A revolution in scientific communication is underway, as the web becomes the predominant medium of publication and as individual papers are gradually linked together into a continuous network of information. The benefits will be enormous, as almost everyone agrees, but in discussing the advantages of this new technology, it is easy to forget the law of unintended consequences. The scientific community is a vast network of individual researchers who communicate with each other largely through the scientific literature. When the fundamental rules of connectivity change, we should expect emergent changes in the system itself, not all foreseeable and not all necessarily good.

The changing pattern of scientific communication is part of a larger trend, the implications of which are discussed by Cass Sunstein, a prominent American legal scholar, in an interesting new book entitled Republic.com. As technology advances, he argues, consumers have greater power to choose their own news, entertainment and other sources of information. Web filtering technology allows individuals to customize their news according to their tastes and interests, often with great precision and sophistication. As a result, people are increasingly exposed to an information diet that is largely of their own design, a 'Daily Me' in place of a daily newspaper.

This is not necessarily a good thing, according to Sunstein, whose thesis is that unfettered consumer sovereignty leads to an isolationism that is unhealthy for a democratic society. The more people filter the news they receive, the less they will be exposed to anything that they did not plan to see. As a result, they will be more ignorant of other viewpoints, and of the interests of other people. Moreover, the tendency will be self-amplifying, because people become interested in things they are exposed to, and are typically less interested in things about which they have no knowledge. Filtering also reinforces biases; people who get their news from liberal web sites, for instance, will underestimate the extent to which left-wing views depart from the political mainstream. In such a system, extremism will thrive, political consensus will be weakened, and the sense of community based on shared experience will be diminished.

Sunstein's focus is on national (particularly US) politics, but his argument seems equally applicable to the scientific community, which has embraced the idea of consumer sovereignty with almost unqualified enthusiasm. Alerting services allow readers to be notified whenever a paper appears on a given topic, from a given author, or that cites a particular paper. These recommendation services will become more sophisticated in future, as they start to incorporate user profiles and statistical comparisons with other users (“People who read paper A also read papers B, C and D”). Meanwhile, institutional site licenses and standardized formats for reference linking will remove the arbitrary barriers between journals, making it easy to move from any given paper to those that cite or are cited by it.

These changes can hardly fail to alter people's reading habits, as the 'browsing impulse' is forced to compete for mental energy with an ever-increasing volume of specialized yet highly relevant information, a scientific version of the 'Daily Me'. This will lead to two effects. First, there will be a loss of serendipity; readers will be less likely to be exposed to topics that they did not plan to read about. Second, there will be a loss of shared experience. If researchers read mainly what they have selected for themselves, the chances that two people from even slightly different fields will have read the same papers will be reduced.

This is exactly the scenario Sunstein warns us of, and the political analogy can be pushed even further. The traditional system of science communication resembles a deliberative democracy, in which journal editors serve roughly the same function as elected representatives. (Editors are 'elected' in the sense that their influence depends on their journal's reputation in the eyes of contributors and readers.) In contrast, a system based on citation data and usage statistics is more akin to direct democracy, government by referendum. The classic argument against the latter system is the danger of 'tyranny of the majority', and in any numerical method for identifying papers of interest, it seems inevitable that small disciplines will be swamped by bigger ones: as (say) Rhode Island is to California, so is psychophysics to molecular biology.

What does this mean in concrete terms? Several dangers can be identified: wide reading is often a stimulus to creativity, and fewer serendipitous encounters with papers from other disciplines may mean fewer good ideas. A lack of exposure to a broad range of papers will also have a ripple effect on other forms of scientific communication; who, for example, will invite psychophysicists to speak at departmental colloquia, or support their funding applications, or vote to hire them as colleagues, if nobody knows what they are doing? Moreover, if the importance of science as news is diminished, then so too will be the sense of community that arises from sharing it; simply put, if researchers are not reading the same papers, they will find less to talk about.

Less tangible, but perhaps no less important, are the possible long-term effects on how the scientific community sees itself. Science has been compared to a republic; lacking any central authority, the research enterprise is self-governing through the links between its members. These links exist because an expert in one area knows enough to evaluate the validity and significance of work in adjacent areas, and by extrapolation, s/he assumes that experts in those areas have similar competences that lead further afield. Through the resulting network of trust, neurophysiologists are willing to accept claims in fields ranging from paleontology to particle physics, even if they could not begin to evaluate those claims for themselves. This network of trust, however, is only as strong as the links that it comprises, and if these are weakened through increasing specialization, it seems inevitable that researchers will eventually cease to perceive themselves as part of a unified community.