6.3 Digital is different

Social research in the digital age has different characteristics and therefore raises different ethical questions.

Most social research in the analog age struck an appropriate ethical balance. For example, in a review of lab experiments that collectively involved more than 100,000 people, Plott (2013) found only one adverse event, a student who became upset because of losing money in an economic game. As the previous three digital age examples illustrate, however, researchers now face ethical challenges that are different from those in the past. Generalizing from these three studies, I think that the main problem that well-meaning researchers face is that capabilities are changing faster than rules, laws, and norms. More specifically, researchers—often in collaboration with companies and governments—have more power over participants than in the past. By power, I mean simply the ability to do things to people without their consent or even awareness. The things I’m talking about could be either observing their behavior or enrolling them in experiments. As the power of researchers to observe and perturb is increasing, there is not an equivalent increase in clarity about how that power should be used. In fact, researchers must decide how to exercise their power based on inconsistent and overlapping rules, laws, and norms. To be clear, this does not mean that most digital age research is unethical. In fact, given this situation, I think that researchers have shown remarkably good judgment. The combination of powerful capabilities and vague guidelines, however, puts well-meaning researchers in a difficult situation.

Although you personally might not feel especially powerful in terms of your ability to do things to people, increasingly researchers—often in collaboration with companies and governments—have the ability to observe and perturb people without their consent or awareness. For example, imagine following a person around and recording everything that they do. This would include tracking things such as where they go, what they buy, who they talk to, and what they read. Monitoring people like this in the analog age used to be the stuff of governments with enormous budgets. Now, all of this information is routinely and automatically recorded about millions and soon to be billions of people. Further, because all of this information is stored digitally, it is easy to copy, search, transmit, merge, and store. In other words, what is routinely done today would shock and amaze Cold War spy agencies like the KGB, CIA, and Stasi. Further, much of this behavioral tracking is taking place without the full understanding of those who are being surveilled.

A vivid metaphor that partially captures this situation of mass surveillance is the panopticon. First proposed in late 18th century by Jeremy Bentham as an architecture for prisons, the panopticon is the physical manifestation of surveillance (Figure 6.3). The panopticon is a circular building with rooms oriented around a central watchtower. Whoever occupies this watchtower can observe the behavior of all the people in the rooms. And, critically, the people in the rooms cannot observe the person in the watchtower. The person in the watchtower is thus an unseen seer (Foucault 1995).

Figure 6.3: Design from the panopticon prison, first proposed by Jeremy Bentham. In the center, there is an unseen seer who can observe the behavior of everyone and cannot be observed. Drawing by Willey Reveley, 1791. Source: Wikimedia Commons.

In fact, digital surveillance is even more extreme than a person in a watchtower because it can produce a complete digital record of behavior that can be stored forever (Mayer-Schönberger 2009). While there is not yet a full recording of all human behavior merged into a master database, things are moving in that direction. And, that movement will most likely continue as long as the capabilities of sensors continue to increase, the cost of storage continues to decrease, and more of our lives become computer-mediated.

To many social researchers this master database might initially sound exciting, and it could certainly be used for a lot of important research. Legal scholars, however, have given a different name to this master database: the database of ruin (Ohm 2010). The creation of even an incomplete master database could have a chilling effect on social and political life if people become unwilling to read certain materials or discuss certain topics (Schauer 1978; Penney 2016). There is also a risk that the master database, while created for one purpose—say targeting ads—might one day be used for a different purpose, a situation called secondary-use. A horrific example of unanticipated secondary-use happened during the Second World War when government census data—the master database of that time—was used to facilitate the genocide that was taking place against Jews, Roma, and others (Table 6.1) (Seltzer and Anderson 2008). The statisticians who collected the data during peaceful times almost certainly had good intentions. But, when the world changed—when the Nazis came to power in Germany and neighboring countries—this data enabled a secondary-use was never intended. Once a master database exists, it is hard to anticipate who may gain access to it and how it will be used.

Ordinary social researchers are very, very far from anything like creating chilling effects on society or participating in human right abuses through secondary-use. I’ve chosen to discuss these topics, however, because I think they will help social researchers understand the lens through which some people will see their work. Let’s return to the Taste, Ties, and Time project, for example. By merging together complete and granular data from Facebook with complete and granular data from Harvard, the researchers created an amazingly rich view of the social and cultural life of the students (Lewis et al. 2008). To many social researchers this seems like the master database, which could be used for good. But, to some others, it looks like the beginning of the database of ruin that was created without the consent of the participants. The Taste, Ties, and Time project began in 2006, and the information that researchers had was not particularly private. But, if you look forward a bit you can imagine that these issues are likely to get more complex. What kind of digital mosaic will researchers be able to construct about students in 2026 or 2046?

In addition to this mass surveillance, researchers—again in collaboration with companies and governments—can increasingly systematically intervene in people’s lives in order to create randomized controlled experiments. For example, in Emotional Contagion, the researchers enrolled 700,000 people in an experiment without their consent or awareness. And, as I described in Chapter 5 (Running experiments), this kind of secret conscription of participants into experiments is not uncommon. Further, it does not require the cooperation of large companies. As I described in Chapter 5, researchers can increasingly design and build digital experiments with zero variable costs, a cost structure that enables extremely large experiments. Like the ability to observe, the ability to systematically perturb will likely continue to grow.

In the face of this increased power, researchers face inconsistent and overlapping rules, laws, and norms. One source of this inconsistency is that the capabilities of the digital age are changing more quickly than rules, laws, and norms. For example, the Common Rule (the set of regulations governing most government funded research in the United States) has changed little since 1981. An effort to modernize the Common Rule began in 2011 but was not complete as of the summer of 2016. A second source of inconsistency is that norms around abstract concepts like privacy are still being actively debated by researchers, policy makers, and activist. If specialists in these areas cannot reach uniform consensus, we should not expect that empirical researchers or participants will reach consensus either. A final source of inconsistency is that digital age research is increasingly mixed into other contexts, which leads to potentially overlapping norms and rules. For example, Emotional Contagion was a collaboration between a data scientist at Facebook and a professor and graduate student at Cornell. At Facebook running large experiments is routine as long as they comply with Facebook’s terms of service, and at that time, there was no third-party review of experiments. At Cornell the norms and rules are quite different; virtually all experiments must be reviewed by the Cornell IRB. So, which set of rules should govern Emotional Contagion—Facebook’s or Cornell’s? When there are inconsistent and overlapping rules, laws, and norms even well-meaning researchers might have trouble doing the right thing. In fact, because of the inconsistency, there might not even be a single right thing.

Overall, these two features—increasing power and lack of agreement about how that power should be used—mean that researchers working in the digital age are going to face ethical challenges for the foreseeable future. Fortunately, researchers facing these challenges do not need to start from scratch. Instead, researchers can draw wisdom from previously developed ethical principles and frameworks, the topics of the next two sections.