NETWORKING NOT WORKING

Even if political polarization did not wait for the internet, it is hard to be entirely sanguine about the effects of the virtual social networks and the internet on our policy preferences, and the ways they are expressed.

There are other concerns as well. The first is that the circulation of news on social media is killing the production of reliable news and analysis. Producing fake news is of course very cheap and very rewarding economically since, unconstrained by reality, it is easy to serve to your readership exactly what they want to read. But if you don’t want to make things up, you can also just copy it from elsewhere. A study found that 55 percent of the content diffused by news sites and media in France is almost entirely cut and pasted, but the source is only mentioned in less than 5 percent of the cases.⁷⁶ If a piece of news produced by a team of journalists is immediately cut and pasted onto many other sites, how does the original source get rewarded for its production? It is no surprise that the number of journalists in the United States has plummeted in the last few years, going from nearly 57,000 in 2007 to almost 33,000 in 2015.⁷⁷ There are both fewer journalists in total and fewer journalists per newspaper.

The second concern is that the internet allows for endless repetition. The problem with echo chambers is not just that we are only exposed to ideas we like; we are also exposed to them again and again and again, endlessly, throughout the day. The fake users used to “boost” stories on Facebook plus the real persons paid to “like” content accentuate the natural tendency for some messages to be repeated and acquire a life of their own. The endless repetition whips people into a frenzy (much like the way political demonstrations use repeated chants), making it harder for them to stop and check the stories.

And even if the truth eventually gets out, the many repetitions of a falsehood can raise the salience of a divisive issue and harden extremist views. We remember only the endless talk about Mexicans (who we never trusted in any case) and not the fact that first-generation immigrants, legal or otherwise, are actually less likely to be criminals than native-born Americans.⁷⁸ This of course creates a very strong reason to flood the markets with alternative facts. A hundred and fifteen pro-Trump fake news stories that circulated before the 2016 presidential election were viewed thirty million times (pro-Clinton fake news stories also existed, but they were viewed only eight million times).⁷⁹

The third is that the crabbed language of internet communication (which Twitter takes to an extreme) encourages directness and abbreviation, contributing to the erosion of the norms of civic discourse. As a result, Twitter has turned into a lab for trying out the latest nasty pitch. Political entrepreneurs are happy to plant their wildest claims on Twitter and watch them play out, with an eye to whether they have gone too far. If it seems to be working, at least among the targeted group (as measured by retweets or likes, for example), they add it to the pack of potential strategies for the future.

Fourth, there is automatic customization. In 2001, when Sunstein was writing about echo chambers, he was worried about the opportunity users have to choose the news they consume. Increasingly, there is no need to choose. Sophisticated algorithms use machine-learning prediction techniques to try to figure out what we might like based on who we are, what we have searched before, etc. The objective, quite explicitly, is to get people what they like so they spend more time on it.

Facebook came under pressure for the algorithm it used to push stories to users, and in 2018 it promised to reprioritize its feeds, putting posts from friends and family ahead of media content. But you do not need to be on Facebook for this to happen. On Esther’s Google home page on July 2, 2018, there was one article from the Atlantic, “The Trade Deficit Is China’s Problem”; Paul Krugman’s latest op-ed in the New York Times; one article from the New York Times on millennial socialists; one article on the soccer World Cup; one article from the Boston Globe on Lawrence Bacow, the new president of Harvard; one article on Simone Veil’s burial; a Huffington Post article on Senator Susan Collins’s view on the choice of the latest Supreme Court justice; and the unavoidable article about the Pixel Watch. There were only two stories she was not obviously interested in: one about a criminal escaping a French prison by helicopter (which turned out to be lots of fun) and a piece on Fox News about Busy Philipps fighting with Delta Air Lines for rebooking her and her child on different flights. That last piece was her entire exposure to right-wing media for the day. Such customization is ubiquitous. Even the National Public Radio app (“NPR One” to the cognoscenti) calls itself “Pandora for Public Radio,” referring to the app that gives you the music you like based on what you have listened to in the past. Within the echo chamber for liberal ideas that is NPR, an algorithm will filter for the user exactly what the user likely will want to hear.

This matters because when users actively choose what they are reading, they are at least conscious of what they are doing. They may prefer to read articles from familiar sources, but be sophisticated enough to acknowledge their own biases reflected in those sources. An unusual experiment in South Korea demonstrated that this kind of sophistication is very real. From February to November 2016, two young Koreans created an app offering users access to curated articles from the press on topical issues and regularly asked them their opinions on the articles and on the issues themselves. At first, all users received a randomly selected article about each issue. After a number of rounds, some randomly selected users got to choose the news sources from which they received their articles, while others continued to receive randomly selected articles. The experiment yielded three important results. First, users did respond to what they read: they updated their views in the direction of what was being presented to them. Second, as expected, those given the option chose articles generally aligned with their partisan preferences. Third, however, at the end of the experiment, those who got to choose their articles had updated their preferences more than those who did not, and they had generally updated toward the center! This is the opposite of the echo chamber effect. On balance, the option to choose slanted material made users less partisan. The reason is they understood exactly how biased the source they chose was, and partly undid the bias, while being receptive to the information; whereas with randomly assigned stories, users did not recognize the bias and therefore remained skeptical about the content, not changing their opinion much.⁸⁰

It would be very interesting to replicate this experiment in the United States. The effect may also depend on how politically engaged the reader is. It is not entirely clear that many internet readers in the US make a conscious effort to correct the bias in what they read. But this study suggests a key problem of seamless customization: its very seamlessness. Correcting slant requires an understanding of what the source’s slant is. When we always read news from the same source, we are familiar with it. But when an algorithm is serving us articles from all over the internet, some of which comes from known sources and some from more unfamiliar corners, and some of which may be entirely fake, we won’t know how to read those signals. Moreover, because we have not made the choice ourselves, we may not even remember to make the correction.