Peripheral Forces.On the Relevance of Marginality
This article is about ranking. Hidden in the search engine, ranking is the elusive, complicated machinery of magic which structures our encounters with information on the Web. Despite the intricate system of ranking, most engines make search look deceptively simple: enter a term into the empty text field where the cursor is blinking, click “search”, and a list of results unfolds. Depending on what you entered, the list will be long or short. But no search engine either indexes or shows results without applying some kind of hierarchy to them. This initially seems like a normal, everyday procedure, comparable to the ways we judge between relevant and trivial, foreground and background information in everyday life; after all, our own hierarchies of visibility are also shaped according to certain needs, beliefs, and limitations. Often, the hierarchy applied by ranking rewards what is already popular. Like a forward echo, the search seems to already know what we “want” before we have run the query. The threshold of ranking also tends to suppress less often viewed currents and opinions in broad, public topics.
Redesigning the search engine begins with challenging the principles of relevance and popularity inherent to ranking. In this essay, we argue how ranking mechanisms translate as phenomena of sociability, and how a different take on the sociability of “weak ties” may bring a different appreciation of their relevance to networks.
On the web, choices we make – and links we click – have far-reaching consequences for what others find, see and click. We are not neutral bystanders to an unfolding natural process on which we exert no influence. What we choose to look at and link to is what we reward and recommend to others. What we ignore, we eventually obscure – if others concur. This process entails a form of political power exercised through sociability, where networks, protocols and ranking mechanisms supplant the exercise of direct force, and algorithms become wholesale distributors of visibility and obsolescence. Results displayed at the top of a ranked list are “naturally” considered more relevant than ones at the bottom.
But the properties defined by the algorithm or the values of the ‘relevance decision’ remain kept from public view and user intervention. Most search engines are designed to cater to the impossibly large target audience of “every other web user”. Just as a global fastfood chain offers its clientele little insight into the intricate fabrication of a hamburger, there seems to be no need for the world’s mass markets to tweak the controls of its web search results – or so it seems – as long as the results list is stunningly long, impossibly up-to-date and loading at the speed of light. After Google, since that is what we want, that is what we get.
The public’s engagement with the web is mostly made on terms of free choice; clicking, tagging, linking and browsing1 are social actions, often with reciprocal forms of reward – which, in online social networking platforms such as Facebook and MySpace, have become systematized and automated by means of the “friend” ranking as a mutually affirmed social bond. On a larger scale, these forms of reward, subscription and consent constitute a genuine form of “network power”. “The idea of network power argues that we are pulled by our choices along avenues smoothed by the prior choices of others,”2 as David Grewal explains. In networks, power by sociability is immanent.
While networks can be viewed as the natural arena for the exercise of a global free will – producing an aggregative, collective form of giant decision on “what we all want” – we suspect that the most relevant structural components of networks are their gaps. These “structural holes” are not just the limitations of sociability within clusters of mutually affirmed alliances and links; they are also gaps of knowledge and power.3 We will argue how weak ties spanning structural holes provide a critical alternative to ranking models based on aggregative forms of sociability, by being unpopular, autonomous and peripheral. We will explain the argument by an up-close exploration of the theory of structural holes as developed by Ronald Burt. By way of example, we will assess notions of social power present in U.S. president Barack Obama’s election campaign, as well as his initiatives for the reinvention of the U.S. government’s relationship to the people. These will ultimately be structurally compared to the semi-independent terrorist cells which constitute the Al Qaeda network after central command has been taken down.
From relevance to authority, and back
Ranking mechanisms may constitute a form of power both more sneaky and more structural than old-fashioned coercion; the affirmative force of sociability, present in the reciprocal processes of cross-linking and tagging, may have become the vehicle of a kind of power which suppresses alternatives without coercion being needed.
Currently the hierarchy of relevance applied by ranking algorithms propels a hierarchy of influence: the initial relevance of results topping the list is reaffirmed by the public’s browsing behavior. The more relevant a result, the easier it is found and clicked on, and the more it eventually will get seen, read and linked to. Such a ranking process risks constituting a self-fulfilling prophecy of relevance. The most well known example of this type of algorithm – or mode of decisionmaking – is called PageRank, the algorithm of the Google search engine. Google’s founders Sergey Brin and Larry Page state that
a user might prefer a news story simply because it is linked directly from the New York Times homepage. Of course such a story will receive quite a high PageRank simply because it is mentioned by a very important page. This seems to capture a kind of collaborative trust, since if a page was mentioned by a trustworthy or authoritative source, it is more likely to be trustworthy or authoritative.4
When Brin and Page assert that “quality or importance seems to fit within this kind of circular definition”5, they refer to “a kind of peer review”6 – borrowing the name from the process by which impartial experts scrutinize the work of fellow scholars; often, the same assumption is made when the efficiency of algorithms is tested on academic datasets where citations count as links. Brin and Page continue:
[I]f a web page has a link off the Yahoo home page, it may be just one link but it is a very important one. This page should be ranked higher than many pages with more links but from obscure places. (…) [A] page has a high rank if the sum of the ranks of its backlinks is high. This covers both the case when a page has many backlinks and when a page has a few highly ranked backlinks.7
The Google founders leap across these two contradictory logics — the academic peer and the social actor. With the PageRank model, “more important and central web pages are given preference”.8 Since highly ranked links exert more influence on the distribution of authority, there is a higher payoff for “proximity” to the authoritative source.
Nevertheless the web culture of the amateur is not bound to academic impartiality. Unlike the academic processes of peer reviewing and referrals, web users are social actors who don’t need to discuss or verify prior assumptions. Actors often seek to bond with others “like them” and confirm what they already consider true or valuable. In other words, there is an incentive to link with others who share certain (ideological) standards – or simply the same protocol. The advent of blogs, amateur journalism, social tagging and other forms of user-generated content puts any prior definition of authoritative sources under further pressure. Because of their basis in the handshake of protocol and the aggregation of user benefits in joining mutually affirmative standards of cross-linking, blogs and social networking platforms are acquiring a high level of PageRank despite the fact that they do not possess factual authority comparable to The New York Times. In other words, what began as a peer reviewed quality standard is transforming into a “peer pressured” sociability standard. Links function as a way for users to associate themselves with what they like, voting for those they agree with; linkage – sociability – is becoming the stand-in for authority. It follows that dense social spheres, tied by consent, often disregard the unpopular information offered by the weak ties at their peripheries.9
Politics of structural holes
In his book Structural Holes: The Social Structure of Competition, Ronald Burt addresses precisely these features which reinforce central actors. Running counter to the aggregative logic of authority, Burt argues for the relevance of weak ties. In his approach, social cohesion is an indicator of redundancy. He sees a high probability that tightly-knit clusters circulate the same information among their actors. Burt is particularly concerned with defining the conditions under which a networked actor may benefit from putting other actors into contact.
Structural holes are the gaps between non-redundant contacts. As a result of the hole between them, [once bridged] the two contacts provide network benefits that are in some degree additive rather than overlapping.10
The main cost of aggregative sociability lies in the redundancy of information. There are also structural constraints on the freedom to opt out or relate to unpopular and peripheral ideas. At the level of the wholesale distribution of visibility via algorithms, the capacity of peripheral ideas to surface in a list of results becomes structurally affected by the relative weight of heavily cross-linked, socially coherent spheres of consent, combined with the algorithmic preference for these spheres as more “authoritative”. This preference was first based on an “extrinsic” definition of authority taken from printed media and the academic peer review, but then gradually became “intrinsic” to the aggregative economy of linking on the web, with standards of sociability supplanting old-fashioned authority. Therefore, the “design of consent” of Google’s list – in particular the first 10 results – harnesses a preference for sources, many of which have become authoritative via their social structure (like Wikipedia).
The disregard that social networks tend to display for Burt’s structural holes translates into a further suppression of marginalized ideas, some of which may hold valid criticality of the popular theses. The structure of a critical argument and its difference from the mainstream may be linguistic or semantic in nature. For example, the name “Radovan Karadzic”, on English-language news sites, is mostly surrounded by a “semantic escort” of words like “criminal”, “tribunal” and “Bin Laden”, whereas on Serbian-language blogs words like “narodni” and “heroj” (“national” and “hero”) may be more frequent.11 The critical knowledge at hand is not that Karadzic is not a criminal, but that users get confronted with both agreement and disagreement with the general thesis – whether or not they consider Karadzic a criminal. With the broad adoption of the web, opposing points of view can be directly consulted rather than just being heard of. The challenge of a search engine algorithm emulating something of a peer review need not avoid the inherent partiality bound to political questions. If sources of alternative views are systematically irrelevant because of their lack of authority- sociability (formalized through semantics), the “giant decision” through Google-like ranking becomes dependent on actors able to — both semantically and socially — reframe an issue on a large scale.
The propagation model rewarding central actors with more influential “votes” produces as a consequence that, when an actor’s ranking increases, his increase in authority propagates to his neighboring actors; all linked actors get gradually promoted along. This situation, where self-interest promotes “peer-interest”, has an impact on social formations such as online communities but also on political practice.
Studies on how social actors influence each other’s political views revealed similar effects even prior to the use of social networking platforms for large-scale political campaigns, such as the one for the election of Barack Obama. They show that voting patterns are highly influenced by the choices made by one’s acquaintances, resulting in cascading effects along the social hierarchy.12 The political scientist Cass R. Sunstein warns that these “cybercascades” might lead to the fragmentation of public discourses, with a further polarization into completely disconnected political spheres. Sunstein speaks of “communication worlds” where people “hear echoes from their own voices and wall themselves off from others”;13 each sphere is reaffirming its own bias towards like-minded convictions.
The “mere exposure” effect means that [group] polarization is likely to be a common phenomenon in a balkanized speech market. (…) To the extent that these exposures are not complemented by exposure to competing views, group polarization will be the inevitable consequence.14
Through bypassing group polarization, current ranking models deflect these “cybercascades” away from each other, avoiding their confrontation. Organizational models based on sociability — rather than authority — rely on aspects that have not yet been adopted by the search engines market. In the next paragraphs we will see how these networks have fully integrated the consequences of fragmentation, making these aspects the key incentives for building their networks.
When Barack Obama ran for the U.S. presidency in 2008, his campaign displayed an unprecedented ability to reach, influence and mobilize actors who were peripheral to the Democratic points of view, or to the elections at large. The cunning of Obama’s team, directed by David Plouffe, was to address the division not from within his camp but from its border with Republican and undecided voters. This campaign turned voters into social actors by relying foremost on their own abilities at communicating, linking and forging other actors. “Neighbor to neighbor canvassing actions” were organized in every quarter, town, city and throughout all states. To succeed, an actor had to get others’ consent at voting for the same candidate. By including new voters as part of the network, they diffused the social border between the two camps. Field organizer Patrick Frank, addressing campaigners, tells them to “[b]uild yourself a team and be organizers too. There’s no end to it.”15 The diffused process is taken under control, but unlike other campaigns, the structural advantage of small, autonomous clusters is played out. Gene Koo drafts the following model:
The superstructure of the campaign is traditional, top-down commandand- control (with information flowing upwards, of course). At the roots the campaign – as is typical for most volunteer efforts – comprises ad hoc mesh networks. It’s in inserting strong, tightly-knit teams that the campaign has made the greatest innovation.16
The model is optimized since redundant information is intentionally kept within the tightly-knit teams; the only critical information that is coming from the topdown hierarchy is for avoiding overlap. The teams can fully act on their structural autonomy and social opportunities. This means that the political practice of inclusion and exclusion to and from the decisive demos17 is no longer in the hands of a representative system of authority. A new level is inserted, formed by clusters of actors pushing local snowball effects. Koo writes that “each team, as a whole, functions like a paid staffer, with similar responsibilities and accountability”. 18 Their advantage is that “the teams also have a deeper and wider network” 19, since each actor can rely on his own network of contacts. The teams are close enough to others so as to easily deal with new information, which allows them to broker bridges between conflicting ideas. Their autonomy abounds with structural holes, and their crucial role in the campaign is to connect otherwise disparate spheres of actors.20
Centrality-based ranking would have a hard time detecting these clusters as the key components of the campaign, simply because they are at the periphery of their social sphere. Community organizers and field managers have much more authority than the teams and ad-hoc volunteers; they exert their power and take decisions which have far-reaching influence. Their role is to manage the network internally, from its central positions. They have to avoid information overlaps which would otherwise make them “structurally equivalent” for the campaign.21 Regardless of their communication, they rely on the small teams and volunteers for accessing new (outside) information.
Cancelling out central power
Along with the notion of “authority” that influenced Brin and Page’s model for Google search, Jon Kleinberg has also introduced the concept of “principal and non-principal communities”. Kleinberg’s model situates each actor in a community and measures an actor’s relevance ranking according to that community’s size. Large (principal) communities tend to suppress actors engaged in smaller ones. Brin and Page did not include the mechanisms Kleinberg proposed to avoid this effect22 into Google; actors engaged in small, peripheral communities typically have low PageRanks. The difference between these two models can be traced when comparing the campaign structure, facilitated through the social networking site my.barackobama.com, with the one used in the social-governmental web site: change.gov. Change.gov was developed for Obama’s transition to the U.S. presidency. It repeated on a governmental level the social voting logic by putting to use the “user generated content” of the people itself as a source of policy change. The principle of Change.gov is a PageRank devoid of influential differences; since it is a direct voting system, there is no social network through which authority can propagate. As advanced by Kleinberg and Burt, the cost of the excess of centrally-oriented actors is the circulation – and dominance – of redundant information. Regardless of the influence that each tightly-knit social cluster would be able to cover, as long as they remain non-principal and nonoverlapping, all new (formerly weak) ties would be non-redundant. The actors who are the least constrained socially and the most autonomous in their ideas are the ones dealing with outside, conflictual information. Peripheral actors occupy the border between the two political spheres.
Valdis Krebs, chief scientist and founder of orgnet.com, has extensively studied political instances when bridging affects political choices. In an article published in the book Extreme Democracy he assesses how weak ties function across spheres holding different political views.
Not only do individuals have [motivational] thresholds, but communities and networks also have thresholds. These are barriers to minor opinions taking hold. Opinions and information that run counter to the dominant view within the network are usually dismissed.23
By looking at how actors’ motivations affect their ability to influence their acquaintances, this study reveals how the threshold for accepting new opinions is a property of the community rather than one of each actor independently. Ronald Burt questions the role that motivation would have on the behavior of actors in networks and their engagement in communities. Whether actors engage in relations because they are motivated or because the opportunity presents itself is, for Burt, a non-issue. He considers “motivation and opportunity as one and the same”, since an actor is author of his network just as the nature of the actor’s environment is.
[A] network rich in entrepreneurial opportunity surrounds a player motivated to be entrepreneurial. At the other extreme, a player innocent of entrepreneurial motives lives in a network devoid of entrepreneurial opportunity.24
This means that a network can be emergent or adjusted in order to exploit the advantages of structural holes. Krebs proposes to evaluate the returns of Plouffe’s ground strategy in a sharp comparison with previous elections. During the 2004 US elections the New York Times published an article based on a cluster analysis of online book purchases, which corroborates Sunstein’s polarization argument.
[Krebs] calls this pattern the “echo chamber” effect: for the most part, he found, buyers of liberal books buy only other liberal books, while buyers of conservative books buy only other conservative books.25
Krebs ran the same analysis towards the end of the 2008 elections. This time the graph showed a new tendency for books read by both liberals and conservatives. Interestingly these books were mostly read at the periphery of each political sphere.
The anti-Obama books – The Obama Nation and The Case against Barack Obama – are mostly being read by people who are already against Obama. One of the anti-Obama books is connected to one of the [bridging] books. (…) Could some undecided voters be reading this to make up their mind on Obama?26
Here, weak ties stretch over those parts of the network where we find content that is unknown to the center of each group. In political terms, bridging these holes and strengthening these ties translates as conflict or friction. As Sunstein argues, such outside information — unpopular within one’s sphere — is essential for bridging polarized fragments of public discourses. Obama plans to carry out his politics along the same logic. When accepting the presidency on November 4th 2008, he proclaimed that “I will listen to you, especially when we disagree.”27
Burt understands relations not only in their positive properties — such as their informational benefits — but also in the constraints limiting the level of structural autonomy. He asserts that the most dramatic change in structural autonomy happens with the first signs of constraints.
[S]tructural holes have their greatest effect as unconstrained actions begins to be constrained. Once action is constrained beyond a low level … further increases in constraint are almost superfluous.28
The opportunity for bridging is higher for non-central, unconstrained and peripheral positions, since they are able to play conflicting demands against each other, while non-conflicting demands end up in a self-referential core. This core is enclosed by the alignment of the structural holes around actors. In more recent works, Burt compares structural holes to “network closures”,29 which at a higher scale of collectivity are the sum of holes, the structural periphery and social border of a consensual sphere of actors. Closely following Burt, Jon Kleinberg has recently developed an optimized algorithmic model testing the extent to which social differences can be resolved when all actors use their network power to bridge the holes surrounding them.
[T]he symmetry-breaking [is] putting different nodes at different social levels in which they obtain different payoffs. (…)[S]ince the advantages of bridging decreases rapidly with the number of people doing it, there are “first-mover” advantages that translate into broken symmetry in equilibrium. 30 Kleinberg’s model pushes structural holes to extreme states of equilibrium. His findings corroborate with studies having similar aims; using different methods, these reveal that social stratification is a key example of stable networks even when actors keep acting on their structural opportunities. At the other extreme of political strategies, there are networks of terrorist cells; although these fully rely on the structural holes between their cells, the latter are only rarely acted upon. Krebs provides a much-cited argument in his study of Osama bin Laden’s Al Qaeda structure, as the design of stealth networks and their ability to function even without central command at all.
In a normal social network, strong ties reveal the cluster of network players – it is easy to see who is in the group and who is not. In a covert network, because of their low frequency of activation, strong ties may appear to be weak ties.31
Krebs shows how weak ties — which typically connect actors that are socially distant from each other — when used for coordinated action, are able to bridge “deep structural holes”. The depth of a structural hole increases with the weakness of the tie between actors of different cells.32 Once overcome, deep holes function as “ports of information” with the benefit of exclusive access.33 Krebs further explains the strategy behind fostering strong yet dormant relations.
The hijacker’s network had a hidden strength — massive redundancy through trusted prior contacts. The ties forged in school, through kinship, and training/fighting in Afghanistan made this network very resilient.34
Both Obama and Osama’s use of network power can be explained through the lens of structural holes; both have highly resilient networks since their key components are highly autonomous. But their difference lies in their use of standards – the central conventions common to all of a network’s actors. With his definition of network power, David Grewal distinguishes mediating standards from membership standards:
[A] mediating standard is a standard that governs access to others by its very nature; some particular social activity is inherently regulated by it … they form a part of that very activity itself. [A membership] standard is not necessarily basic or inherent to a given activity … but rather establishes an ideal or target.35
Grewal asserts that mediating standards are self-enforced, which explains why the activity of networking – and bridging structural holes – empowers those engaged in it, while the network gets richer and more diverse in information. With membership standards, revealing the network’s structure presents a risk for its own activity and survival.
A tentative conclusion to our structural comparison comprises two negations. Obama’s network seems to be driven by a renewed ideology with a consistent brand identity. Following Grewal’s prescriptions, this network is governed by a mediating standard; one becomes part of the network through networking. The system itself is self-referential, but all efforts go into the setting of the adhoc peripheral meshes to absorb potential conflict at the border of the political sphere. Osama’s network, by contrast, is composed of isolated cells separated by dormant ties. Such a structure is designed to create endless opportunities for the setting of powerful strategies by bridging any number of cells. Osama’s network need not rely on active peripheral networking structures, since it is governed by a membership standard. In this article we argue that the most critical network positions lay at the periphery and in between the dominant spheres of information networks. Current models of ranking are unable to capture the power of weak ties, making their design less suited for searching for actors with peripheral yet substantial social engagement. Future search engine designs have to confront both the back-end – with the algorithm, and the front-end – with the user interface, with the weak ties and structural holes in networks. With gamechanging ranking algorithms no longer based on central cores but on peripheral forces, the design of search engines is at a promising new inroad for change. Users need to be made more aware of what the currents are around the page they are looking at, clicking on, and choosing. They need to be able to grasp where the borders between these social currents persist, and where they consent or diverge. Ultimately, network power induced by ranking models that value the spanning of social bridges, will hold an incentive for resolving polarized views instead of reinforcing them.
1 Yuting Liu, Bin Gao, Tie-Yan Liu, et al., “Microsoft Research Asia, BrowseRank: Letting Web Users Vote for Page Importance” (Proc. 31st Annual International ACM SIGIR Conference, 2008), 451-8
2 David S. Grewal, Network Power: The Social Dynamics of Globalization (New Haven: Yale University Press, 2008), 140
3 Ronald S. Burt, Structural Holes: The Social Structure of Competition (Cambridge: Harvard University Press, 1992), 7
4 Larry Page, Sergey Brin, Rajeev Motwani, and Terry Winograd, “The PageRank citation ranking: Bringing order to the web”, technical report (Stanford: Stanford University, 1998), 11
6 Ibid, 15
7 Ibid, 3
8 Ibid, 15
9 Mark Granovetter, “The Strength of Weak Ties”, in American Journal of Sociology Vol. 78 Issue 6
(Chicago: University of Chicago Press, 1973), 1360-80
10 Burt, 1992, 47
11 Metahaven, “Multipolar Search: EXODVS” (Pancevo Republic! 13th Biennial of Art, Pancevo, Serbia, 2008) Under development together with Tsila Hassine, this model follows prior research conducted at the Lab for Media Search at the National University of Singapore
12 James H. Fowler, “Turnout in a Small World”, in A. Zuckerman (ed.) Social Logic of Politics (Philadelphia: Temple University Press, 2005), 21
13 Cass R. Sunstein, Republic.com (Princeton: Princeton University Press, 2002), 44
14 Ibid, 66
15 Zack Exley, “The New Organizers: What’s really behind Obama’s ground game” (Huffington Post, 2008), http://www.huffingtonpost.com/zack-exley/the-new-organizers-part-1_b_132782. html
16 Gene Koo, 2008, “A network analysis of the Obama 08 campaign”, http://blogs.law.harvard.edu/ anderkoo/2008/10/14/a-network-analysis-of-the-obama-08-campaign/ 197
17 Jacques Rancière, Disagreement: Politics and Philosophy, translated by Julie Rose (Minneapolis: University of Minnesota Press, 1998), 10.
18 Koo, 2008.
20 Burt, 1992, 30. By bridging between conflicting demands, these teams set the conditions for what Burt refers to as the “tertius gaudens”. The “tertius” is the third who benefits from the disunion of others. On page 48 he states the following. “In the swirling mix of preferences characteristic to social networks, where no demands have absolute authority, the tertius negotiates for favorable terms. (…) Structural holes are the setting of tertius strategies. Information is the substance. Accurate, ambiguous, or distorted information is moved between contacts by the tertius.”
21 Burt, 1992, 19
22 Jon M. Kleinberg, “Authoritative Sources in a Hyperlinked Environment”, IBM Research Report RJ 10076 (Proc. ACM-SIAM Symposium on Discrete Algorithms, 1997), 15.
23 Jon Lebkowsky and Mitch Ratcliffe, Extreme Democracy (Lulu.com, 2005), 119
24 Burt, 1992, 36
25 Emily Eakin, “Study Finds a Nation of Polarized Readers” (The New York Times, March 13 2004), http://query.nytimes.com/gst/fullpage.html?res=9F01EFD6103EF930A25750C0A9629C8B63
26 Valdis E. Krebs, “New Political Patterns” (orgnet.com, 2008), http://www.orgnet.com/divided. html
27 Barack Obama’s speech as president-elect of the United States, Chicago, 4 November 2008
28 Burt, 1992, 94
29 Ronald S. Burt, Brokerage and Closure: An Introduction to Social Capital (Oxford: Oxford University Press, 2005)
30 Jon M. Kleinberg, Siddharth Suri, Éva Tardos and Tom Wexler, “Strategic Network Formation with Structural Holes” (Proc. 9th ACM Conference on Electronic Commerce, 2008), 3.
31 Valdis E. Krebs, Mapping Networks of Terrorist Cells (Connections 24-3, 2002), 49
32 Burt, 1992, 42
33 Ibid, 43
34 Krebs, 2002, 49-50
35 Grewal, 2008, 22-3