modified, by whom and what the context of the vote was at the time.

For example, if a decision about what the city should do with a newly acquired piece of land was coming before the city council, the application might present and synthesize to users the land use history of the specific site, the various proposals that had been laid out in the media, and what had been decided about similar parcels in nearby jurisdictions. The relevant context would need to be determined on a case-by-case basis, likely by civil servants, perhaps with assistance from LLMs.

An example of what such an open-source publicly available app might look like. Matt Irvine Brown for Noema Magazine

These interconnected graphs of legislative activity can also serve as input vectors for LLMs for the continuous transformation of any legislative code. LLMs like ChatGPT, Bard and others, could potentially be made available as companion bots to help explore the legal and cultural context of past decisions and upcoming votes, and perhaps even help constituents draft new bills and ordinances based on precedent.

Central to the inquiry of our prototyping is how the function and provenance of these tools can be made transparent and how they can be tested for accuracy, bias and misinformation. Solving these problems is perhaps the largest hurdle to the potential uptake of these tools as services offered by public bodies themselves.

The second part of our prototype stack uses this data graph to present a new type of continuous engagement experience for constituents. Our initial goal here is to create a lightweight and accessible mobile app for residents of a municipality that enables them to see what has been and will be voted on by the legislature, to track what they care about and signal their preferences on each issue in an aggregated and anonymous way to their political representative.

“Our initial goal here is to create a lightweight and accessible mobile app for residents of a municipality to see what has been and will be voted on by the legislature, to track what they care about and signal their preferences on each issue in an aggregated and anonymous way to their political representative.”

We hope to iterate on this experience as fast as possible and make it highly configurable to suit each unique political context and allow for experimentation with techniques like data visualization and machine learning. We are currently open to conversations with municipal legislators who want to participate in this collaborative prototyping and feedback process.

The semantic graph underlying the decision history tools allow the constituent to explore the timeline of past and future voting issues on ordinances 2744 and 2745, for example, by topic and to filter by specific political actor choices. Matt Irvine Brown for Noema Magazine

As a constituent exploring the issues, you will see how your councilmember voted on prior ordinances and can also signal your own position on any proposal, which will be aggregated into an anonymized data visualization that the councilmember can use as community input. There will need to be public discussions about whether the results should be sortable based on demographic data about the constituents, or if such data should be collected to begin with.

Matt Irvine Brown for Noema Magazine

On the one hand, though the current system is not very representative as it is, without such data from the app, politicians might doubt the representativeness of the expressed opinions; on the other hand, such demographic data, if sufficiently granular, might compromise the anonymity of the constituents. Likewise, different communities will have different preferences about whether the results should be made available only to the politicians or to the public at large.

This is a potential feedback dashboard that legislators could consult, which graphs constituent input. Matt Irvine Brown for Noema Magazine

A New Model For Democracy

Over time, we can imagine Open Insight might become good enough at learning about constituent preferences that it could provide recommendations to them about how they might feel about issues being debated in their city council. This vision opens the wild possibility of each of us having a “personal political avatar,” that is, a continuously updated digital representation of our political preferences, based on our formally expressed and perhaps now implicitly understood political opinions.

This political avatar might “advise” you as to your likely thoughts regarding an upcoming vote on a matter before your city council and also explain to you why this is. Of course, upon seeing its rationale, you might decide that in fact this is not your opinion on a particular matter of public policy and choose to express the opposite opinion. Your political avatar would then update its understanding of your preferences. Over time, your avatar would learn more about you, and its ability to anticipate your opinions would become increasingly accurate and thus labor-saving.

Needless to say, this last vision of a radically new form of digitally enabled democratic participation raises a series of potentially alarming new challenges for how we need to regulate our democratic practices, including questions that are simultaneously technical and ethical. Who will control and vet this codebase? Should the preferences expressed by constituents through the app be made public? What sort of obligation should they impose on the reelected representatives who are ultimately responsible for the decisions? How do we ensure the privacy of those expressing their preferences through the Open Insight app?

The privacy of the formal ballot in voting is a venerable tradition in most democracies — how does that translate into online expression or pre-voting, during the legislative process? Who will ensure both the accuracy of the anonymized data and that it isn’t breached or altered in some way? These are all questions that are central to any form of democratic practice, but which will only be intensified in the technologically enhanced form of democracy we are imagining here.

Likewise, if the app includes a recommendation engine that is continuously learning each constituent’s specific political and policy preferences, how do we ensure that the algorithm is itself transparent and that users do not come to over-rely on these recommendations, giving up agency to the app even as they regain it from their representatives? Should the constituents’ “votes” be binding on policymakers or simply serve as recommendations for them to consider?

To begin with, we believe that there should be recommendations for elected officials to be able to gauge public opinion, but if the app becomes transparently governed and effective enough, possibilities for political disintermediation might arise, such that technology could enable a new form of direct democracy.

“Built and deployed as open source public code, an app like Open Insight can be designed to ensure both transparency and guardrails against the worst outcomes.”

Such an app, if deployed at scale, might change democracy in more profound ways. Imagine, for example, if the app achieved the following three adoption milestones: (a) 100% adoption by constituents in a given jurisdiction; (b) that constituents are so satisfied with how well their “personal political avatar” represents their interests that they automatically hit “approve” on all recommendations; and (c) political representatives become so confident that these expressions of opinion accurately represent the Rousseauvian “general will” that they automatically defer to them.

Taken together, you end up with a political decision-making process that is “radically democratic” (in the sense of being responsive to the general will) but paradoxically also one in which humans have been completely removed from the decision-loop of the policy-making process. In this scenario, policymakers would merely propose laws, submit them for evaluation by their stakeholders’ personal political avatars and then implement what gets approved. In other words: an automated form of direct democracy.

Whether you consider such an endgame a dream or a nightmare, however, is not a good argument to avoid building such an application. Rather, it is an argument in favor of building it and deploying it as public code. It is next to inevitable that something like this will eventually be built, but the choice, societally, is whether it be built and owned by a private vendor whose incentives will likely be more profit-driven and proprietary in nature, or whether it is ultimately owned by all of us.

Built and deployed as open-source public code, an app like Open Insight can be designed to ensure both transparency and guardrails against the worst outcomes. What might such guardrails look like? It might be requiring users to review relevant materials before expressing their opinions, listening to both proponents and opponents of a particular piece of legislation, or coupling such software-powered harvesting of opinions to other participatory democratic reforms such as sortition and citizens’ assemblies.

Part of the reason we are looking to engage with software domains that address the core functions of democracy is to underscore the importance of literacy in and governance of such potentially transformative tools, which are now being generally referred to as digital public infrastructure.

We believe that software that serves the public interest deployed as a public service — in the way that Open Insight might be — should be conceived of as public code, an approach to the production of software that is not only situated in traditional open source principles, but further mandates explainability, good governance, sustainability, and accountability, among other criteria spelled out in the Standard for Public Code.

This standard, and an accompanying approach to collaborative codebase stewardship, have been developed by the Foundation for Public Code, a Netherlands-based, nonprofit association that helps public administrations around the world build open digital infrastructure together.

International collaboration on the development of such large-scale open Digital Public Infrastructure has already yielded projects like MOSIP, India’s digital identity service that gives nearly one hundred million registered users access to government services, and LEOS, a project of the European Commission to provide collaborative online editing of proposed legislation that is used in the European Parliamentary process, as well as among member states.

The Digital Public Goods Alliance, which is endorsed by the United Nations, identifies a whole range of open digital tools and platforms that can be deployed by cities and states to advance sustainable development goals, like operating school systems or building out transit networks.

As we develop prototype projects like Open Insight, we hope to eventually convene an ever-expanding network of cities and states working collaboratively to test and build highly functional systems and tooling. This ecosystem of open distributed procurement will create normative standards for processes like those of continuous political participation. An institution that wants to offer a Digital Public Infrastructure solution can produce a “reference implementation” of a proposed tool or system — that is, a program that implements all requirements from a corresponding specification and thereby serves as a model for others.

This initial implementation can then be iteratively prototyped in open collaboration with implementing partners. If successful, this solution can become public code, with an open license, a collaboratively generated governance model and a technical roadmap of potential improvements that is continually implemented, maintained and evolved by a thriving community of public organizations.

“Technological tools like continuous political participation may help us become more responsive to fundamental societal questions as a culture, without compromising the core values that define our civilization.”

Moving from a democratic system where voting occurs once every couple of years after which decision-making is completely delegated to the elected officials — essentially our current model — to one based on continuous technologically enabled engagement with the nuts and bolts of government decision-making, represents a radical revision to our standard assumptions not just about how democracy works but about what democracy even is.

The model of democracy envisioned here involves going far beyond the vision of democracy promoted, for example, by the Open Society Foundations (OSF), that is, a vision of democracy centered on political parties competing in free and fair elections, with peaceful transitions of power from one set of elected representatives to another. That OSF version of democracy is of course vitally important, and the vision we propose here does not displace it, but rather supplements and enriches the traditional model of representative electoral democracy with a technologically enabled system for the continuous expression of political sentiments and opinions.

As we head deeper into the intensely nonlinear 21st century, with its rapid onset of epidemics and climate catastrophes, it is clear that human society and governance must become adaptive at a higher rate than we can currently achieve. We must begin to understand the possibilities of technological tools like continuous political participation to explore how to become more responsive to fundamental societal questions as a culture, without losing our integrity to the core values that define our civilization.

How can we give the Oscar Wildes of today back their evenings while keeping our institutions and planet intact? We can do this by using the process of prototyping participatory software itself as a site for reimagining practices of democratic deliberation. In this way, software design methods based on iterative experimentation, collaborative understanding, and continuous integration of multiple points of view can serve as a model for a dynamic new concept of democracy.