party in each topic area, calculating weighting variables to calibrate the relative power of that party to decide the next specific vote in that area. Thus a party’s strength in Parliament would be decided not by how many of its politicians had won seats in the most recent election, but by the momentary fraction of the electorate that had selected it to represent them on the particular issue at hand.
In cases when Voter A delegated to Voter B, there are two possibilities. First, Voter A’s party choice could copy Voter B’s party choice, changing whenever Voter B changed a party selection. Second, if Voter B achieves some threshold number of delegations from other voters, Voter B could become in effect an independent member of Parliament. The balance between party influence in Parliament, versus the influence of individual delegates representing many people but without a party organization, could change over time and across issues. In addition, each voter might have several selection pages in the secure online database, one for local government, one for regional government, one for national government, and ideally even one for world government.
Presumably, each political party, and each unaligned individual delegate, would have a public web page listing positions on the various general issues. It is conceivable that some party or solo delegate might choose to communicate privately, even in secret, with individual voters, and no technical barrier prohibits this. However, democracy generally benefits from broad public discussion, and this system assumes that some kind of public debate has identified what the distinct issue areas are. It is one thing to say that tax policy is logically separate from environmental policy, but when a decision must be made about taxing emissions from a polluting industry, the picture becomes complex.
When it comes time to implement Liquid Democracy, there will be a host of very specific technical questions, including many about the processes used to identify opinion leaders and topic areas. The simple idea just presented of a government database with a private page for each voter is only one of many possible ways to proceed, and a modern political system may require combining several of them. Furthermore, we have not considered yet how a political party would develop its platform, and we should imagine how advanced information technology might manage that difficult process. Without pretending at this early point to know which methods should be used in what combination, we can catalog possible components of a twenty-first century political system based on Internet.
A very large number of information technology methods have been developed recently to support group decision making, and they can be assembled in different ways. Many of them have not generally been presented in political terms, so it will take some imagination even to recognize some of the valuable technological resources available to us. Here we shall consider only three: reputation systems, recommender systems, and online group formation systems.
From a certain perspective, Google is a political entity, ruling world culture by deciding where people will find the information they desire, in terms of the most complex classification system that has ever existed, and a dynamic one at that. It is political because it is based on the equivalent of voting, in the form of links people put on their web pages to other people’s pages. Without getting into details, the Google search engine uses two kinds of data. One is the words written on a web page, and the other is the pattern of links coming to a web page. A key part of the mechanism is the pagerank algorithm — actually a class of algorithms that assign a score to each web page in terms of the links coming to it, adjusted by the ranks of the pages that sent those links (Page et al 1998*; [2]).
For example, consider the English-language Wikipedia page of Pirate Parties International. To find many of the web pages that have links to this particular page, one can enter into Google: “link:en.wikipedia.org/wiki/Pirate_Parties_International.” On October 21, 2011, Google listed 141 such pages, including some belonging to branches of the party, as well as pages in many different languages. Entering “link:www.piratenpartei.de/” turns up fully ten times as many web pages. It is even possible to enter two “link:” URLs, and get a listing of all the pages that link to both of the two target webpages, which can become a metric of how similar those two pages are, in comparison with other pairs of pages that might have more or fewer common in-coming links.
Thus Google page rank is first of all a measure of popularity, but also data that can be used to map web pages in terms of similarity. Of course we should be cautious about using Google as our voting system. Yes, one can easily tabulate the relative numbers of in-coming links for the web pages of politicians, but this is not the same thing as their popularity with voters. Many of the highly ranked pages sending links may belong to ideological organizations, venial corporations, or crazy fanatics who put up many webpages that draw attention for being bizarre, not for being wise. Yet as a technical method akin to a voting system, the Google search engine has been remarkably successful and may have lessons for those who wish to reform the political system in the light of advanced communication technology.
In a sense, Google is a reputation system, and its methods can be adopted to measure the reputations of political leaders, or to cluster them into parties if they have not already organized. The original area in which such network-based techniques were developed was bibliometrics — specifically studying the pattern of literature citations to identify the most influential publications and scientists (Borner 2010*; 2011*). Similar methods are now used in a number of fields, using a range of computational methods, to identify leaders in a network of communication.
A recommender system is a database and statistical analysis engine that recommends future actions to the user — typically what movies to rent or books to buy — based on the user's prior behavior or expressed preferences (Basu et al 1998*; Canny 2002*; Herlocker et al 2004*). These systems are widely used in Internet advertising, in order to customize the sales effort to fit the interests of the audience, but can be developed not only to cluster small issues into coherent political programs, but also even to conduct a form of science-moderated direct voting. The distinction between reputation systems and recommender systems is unclear, and the two share many technical features. But the best way to get the idea across is to look at one of the best-known pure recommender systems, the Netflix movie rating system. [3]
After people rent a movie from Netflix, they are encouraged to rate it on a preference scale from 1 to 5, and their responses are used to determine which movies Netflix will recommend they should rent. Starting in 2006, Netflix held a contest, providing a huge training subset of their data, based on hundreds of thousands of raters, and challenging contestants to devise an algorithm that would best predict customers’ ratings on movies for which the data were not in the training set. I entered the contest, not intending to compete, but to explore how such data might be used to map the styles and ideological orientations of movies. I knew from my earlier research, that people’s preferences were often largely shaped by the visual style of a movie, the leading actors in it, and the year in which it was released — but modulo all these extraneous factors ideology could sometimes be detected (Bainbridge 1992: 470-481*, 2007*).
To illustrate the methods here, I have selected 15 movies that concern artificial intelligence or virtual realities — topics close to fluid democracies in their reliance on information technology for radical social purposes. One consequence is that these films may not differ much from each other in term of ideologies, precisely because they have so much in common. The first methodological challenge is that many respondents rate very few movies, so to get robust results I focused on the 6,551 respondents who had rated at least 10 of the 15, only 110 of whom had rated all 15. They are all diehard sci-fi fans, but if the data concerned politics rather than films, we would be dealing with knowledgeable experts on that very different topic. Table 1 lists the films, the year each was released, the average ratings, and the results of a factor analysis of the data.
Table 1: Fifteen Movies about Advanced Information Technology
| Title | Year | Netflix Raters | Mean Netflix Rating (1-5) | Factor 1 |
|---|
