Happy accidents

1 Introduction

In this paper we wish to make a contribution to the study of the deliberative potential of Internet communication. In the first, analytical, part of the paper we draw a distinction between diverse and opposing views. We argue that the study of the deliberative potentials of Internet communication should turn on the effective possibilities to expose users to opposing, not just diverse views.

In the second and more exploratory part of the paper we ask if online experiences facilitate exposure to opposing views. Using recent empirical findings, we argue that Internet communication is a “mixed blessing” for deliberation, as it generates both unintentional exposure to opposing views, as well as “drivers” that channel users away from opposing views.

2 Distinguishing opposing from diverse views

Proper deliberation extends beyond the mere articulation of reasons or arguments in support of actions; it is the seeking and the weighing of pros and cons by the deliberating agent/s that distinguishes deliberation from other forms of reasoning.1 We say that we deliberate, individually or collectively, when we engage in a distinctive mode of reasoning.2 We deliberate about a given course of action when we suspect that there might be reasons against it as well as reasons for it. If we did not think that there might be, at least potentially, reasons for not doing X alongside reasons for doing it, we would use reason in a different way. We would seek to establish that X is the right course of action by supplying solid arguments for it. We would not actively seek potential counterarguments.

In this section we wish to emphasize the distinction between diverse and opposing views.3 A long tradition in liberal theory has praised the benefits of diverse and opposing views for adequate deliberation. It has often been argued that a necessary and sufficient condition for the benefits of deliberation to materialize is that participants in discussion hold diverse views and articulate a variety of perspectives. That tradition ranges from Mill, to Popper, to Sunstein, to many others.

The problem with this line of thinking is that “diversity of views” and “opposing views” get treated as roughly interchangeable notions. It is our contention that these notions are not interchangeable. While both opposing and diverse opinions may be needed for adequate deliberation, diversity of opinions alone is insufficient for adequate deliberation.

Elsewhere, Manin4 elaborated on possible reasons for the reluctance to search for the potential downsides of a measure supported by a good argument and to articulate them, even among agents coming from diverse backgrounds and holding a variety of perspectives. For example, the costs of information search may lead people to use “satisfying” heuristics and stop the search for reasons once a good argument has been found. Others may not wish to be seen as opponents of a measure that arguably promotes a common goal. Yet others may surrender to conformity pressures. As a result, few, if any, arguments pointing to the potential downsides of a proposed measure can be heard in deliberative settings, even if members of the deliberating body hold diverse views and the set of arguments is lopsided.

Two further considerations lend additional weight to our claim that diversity of views per se is insufficient for adequate deliberation. These two points, regarding cognitive processes and selection effects, are especially relevant to our discussion below about exposure to opposing views online.

Social and cognitive psychological research shows that people do not process information in a neutral and unbiased manner, but instead systematically misperceive and misinterpret evidence that is counter to their prior beliefs, in order to make the new information consistent with these prior beliefs. Even if decision-makers are exposed to a variety of arguments against a certain view, they can still fail to consider properly the new evidence on its merits, when it runs counter to their prior beliefs. There is, however, some experimental evidence that the most effective way of countering the effects of such biases is to give greater salience to information that runs directly counter to prior beliefs.5

Most importantly, and most relevant to the Internet, is the possibility that mere diversity of views may result in the generation of enclaves of like-minded people. A robust finding from a large body of research on social and political behaviour is that when choice is available, agents prefer to interact and organize with, and receive information from like-minded others, a phenomena known as homophily.6 Below we show that this tendency is manifest in a variety of spheres online.

When diversity of views is combined with freedom of speech and association, and especially with enhanced abilities to locate like-minded others and filter out opposing views, the result may be enclaves of like-minded people talking to one another, even in a context of a wide multiplicity of views. In the light of trends such as residential segregation, fragmentation of the media and narrowcasting, the consequences of segmentation seem to be of prime concern from a deliberative standpoint.

The deliberative potential of a given environment or medium should be assessed by looking at the probability that agents will be confronted with opposing views, and will give them due consideration. Thus, in seeking to estimate the deliberative potential of Internet communication we should focus on the probability that users will be exposed to opposing views online, and on the probability that such exposure will trigger the distinct deliberative mode of reasoning “within” individuals.

3 Generating exposure to opposing views

The heterogeneity of participants’ backgrounds and opinions does not necessarily entail the articulation of arguments for and against particular courses of action during deliberation. It is the opposition of views and reasons that is necessary for deliberation, not just their diversity. Diversity of views is not a sufficient condition for deliberation because it may fail to bring opposing views into contact.

But exposing agents to opposing views during deliberation entails a number of challenges. First, typically there are substantial opportunity costs for the deliberating agents, as deliberation takes time and cognitive resources that may be devoted to other issues, more aligned with the deliberants’ interests and concerns. Hence, debates on issues of public concern may have to be actively promoted.

Second, debates with an adversarial character need “enhanced” promotion and organization, since they require participants to face conflict and generate talk across cleavages. But research shows that people tend to avoid the psychic discomfort of expressing opposing views and becoming involved in contentious discussions. Whereas learned scripts largely regulate recurring interactions with others, a cognitive shift occurs when others challenge one’s views or when one feels the need to challenge others’ views (Ryfe 2005).7 Such a cognitive shift disrupts individual reasoning routines and generates anxiety, and therefore people are reluctant to experience it and try to avoid it in their daily routine (Ryfe 2005, see Eliasoph 1998; Marcus, Neuman and Mackuen 2000).8

As a result, people tend to carefully select their conversation partners. Research indeed shows that offline political talk occurs mostly among friends, family, and like-minded others.9 Even the voluntary associations that people choose to join evolve to become rather homogenous ideologically.10 Further discussion of these phenomena is beyond the scope of this paper.

One cannot expect adversarial debates to arise spontaneously in a diverse society with freedom of speech. Public deliberation is a complex public good whose facilitation has to overcome a number of obstacles (opportunity costs, generating cross-cleavage communication, overcoming conflict avoidance), and requires extensive organizational work. When organizational costs are borne by interested parties, the hazard is that they skew the deliberation so the outcomes favour their interests (Przeworski 1998),11 for example by manipulating agendas, argument pools and procedures. Presenting “devil’s advocate” arguments may be especially challenging if the organizers of deliberation feel that allowing them may have adverse consequences.

Let us now move to discussing the possibilities of exposing agents to opposing views online. One should steer away from simplistic arguments that directly link, for example, abundance of information to familiarity with opposing views (for similar points see Bimber 1998; Delli Carpini and Keeter 2002).12 When information is abundant but the attention of information consumers is scarce, agents use selection strategies and short-cuts in exposure to information sources, or even choose to remain uninformed. The logic of “rational ignorance” still prevails even if, as Lippmann puts it, “by some development of the radio every man could see and hear all that was happening everywhere, if publicity, in other words, became absolute.”13 Scale and accessibility are by themselves insufficient to account for the deliberative possibilities of Internet communication. The effective possibilities to get exposed to opposing views are also determined by such factors as the organization of content and links, and the ideological makeup of deliberative spheres online.

Before getting there, let us note that the literature on online deliberation focuses on facilitated settings. Organizing such forms of deliberation online is substantially less expensive than off-line. Participants can deliberate from the comfort of their homes, without necessarily limiting themselves to very specific times and places. It is also significantly less expensive to create a representative sample of a decision-making body online (due to reduction of coordination costs, transportation costs, and so on). Offline, when organizers aim at achieving a representative sample of a geographically dispersed population, they must bring participants to a common physical location at a specific time, which can be extremely expensive.14 Even more expensive to organize are off-line longitudinal deliberation settings, which require multiple sessions separated by long intervals of time.

Experiments in online deliberation have produced encouraging results from a deliberative standpoint, such as lack of polarization and radicalization, knowledge gains, more considered opinions, satisfaction from the deliberative process, and enhanced feelings of efficacy.15 Such experiments point to the continuing promise of utilizing the Internet to support facilitated deliberative arenas to discuss the problems of heterogeneous publics.16

However, these settings represent just one instance of deliberation online. Such deliberative moments of interactive exchange among members of heterogeneous groups are indeed rare; as explained above, they are rare for good reasons: because they are still relatively expensive to organize, they require cross-cleavage communication, and they interrupt regular reasoning habits. We concentrate instead on the large number of contacts or interactions that millions of online users engage in each day. We argue that these online experiences involve both factors that limit exposure to opposing views, and factors that generate unintended contact with such views. We refer to these two sets of factors, for lack of better terms, as “drivers of homogeneity” and “drivers of opposition”, respectively.17 In the following two sections we analyze them using a broad brush.

4 Drivers of homogeneity

Internet communication generates enhanced abilities to locate a variety of communication partners, acquire information from a multiplicity of sources, and “surf” between websites that present diverse and opposing views. These abilities can be utilized in different ways; some users can choose to communicate with and receive information from agents with opposing views, others can choose to communicate with and receive information from those who are like-minded, while others can choose to randomize. However, a robust finding is that the enhanced possibilities for intentional exposure online primarily lead to exposure to like-minded others.

To study the consequences of selective exposure it would be useful to look at some empirical research. Especially telling is research that deduces “macro-regularities” and patterns from the accumulated “micro-behaviours” of large numbers of users. The consequences of homophily are manifest in a variety of settings online: Internet is used for forming clubs of like-minded people, receiving information primarily from like-minded others, and creating homogenous hyperlinked spaces. Let us review these three “drivers of homogeneity” in some detail.

4.1 Associations and normative pressures

A variety of web-based institutions allow agents to tailor their communicative environment and communicate primarily with particular types of agents, creating homogenous clubs of like-minded agents. Of prime concern are the segregating effects of virtual groups. In a 2001 PEW survey, 84 per cent of Internet users indicated that they contacted a virtual group, and 79 per cent of them identified at least one group with which they maintained regular online contact. It should be noted, however, that politics is not a main reason for association for most users; only 22 per cent reported that they contacted a “political” virtual group.18 We will come back to this point later.

Survey work shows that agents join virtual communities for a variety of reasons, but primarily to obtain relevant information at low costs.19 When a large number of agents join for such reasons, the group is essentially composed of members who select to communicate with others with whom they share interests, hobbies, lifestyles, professional interest, health-related or other concerns.

Unlike in more “traditional” off-line communities, exiting Internet-based communities is usually very easy. When members feel their voices are not heard, they may prefer low-cost exit over voice or loyalty, leave the community and establish a new sub-community that is better oriented to their interests and concerns. When such a dynamic occurs, it tends to eliminate not just diversity of views, but opposing views in particular.

Here a short comment may be in order, about the type of discussion one can expect to find in such groups. Research on the social and cognitive effects of CMC (Computer-Mediated Communication) shows, perhaps counter-intuitively, that under some conditions CMC can lead to enhanced normative pressures and generate a “panoptic power.20

CMC environments (particularly text and audio-based) disable a range of contextual cues (social, visual, auditory), but often some group-level social cues remain intact, and are the only cues available for virtual group members. In such conditions, group membership becomes situationally salient. When a CMC environment is characterized by a salient sense of group membership, the lack of other cues leads to stronger influence of social norms on behaviour and to compliance with the situational norms (Postmes, Spears, and Lea 1998). Spears and Lea 21 argue that in such CMC environments, the over-reliance on minimal cues to “cognitively compensate” for the absence of other cues can lead to in-group favouritism, stereotyping and disapproval of out-groups.

This line of research is very relevant to virtual communities, where members are aware of their common group membership, but may be otherwise anonymous to one another. Under such conditions, discussion can become highly normative, leading to suppression of opposing views and radicalization (Sunstein 2001).22

4.2 Collaborative filtering and popular feedback loops

By choosing a group, agents practically select whom to receive information from and communicate with, about a topic they commonly find worth pursuing. This turns out to be an efficient way for agents to sort themselves into clubs. Such clubs can function as efficient information aggregators, and as a scaffolding to organize collective action – particularly for a variety of latent causes.23 But they can also function as efficient information filters, at the price of suppressing opposing views.

Some of the more “savvy” virtual communities enable “collaborative filtering” – they allow group members to prioritize collaboratively the information they are exposed to. Members can rate contributions and contributors; their votes are tallied and weighted to decide the rating of contributions. Automated mechanisms then edit the content of community webpages, and present items according to their ratings. The availability of content is decided by its popularity, and popular content is prioritized over unpopular content. Collaborative filters help people make choices based on the opinions of relevant others; as a result, group members are first exposed to information that other group members think highly of.

This practice of a popular feedback loop has its advantages, as it minimizes information search costs and enables an accessible short-cut to relevant information based on the tastes of (many) relevant others. At its best, when collaborative filtering is based on the force of the better argument/article, the ability to prioritize content based on discussion and evaluations publicly provided by many self-elected “experts” seems very promising.

However, at its worst, collaborative filtering can generate a high-tech version of majority tyranny, amplifying group members’ opinions and muting opposing views. Even if an occasional thought-provoking but nonconforming view is expressed, it can be effectively shunned because of its non-conforming character and in spite of its argumentative value. As a result, for example, members of progressive-leaning groups can not only talk primarily amongst themselves, but also efficiently screen out opposing views expressed by thoughtful conservatives, and vice versa.24 When applied in such ways, collaborative filtering can render opposing views literally invisible.

4.3 Ideologically homogeneous hyperlinked spaces

A third “driver of homogeneity” is apparent in the multiplicity of homophile hyper-linked ideological spaces online, in which surfers are effectively channeled to similar views and away from opposing views.

Let us start with the World Wide Web, which is the largest public communicative space online, with 63 per cent of Americans surfing the web according to 2003 PEW data, 50 per cent of whom surf on a typical day (Madden and Rainie 2003).25 Research suggests that WWW links follow homophile patterns. Hindman, Tsioutsiouliklis and Johnson.26 analyzed the link structure of political issues in the WWW, particularly focusing on themes such as abortion, gun control, and capital punishment. They found clusters of opposing views in each of these categories. The authors also found that each cluster was regulated by power-laws, such that a small number of sites inside clusters emerge as focal sites, while the majority of sites receive a negligible number of inbound-links. These focal sites help to efficiently organize the conversation inside ideological clusters. The consequence is that linking patterns spontaneously generate, for instance, not just a small number of focal sites addressing abortion, but also a small number of focal pro-life and pro-choice sites, with little inter-linking between them.

Research shows that the same homophilic link structure is evident on the blogosphere as well. Adamic and Glance27 studied the linking patterns of political bloggers. They found that the blogosphere is composed of tightly connected clusters of liberal and conservative blogs, with very few links between clusters; the great majority of links are internal to either the liberal or the conservative blog clusters (Adamic and Glance 2005, 11; for similar results, see Ackland 2005).28The authors also found that political blog clusters focus on news articles that support their political views.

Let us repeat that in the cases of the World Wide Web and the blogosphere, the macro-outcome of segmentation (with its adverse consequences for exposure to opposing views) results from the linking micropractices of authors. We can think of links between websites as constituting a form of conversation, where links manifest recognition of the importance of the linked sites and their “legitimacy” as interlocutors. The linking choices of authors direct surfers to potential conversation partners (see Herring et al. 2005). The implication of the homophile structure of these linked spaces is that surfers are likely to come across sites (or blogs) with similar ideological affinities, effectively filtering out sites with opposite views from public deliberation.29

5 Drivers of opposition

We have seen that three distinct mechanisms impede exposure to opposing views online. The research surveyed above demonstrates several manifestations of homophily: communities of like-minded people, homophilic ideological hyperlinked spaces, and enhanced abilities for exposure to information that corresponds to one’s ideological affinities. Preferential attachments allied with enhanced selection abilities are the motors behind such phenomena.

These phenomena demonstrate that, as claimed above, diversity of views is entirely consistent with the formation and persistence of enclaves of like-minded agents. More importantly, they also demonstrate that intentional choices drive out opposing views. Some agents may appreciate and enjoy conversing with others with diverse and opposing views.30 But we should not assume that users, as a general rule, actively look for opposing views. Empirical studies of online communication seem to show that users prefer to organize with, and get their information from like-minded others, when given the opportunity to do so.

This fact, however, suggests a hypothesis that opens up a new line of inquiry. If users’ choices hinder exposure to opposing views, such exposure might still happen unintentionally or even against users’ intentions. We should therefore pose the question of whether Internet communication holds the potential for unintended encounters with opposing views. If this were the case, the Internet would, quite surprisingly, qualify as a deliberative medium for a quality that it is not usually praised for.

In the following sections we argue that such is indeed the case: alongside the enhanced abilities to filter out opposing views, Internet communication also facilitates “happy accidents”, in other words unintended exposure to opposing views. We investigate the factors driving such exposure.

We focus on three factors: the creation of a variety of settings for crosscleavage communication; reduced cognitive pressures to express opposing views in such settings; and imperfect abilities to tailor one’s communicative environment online.31

5.1 Cross-cleavage communication

We claimed above that generating cross-cleavage political communication is a complex public good. Offline, sites of exposure to opposing views and especially interactive discussion with people with opposite opinions are rare; scholars argue that the leading candidates to generate such cross-cleavage exposure are the mass media32 and even the workplace.33 Our proposition is that Internet communication generates a variety of sites which are a welcome addition to such spheres. We focus on online magazines and non-political virtual communities to demonstrate our point.

Currently, the most popular news sources online are the websites of “traditional” general-interest media outlets (such as BBC, CNN and the New York Times), supplemented by additional news portals (like Yahoo News or Google News) and focal political blogs.34 Gradually, such websites include not only news stories but also enhanced “talk-back” features which enable readers to interactively respond to articles and to comments made by others, and to post links to stories published elsewhere. Such sites not only attract general readership, but also enable critical interactive discussions among readers. Such sites seem to support and enhance the role of the mass media as an agent of cross-cleavage exposure35 and seem conducive for exposure to opposing views.

Non-political virtual communities are additional candidates for generating cross-cleavage political communication. As stated before, survey work36 shows that only 22 per cent of the people who contacted virtual groups, contacted “political” virtual communities. Thus, self-described “non-political” communities seem to be much more prevalent than “political” ones.

Let us look at an interesting case study by Lampe37, who examined the characteristics of political conversation in one of the most popular communities, Slashdot. Although functioning as a community for computer hobbyists and professionals (famously providing “news for nerds”), Slashdot became a vivid deliberative forum prior to the 2004 US presidential elections.

Lampe shows that before the elections, more and more political stories were posted to the community portal. Political stories not only received significantly more comments than stories on other topics, but the comments where much more contentious. Commentators on political stories also received significantly more ratings than commentators on other stories, and there were significantly higher inter-moderator disagreements about the value of comments, suggesting that “moderators are using selection bias to judge comment values”.38

Such non-political virtual communities, just as online news magazines, attract large crowds across political cleavages. Some of them evolve to become focal sites for large-scale cross-cleavage communication, among people who did not join for ideological reasons. It is the combination of political heterogeneity, scale and interactivity that contributes to the rise of such new intermediaries for exposure to opposing views.

5.2 Reduced cognitive pressures

Earlier we claimed that two key problems in the organization of deliberation are overcoming selection and conflict avoidance. In the previous section we suggested that a variety of novel and supplementary intermediaries for cross-cleavage exposure are created online, relaxing the selection problem. Now we wish to show that in such settings (and a variety of other online settings) it is also easier to overcome the psychic discomfort that is typically generated by exposure to opposing views.

Why is self-expression easier online, particularly when the communication channel is poor (text- or even audio-based)? Research on the social effects of computer-mediated communication suggests that it should be understood as an “amplifier or magnifier of social psychological and communication phenomena”.39 Earlier, we noted that when CMC environments disable contextual cues but group membership is situationally salient, the result can be stronger influence of situational norms on behaviours.

However, when no cues are available and group membership is not salient, the opposite effect occurs: the total absence of cues generates a reduced sense of social presence, reduced awareness of the social environment, and consequently reduced concerns for social approbation, decreased awareness of, and adherence to social norms, and reduced opportunities for social control and regulation.

When the communication medium is poor, the cognitive discomforts associated with disagreement can be overcome to a large degree. It then becomes easier to express non-conforming and opposing views and even engage in contested debates. Obviously, the expression of dissonant views by some translates into exposure to such views by other members of the deliberative forum. The consequences can vary; in some contexts CMC can encourage anti-normative and uninhibited behaviours such as “flaming” [posting hostile or insulting messages – ed.]. At other times, it can also support the expression of non-conformist views and brainstorming. Further analysis of this topic is, again, outside the scope of this paper (for reviews see; McKenna and Seidman 2005; Postmes, Spears and Lea 1998).40

5.3 Imperfect tailoring and chance encounters

Earlier, we mentioned that the homophile structure of web links can channel users away from opposing views. If hyper-linked spaces were not only homophile but also “hermetically sealed”, surfers would be perfectly locked in them and there would hardly be any possibilities for chance exposures to opposing views. However, the third factor leading to exposure to opposing views is the inability to perfectly tailor exposure to political information online.

Since the link structure of the Web is not created by a “social planner”, but linking decision are made by individual authors, there is always the possibility that sites include links to opposing views. The easiness of following these links makes opposing views more immediate and accessible. Even when people surf the web looking for information to reinforce their prior beliefs, they can at times be routed to or stumble upon opposing views. Even if such cases are not common, when they do occur opposing views are just a click away, unlike access to opposing views off-line.

Search engines demonstrate the imperfect opportunities to tailor one’s communicative environment.41 Search engines are popular starting points for information searches; on any given day, 56 per cent of those online use them.42 Like the websites of traditional media outlets, they attract substantial amounts of traffic and consistently top the lists of popular websites.

An interesting feature of search engines, not often noted by commentators, is that users cannot perfectly tailor the ideological affiliation of the sites towards which they are channeled. For example, users who champion capitalism or globalization and want to learn more about these topics can be channeled to anti-capitalist or anti-globalization sites, respectively.

Elsewhere, Lev-On43 points out that such “tailoring failures” are caused by certain aspects of the process of retrieving information through search engines. First, currently there is no comprehensive and reliable network of keywords that properly describe the content of web documents (a semantic web). Such an absence makes it difficult not just to retrieve information relevant to a query, but also to discriminate between content based on ideological leanings.

Second, the interface of search engines is essentially textual, which mutes the richness of natural language and provides limited interactivity between the searcher and the engine which searches for him (compared to the much richer interaction between a searcher and a human who is asked to do a similar search). This disables a fine-grained understanding of the intentions behind a formal query, and limits the relevance of responses to users’ queries.

The third and last obstacle for a “perfect search” involves the way in which users formulates and articulates their intentions. A number of studies on information-seeking behaviours online reveal that users compose very short queries, hardly use advanced searching options, view a very small number of documents per query, and almost never view more than one page of results (see Spink and Jansen 2004, Machill et al. 2004).44 Spelling mistakes and non-grammatical formulations are frequent. Such information-seeking patterns limit searchers’ abilities to retrieve information tailored to their views and filter out information that opposes them.

Thus, users’ inability to utilize efficient searching strategies, the limitations of the interface between users and search engines, and the absence of a comprehensive indexing scheme online, limit users’ abilities to tailor the “ideological online space” within which they surf. Consequently, when agents use search engines even to locate information that reinforces their views, they can be directed to sites that present information and arguments opposing their views.

6 Provisional conclusions

We started by arguing that deliberation consists in the seeking and weighing of pros and cons concerning a given proposition or course of action. We emphasized the importance of exposure to opposing views. Seen through these conceptual lenses, the driving concerns for the study of the deliberative potentials of online environments are their effective capabilities to confront agents with opposing views, even against their will, and to generate due consideration of such views.

Broadly speaking, our analysis leads to the conclusion that the Internet is a mixed blessing for deliberation. On the one hand, people find it much easier to organize with, and receive information from like-minded others. The homophilic link structure of the most travelled hyper-linked web spaces can further channel agents away from opposing views.

But “drivers of opposition” mitigate the effects of these “drivers of homogeneity”. Perfectly tailoring one’s communicative environment is not all that easy. Furthermore, there are extended opportunities online for communication across political cleavages, as well as reduced cognitive pressures to express opposing views.

What are we to make of all this? The arguments presented here suggest that the deliberative potentials of different online spaces depend on the drivers, whether of homogeneity or of opposition, that dominate in each particular context.

It seems too early to formulate a comprehensive theory of deliberation online. Some of the technologies involved in online communication are still changing at a fairly rapid pace. Access to the medium is spreading, with many people still learning how to use it. Usage patterns are probably not stabilized yet. Finally, research on some of relevant dimensions of exposure to opposing views online is still in its infancy. Instead, we wish to advance a couple of limited and provisional claims.

It is fairly well established by now that intentionality drives segmentation and generates few, if any, contacts with dissimilar people. When users efficiently choose their communicative environment, they tend to build echo chambers as drivers of homogeneity become then dominant. It would be a mistake, however, to view the Internet as a medium offering to all of its users unbounded intentionality in communication. Tailoring one’s online communicative environment is certainly feasible, but it is also costly. It requires time and energy. It also requires skills. We noted that many users do not possess those skills at present. In any event such skills are costly to acquire, too. It seems reasonable to surmise that not all users are equally prepared to incur such costs. In all likelihood, many content themselves with imperfect tailoring, thereby increasing their chances of encountering opposing views. Thus the costs of tailoring one’s communicative environment set limits to intentionality in communication.

Another factor also limits the intentional search for like-minded communication partners or information sources. Like-mindedness is typically not an all-encompassing feature. Users may be of like-mind on one issue or in a given domain while holding opposing views on other issues or in other areas. People are bundles of numerous characteristics such as interests, attitudes, lifestyles, and much else. Similarity along one dimension does not necessarily carry similarity on another. This is especially relevant for online communication. As it has been pointed out many times, the Internet vastly expands the ability of agents to select discussion partners on specialized and narrow criteria. Thus, there is ample possibility that such contacts, while similar along the specific chosen dimension, will be dissimilar along some of the many other, un-chosen, dimensions. These dissimilarities in turn are potential sources of opposing views, and thereby of deliberation, on topics other than that which brought users in contact.

The critical role of intentionality in driving out opposing views suggests one last point concerning political opinions. It seems reasonable to infer that when agents are interested in political issues, sufficiently motivated to incur the costs of tailoring their communicative environment – or of learning how to do so – the drivers of homogeneity become dominant. For such agents, and more broadly for users highly committed to a given cause, the Internet offers the opportunity to build their own effective echo chamber, therefore not enhancing, and even possibly impairing, their deliberative capabilities.

However, for the many agents who do not care much about politics, are incapable of manipulating their communicative environment, or unwilling to put up with the cost of doing so, the mechanisms of segregation may not be efficient enough, and the drivers for opposition can become more dominant. Most likely, online communication enhances the deliberative opportunities for such agents.

It is worth stressing, however, that these conclusions are highly provisional. To better understand the possibilities of exposure to opposing views online we need more empirical research. For example, we need to know more about the implications of preferential attachment as expressed by the ideological composition of web-based discursive genres (such as virtual communities, newsgroups, and meetup-style environments.) We need to know more about the occurrences and characteristics of cross-cleavage communication in various Internet-based spheres, like web-based magazines and virtual communities. We also need to know more on, for example, the effects of collaborative filtering and the patterns of political information seeking online. Such research is necessary to understand if, where and how the promises of improved public deliberation online will become realities.

A slightly different version of this paper is forthcoming in Todd Davies, Beth S. Noveck, (eds.), Online deliberation: Design, Research and Practice, CSLI Publications/University of Chicago Press.