Print Friendly and PDF
only search openDemocracy.net

Liquid democracy, its challenges and its forebears

Until today, particular decisions within software and product development have to be decided by privileged people. LiquidFeedback can democratize the decision-making process. An interview with the founders.

German Pirate Party Federal Congress, 2012. German Pirate Party Federal Congress, 2012. Demotix/ Gregor Fischer. All rights reserved.Whereas a growing amount of decision-making software is currently in use in the political arena, LiquidFeedback's distinctive feature is the possibility for users to delegate their vote to other users by topic. Rather than assuming that all participants are equally knowledgeable and equally invested in every political issue, Liquid Feedback (LF) lets them decide whom to delegate on specific initiatives. Those who hold proxy votes can in turn transfer them to other delegates, facilitating the emergence of networks of trust. Such trust, however, is not a blank check as proxies can be revoked at any given moment. The fluidity of the delegation process implemented by LF is an emerging political protocol, whose roots lie in the decentralized nature of the Internet. As we will see in the following interview, the authors of LF see their software as a concrete instantiation of the idea of liquid democracy, which allows individual constituents to retain their prerogatives without compromising effective decision-making.

Another far-reaching property of LF is that the platform does not allow for the use of secret ballots. In order to ensure transparency and protect against electronic frauds LF implements a voting system that is recorded and verifiable by anybody. The public nature of voting, however, comes at a cost. Because in modern democracies the privacy and anonymity of voting are considered essential to protect individual autonomy and freedom of choice LF is not suitable for consultations where secret voting is desired or required. Yet this limitation has not prevented the German counties of Friesland and Rothenburg and the cities of Wunstorf and Seelze from adopting the software to consult their citizens on a wide range of issues.

For the LF developers, the transparent and public nature of Internet voting is a necessary condition for implementing a system that can be trusted. Even though a country like Estonia has already adopted Internet voting in binding elections, security experts have voiced significant concerns over the trustworthiness of e-voting systems. This does not mean that voting over the Internet does not have its uses. Provided the public and verifiable nature of voting, LF's authors believe that their platform can be adopted to empower political party members, improve internal decision-making in parties and CSOs, and thrust issues rather than leaders into the center of the political process.

And yet LF’s adoption by political parties has yielded mixed results. The most notable case is that of the German Pirate Party. The Berlin chapter of the GPP enthusiastically adopted the platform shortly after its release, in late 2009. Even though Behrens, Kistner, Nitsche, and Swierczek initially joined the Berlin PP, they soon found themselves in the midst of a political controversy over the public nature of voting. Because the Berlin Pirates initially tested LF for nonbinding consultations, they allowed members to use pseudonyms rather than their actual names. “Only very few actually used this possibility,” remembers Nitsche, “but this legacy turned into a major problem once the system was to become more binding.” Initially the Pirates made a productive use of LF for developing the program for the 2011 Berlin elections and won an unexpected 8.9% of the vote. But the unresolved question of how to verify the vote—and hence the identity of voters—became a burden not only for the Berlin chapter but also at a national level. Even though the Pirates made several attempts to resolve the issue, they failed to reach the 2/3 super majority necessary to change the party statute and establish a permanent online assembly ("SMV") based on LF, which would have been able to make legally binding decisions.

Frustrated at the Pirates’ impasse, the four developers decided to focus on the development of the software and to leave the party in January 2011. Since then, the four have continued to develop LF and to promote digital democracy through the Association for Interactive Democracy, which has held workshops in Burma, Pakistan, Georgia, and Colombia. In 2014, the four co-authored The Principles of LiquidFeedback, a book that details the design principles, voting theory, and political philosophy behind the software.

In this interview, Behrens, Kistner, Nitsche, and Swierczek reflect upon the origins of Liquid Democracy, analyze the forms of leadership that emerge through LF, and discuss some of the design principles that have inspired them. Marco Deseriis.

Marco Deseriis (MD): Let us begin from the origins of the idea of Liquid Democracy. To my knowledge, this concept was first formulated in the year 2000 by John Washington Donoso (aka Sayke) in online forums that are no longer online. Donoso claims that he designed Liquid Democracy as a “knowledge sorting system” that could recommend answers to a question based on the shared knowledge of the community. In his mind, LD was meant to determine the best answer to a common problem--e.g. how to maintain civil infrastructure. Sayke also claims that because answer recommendation does not force a community to choose the most recommended answer, it is fundamentally different from an LD system based on proxy voting like LF. In his words, “rather than a mechanism through which we are informed by others, vote proxying (and traditional democratic representation) acts as a mechanism through which we cede power to others.” Do you agree with Sayke that his idea of LD is fundamentally different from yours?


Björn Swierczek (BS): So far as we know, the idea of Liquid Democracy dates as far back as Lewis Caroll’s Principles of Parliamentary Representation (1884), a short book that puts forward the idea of delegate voting in a modern democracy. However, this idea could be utilized and implemented in many different ways for many different purposes. In the text you cite, Sayke says the very same thing: “Liquid Democracy stands as an alternative to direct and representative democracy, but they each can be implemented in 17 hojillion ways - all kinds of voting systems can be designed which use direct, liquid, or representative democracy.”

In 2009, we learned that no useable implementation of Liquid Democracy was available. To give the idea of Liquid Democracy a chance, we started creating LiquidFeedback to actually allow people to use it. As we had political parties in mind, the main task was to create a structured discussion process and the possibility of making solid decisions. Therefore, we created the unique LF proposition development and decision-making process, including predictable scheduling of the four phases of a decision (admission, discussion, verification and voting) and a hierarchical structure of units, areas, issues, initiatives, and suggestions. Furthermore, we created the Harmonic Weighting algorithm to ensure a fair share of representation for minorities by allocating display space. In addition, LF utilizes further algorithms to process the users actions accordingly and features a clone-proof preferential voting system. 

Sayke's paper on how he would like to utilize the idea of Liquid Democracy for answer recommendation is something very different. LF goes far beyond the simple dream of Liquid Democracy and provides a feasible implementation for proposition development and decision-making. 

MD. Still, the admission and discussion phases of LF are a “knowledge sorting system” of sorts in that they empower users to find out for themselves what the best proposals might be and how they might be grouped. Can you explain how knowledge is sorted in LF, especially when the user/voter might have to make a decision among a number of competing initiatives on the same topic?

Andreas Nitsche (AN). Because knowledge representation will always be biased—i.e. knowledge sorting is never entirely neutral—we decided to opt for an initiative-driven approach. Similar to lawyers, LiquidFeedback initiators promote their initiatives by providing the background and the rationale for why people should support them. Whoever likes the idea of an initiative but sees necessary adjustments or potential for improvements can write a suggestion. All participants can assess the suggestion with one click and tell us if they think it is useful or even a necessary condition for their support.

Suggestions are sorted by the Proportional Runoff Algorithm according to the potential positive impact on the degree of support for the initiative. In other words, LF provides initiators with quantified suggestions that are sorted according to their potential impact on gaining a majority. In this way, initiators can gauge possible amendments while retaining the ultimate say on whether to accept a suggestion or not.

When suggestions are not adopted, they can still be integrated within alternative initiatives, which can be started by any participant. The sorting of the alternative (competing) initiatives is done by the Harmonic Weighting Algorithm, which ensures that each interest group gets a fair representation according to its actual size. This protects minorities--i.e., any interest group that does not form a numerical majority for a given issue--by making their point of view adequately visible. At the same time, it protects against the dominance of noisy minorities as they don’t appear larger than they are. 

MD. There is a strong connection in LF between knowledge sorting and the idea of passing on proxies to those who might be more competent on specific issues. Where did these ideas come from?

AN. When we started thinking about developing LF back in 2009 we decided to focus on transitive proxy voting. Transitive proxy voting allows for proportional representation overcoming both the shortcomings of the static representation of representative democracy and the limitations of direct democracy. This is for application fields where a recorded vote is desired - for example, programmatic decisions in political parties. As we considered the creation of a discussion process leading to informed decision-making, one of the challenges was to implement a constructive process that could keep trolling at bay. The classic solution to this problem is moderation but moderation comes with privileges, which usually constitute democratic deficits. To overcome the deficits of top-down or karma-based moderation, we decided for a collective moderation by all participants. Once again transitive proxy voting was the key to actually achieving this goal.

The desire to implement decision structures in a new democratic approach uniting the best of direct and representative democracy was also fueled by the experience of a direct democracy approach of the German Green Party some 30 years before, which unfortunately did not scale up.

MDWas the Green Party's idea of a "non-upscaling” direct democracy ever put in practice?

Axel Kistner (AK.) Like every party with a growing member base, the Green Party of the 1980s needed a delegate system. To solve the problem of over-empowerment of delegates and fixed unchangeable structures, the Greens came up with the idea of introducing a rotation principle so that delegates could not run for office more than once.

They also set up a special representation for women and new members who joined the party from East Germany after the fall of the Berlin Wall. This all happened at a time in which the Internet did not offer technical solutions for remote cooperation and decision-making. Nowadays, the Internet allows for asynchronous collaboration, so the idea that every party member has access and can vote on every topic has resurfaced. In other words, it is precisely because direct democracy does not scale that the idea of Liquid Democracy has to be implemented. This is what LF is all about.

MD. Would you say that LF makes it possible to implement a large-scale participatory democracy that might eventually replace our current representative system?

AN. Not really. While Liquid Democracy can be scaled up, it comes at a price: the vote of every participant is recorded and therefore documented. As far as representatives are concerned, accountability is desired. Liquid Democracy, however, doesn’t differentiate between voters and representatives. A Liquid Democracy society would need to treat every citizen like a representative in the existing parliamentary systems. Furthermore, the system of checks and balances would need to be completely readjusted. It would be irresponsible to give up secret elections – a security mechanism to ensure free elections and protect democracy. This is why we do not endorse calls for replacing representative democracy with Liquid Democracy and conclude: Liquid Democracy provides no alternative to the parliamentary constitutional republic, the presidential republic or the parliamentary constitutional monarchy for that matter. Besides the binding use in political parties and CSOs, it may be used in civic participation as an additional communication channel between citizens and their administration, or in constituency participation for better connecting representatives to their electoral district.

MD. Do you think that political parties should always protect their members’ privacy on certain voting decisions? If so, where do you draw a line between the decisions that should be subject to public scrutiny and those that shouldn’t?

AN. This depends on the situation and the society we are talking about. I would personally wish for a society allowing everybody to express his or her point of view free of fear. But if a party actually feels the need to protect its members, it should not use computers for decision-making as this creates a form of knowledge that can lead to domination. This is a knowledge that is accessible to a few and opens the door for misuse (in the worst case even blackmailing).

MD. Are you suggesting that LD should only be implemented within political formations whose members can firmly trust those in a position of power?

AN. I am only warning against creating a false sense of security. A database containing ‘secrets’ will become a target as soon as interest is high enough--neither the Internet nor any personal computer can be fully trusted. On top of that, all practicable encryption methods have an expiry date, an unknown expiry date, potentially already in the past, which someone may have already exploited, unbeknownst to everyone.

MD. Nicolás Mendoza, a researcher at the Leuphana Inkubator, has described this tension between individual expectations of privacy and public expectations of trustworthiness as the problem of how to respect both the integrity of individuals and of the voting process. In chapter 3 of the Principles of LiquidFeedback you discuss the verifiability of different voting systems, noting that computer systems are more subject to manipulation than traditional voting systems such as voting by show of hands or ballot boxes. Mendoza argues that because electronic voting poses such a threat to both private integrity (anonymity) and public integrity (verifiability) the influence the new decision-making tools can have on the political process is extremely limited. After all, why should citizens rely on systems that, being unreliable, will always have a lower legitimating power—constitutive power, if you wish—as compared to traditional voting systems?

AK. This is a misconception. Electronic voting does not necessarily impose a threat to verifiability. When secret voting is not desired, the public integrity of electronic voting can be even higher than with traditional voting systems. In LiquidFeedback, the proposition development process is fully transparent using recorded vote and can be reviewed by all participants.

As a matter of fact, LF takes “snapshots” of every decision-relevant moment. For example, when an issue passes from admission to discussion, the software archives the supporters including the underlying delegation structure. All participants can review these data at any give moment either manually or using automated checks.

This ensures a credible process with trustworthy and indisputable results. In the instances in which LF can be gainfully used—namely, decision-making in political parties and CSOs as well as agenda setting in civic participation—a credible process can be implemented. You just should not attempt to use computers and the Internet for secret voting.

MD. Even though transitive delegation allows members to revoke their proxy at any given time, LF does not necessarily prevent the emergence of leaders who are considered competent on multiple issues. Based on your experience, how does leadership emerge through this process?

AN. LF’s decision-making is deliberative and driven by initiatives. We acknowledge differences between people and value individual contributions. We also factor in competing ideas. Though we would encourage everybody to cooperate we don’t believe you can always count on cooperation. Members empower each other and the unfolding power structure can be quite stable but will only last as long as those who contribute to individual power are happy with the situation (or more precisely: as long as they don't become unhappy). This also redefines the concept of leadership: less detached and less solitary. In a way we adapt the idea of a multiparty system to the inner-party context. If there are different interest groups they can always work together on one initiative or work on alternatives.

Further, there are different forms of leadership. We can see three types of leaders. Type A are opinion leaders who shape the program of the party but very often don't need the limelight, don’t run for any office and are widely unknown to the public (although the public could know these people if they wanted). (In their function they are comparable to the people referred to as gray eminence in more traditional parties.) Type B leaders are mostly "presenters" who sometimes appear as the actual leaders in the media. And type C would be formal leaders such as a board member or the chairman, mainly responsible for administrative tasks such as registering the party for public elections, campaign logistics, and keeping the member database up to date. All types are important, sometimes overlapping in a person and form a symbiosis. Traditional media reception practices sometimes valorize a combination of types B and C. Only type A emerges in and through LF.   

AK. Leadership follows psychological aspects. Usually human conflicts in the "real world" can be only solved partly by technical solutions and need an agreement of the participants to accept certain rules of procedure provided by a system like LF.

We implemented the four main aspects into the software that are essential for true self-organization in an equal discussion process: 1) Scalability through division of labor (realized by transitive delegations / liquid democracy); 2) Proportional representation of minorities (realized by collective moderation = no moderator / no leader needed); 3) Protection against non-transparent lobbying (realized by a fully transparent decision process); 4) Equal treatment of competing alternatives (realized by preferential voting).

These four essential principles of LiquidFeedback guarantee a fair process of proposition development and decision-making. Even the "delegates" (people who receive many proxies through LF) are not leaders in a traditional sense. The division of labor does not empower people who give voice to an opinion—there are no “opinion leaders.” The status of opinion-leader is a real-world experience based on psychological aspects of life has nothing to do with LF. Any person within the system may lose delegations immediately if he or she starts to do odd things that are not accepted by the participants delegating their vote to them. That is why the system is not static but dynamic ("liquid"). 

MD. Are you suggesting that leadership is based exclusively on the force of an argument? If that is the case, LF would be the ultimate instantiation of the Habermasian ideal of a public sphere based on rational-critical debate. 

AK. Yes, especially Habermas’ "Concept of Political Participation" and his critique published in the book Kultur und Kritik identifies the basic problems of European society. All turmoil in contemporary Europe has its roots in the problems described by Habermas back in the early 1970s.

AN. I am inclined to agree but we do not ignore the existence of interfering interests and dummy arguments. However, these are less likely to harm the debate in a fully transparent process. Likewise we do not think there are only rational reasons for delegations. The reason for choosing a specific delegate is a personal mix of assumed expertise, reputation, trust and sympathy (pretty similar to elections).

MD. One of the arguments in favor of embodied forms of decision-making is that humans need bodily and emotional cues in order to really trust each other. What is your view on this topic?

AN. LF users know each other, can identify each other, and meet in the real world. To make an example, the Berlin pirates organized proposition conferences for their upcoming party convention and focused each event on a certain topic area. Right after the meetings, the impact of these discussions became visible in LF. Real-life discussion formats—e.g. among co-workers, in town hall meetings, in talk shows—have their own pros and cons. They are either limited in the number of active participants or they are moderated, which limits the ability of participants to speak as they see fit. They also usually do not embed a quantification of the support for the points of view being expressed. On the other hand, these comparatively unstructured discussions have a great creative potential. We think it is a perfect symbiosis to see results of all these discussion formats reflected and measured in LF's structured discourse. 

MD. The question of measuring consensus is central to LF. Can you explain how consensus is measured through the various stages of the decision-making process?

Jan Behrens (JB.) LF does not measure "consensus." It measures a collective preference based on the democratic principle of majority rule. As we explain in The Principles of LiquidFeedback, decisions are made by majorities in LF, while every minority is able to put their point of view up for discussion. This allows people to reach a consensus when it is possible [edited]. But demanding a consensus, in the sense of “unanimity” or large supermajority requirements, is undemocratic because the majority can be taken hostage by a minority (which would assign more power to members of the minority than to other individuals). As explained by Anthony J. McGann at the University of California in Irvine in his paper “The Tyranny of the Super-Majority: How Majority Rule Protects Minorities”, only majority rule satisfies political equality. Consensus requirements increase the risk of resentment, hidden conflicts, stagnation, and, last but not least, preservation of existing power structures.

Therefore, LF uses proportional representation algorithms during the discussion of an issue and a variant of the Condorcet Method (the Schulze Method) for its final decision. The proportional representation in admission, discussion, and verification phase enables every minority to put their point of view to discussion. The Schulze Method ensures that minorities cannot enforce the status quo against a majority of eligible voters.

MD. Can you explain how the Schulze method works and why or when is it better than proportional or single-winner voting systems?

AK. The Schulze method as used in LF is a single winner voting system determining the collective preference by pairwise comparison of all alternatives. The Schulze method is a state of the art preferential voting system. Its scientific roots go back as far as the work of Condorcet during the French revolution.

MD. Can you say a little bit about the connection between the Condorcet method and the French Revolution?

BS. The development of the Condorcet Method and the French Revolution were both driven by the cause of justice. And Condorcet was involved in both as well. As many of his time, Condorcet saw a deep injustice in the treatment of the people, and this perception influenced his entire work. As a mathematician, he had proved that juries with more members (statistically) made less mistakes (and vice versa, nowadays known as Condorcet Jury Theorem). He worked on the concept of the "pairwise comparison" (Condorcet Method) and discovered "collective cyclic preferences" (Condorcet Paradox). As a rationalist politician he promoted his own findings as well as progressive (at least for his time) ideas, such as rights for women and for blacks, immediate freedom for all slaves and abolition of the death penalty, and he fought on the side of the French revolutionaries.

MD. Do you see a similar connection nowadays between the attempts to use the Schulze method for decision-making and the cause of justice?

AN. Absolutely, it is all about justice and fairness. The aim is to determine the collective preference without the encouragement of tactical voting. Every voter shall be able to express his or her true preference without being manipulated by tactical considerations.

MD. My last question is about the current and future development plans for LF.

BS. We will combine the concepts of LiquidFeedback with revision control systems, such as the software Git developed by Linus Torvalds, the [first] developer of Linux. This allows a democratic software and product development as well as broad democratizing of collaborative knowledge management. Until today, particular decisions within software and product development have to be decided by privileged people. LiquidFeedback can democratize the decision-making process. The same is true for conflict resolution in collaborative projects, such as the free encyclopedia Wikipedia. The required add-on will be integrated in LF soon.

If you enjoyed this article then please consider liking Can Europe Make it? on Facebook and following us on Twitter @oD_Europe

About the authors

Jan Behrens, Axel Kistner, Andreas Nitsche, and Björn Swierczek are four Berlin-based systems developers, who have been collaborating since the early 2000s. In June 2009, the four founded the Public Software Group to release their open source collaborations. So far, the best-known PSG project is LiquidFeedback, a decision-making software for political parties and civil society organizations. Released under MIT License in September 2009, LiquidFeedback (LF) has been adopted by various parties at a local and national level, including the German, Austrian, and Italian Pirate Parties and the Italian Movement Five Star.

Marco Deseriis is a Marie Curie Fellow at the Scuola Normale Superiore in Florence and an assistant professor at Northeastern University in Boston. His current research project examines the rise of participatory forms of networked democracy through a comparative analysis of a new generation of decision-making software for political participation. 


We encourage anyone to comment, please consult the
oD commenting guidelines if you have any questions.