The big data divide between companies and customers

17. Mai 2022

“Big data divide” describe the asymmetric relationship between those who collect, store, and mine large quantities of data (usually companies), and those who are the target of data collection (e.g., customers). This contribution claims that overcoming the big data divide should not be the focus of ethical concern. Rather, the goal should be to safeguard those potentially disadvantaged by this divide.

One of the fundamental ethical issues associated with big data is the “big data divide”—a term used to describe the asymmetric relationship between “those who collect, store, and mine large quantities of data, and those who are the target of data collection” (Andrejevic 2014). In this section, I sketch the ethical nature of this problem by highlighting the values that make the big data divide, and transparency asymmetry in particular, problematic. We argue that, although those values provide orientation in an ethically justified direction, overcoming the big data divide is the wrong goal due to the structure of this problem. Instead, the focus should be on safeguarding those potentially disadvantaged by this divide from its negative consequences.

Structurally, the big data divide (with a focus on transparency asymmetry) involves (at least) two actors: data givers and data owners. They must be distinct to some degree such that the notions of “divide” and “asymmetry” make sense. Data givers disclose information of various types and are aware to varying degrees that they are doing so. Data owners collect this data and make use of it in various ways. In the early years of data protection (the 1970s or earlier), this asymmetry was described as mainly present between states and their citizens (Hornung and Schnabel 2009). Currently, the asymmetry is usually conceived of as existing between a very specific group of companies (those controlling large platforms such as search engines and social networks) and the customers of those companies (Coley 2017). Obviously, the situation can be much more complex. For example, states may demand by law access to big data collected by tech companies (e.g., Apple is required to store data from Chinese customers in China, and the Chinese state may access this data). However, let us for the sake of the argument restrict this discussion of the big data divide between tech companies and their customers to the Western context. In that respect, it is important to note that the big data divide concerns not only access to data but also the capacity to make sense of it—by machine learning, for example—a point that will become relevant below.

Furthermore, the digital divide is closely linked to a transparency problem—namely, the claim that the data givers (the customers) lack transparency in terms of knowledge about the existence and consequences of this divide. Thus, the notion of transparency has both a content (what data are collected?) and a process (what could be done with this data?) dimension. In that respect, transparency is a tricky concept. First, it is observer dependent (the cognitive capabilities of the person determine whether transparency has been achieved). Second, it is embedded in a temporal structure: having achieved transparency as to the content and process of data at time t1 does not mean that transparency is still present at time t2 because, for example, new methods to analyse data may have been developed at t2 that allow for insights that could not have been foreseen at t1.

From an ethical point of view, this data divide problem can be described as a combination of violating the value of equality (with respect to access to information about the content and process dimensions of big data) and violating the autonomy of those who are the objects of data collection. Arguments for equality in access to and use of big data are often consequentialist (e.g., the claim that equal access to big data may increase welfare because markets can become more efficient (Zuiderwijk et al. 2012). This consequentialist framing of equality denies that “big data equality” is an ethical goal per se (a point that will become clearer below). The reference to autonomy is usually justified by deontological reasoning—for example, by making the argument that having control over personal data is an expression of privacy, which is a fundamental human right (Rouvroy and Poullet 2009). Again, this sketch simplifies the situation; real-world situations may have additional ethical aspects that should be considered. However, let us for the sake of the argument consider only two values, namely, (1) equality as a means to increase welfare and (2) autonomy. The value of transparency has here mainly a functional role. First, it is necessary to check to what extent equality has been achieved. Second, it is a condition for acting autonomously. Unless the data giver is aware of the content and process dimensions of the data to be disclosed, disclosing them cannot be considered an autonomous act of the data giver. In the following paragraphs, we discuss the question of whether these two guiding values provide sufficient support for a goal to eliminate the big data divide. The answer is negative.

The goal to eliminate the big data divide can have two different meanings. First, it could refer to an actual elimination: all data givers are also data owners—that is, everybody has access to his or her data (knows what data are there and can understand them) and can process this data in a meaningful way. This goal is conceptually unclear, factually not achievable, and ethically wrong. It is conceptually unclear because the ontological status of many of the involved data is unclear. For example, linking data in a social network (who knows who) do not belong to a single individual, and their meaning is context dependent. The fact that person C knows that person A knows person B has a different meaning than the fact that person D knows that person A knows person B because the interaction histories of C (with A and B) and D (with the same persons) are different. Therefore, the notion of “my data” is unclear for many data types that are particularly relevant in the digital age. Furthermore, C has no access into the mind of D; this means that there will always be a data divide. Given that C and D are also likely to differ in their interests in and capabilities of achieving transparency, transparency asymmetries in particular will remain in any case. Removing those asymmetries would require forcing C and D to have equal interests and gain equal data processing capabilities: an obvious violation of autonomy and thus unethical. This argument also remains valid when C and D are legal persons (i.e., companies). Thus, factual elimination of the big data divide is the wrong goal.

A second interpretation of the goal is to potentially eliminate the big data divide. Namely, measures are put in place such that those who want equal access to big data and have (or are willing to obtain) the capabilities to reach content and process transparency can achieve this goal.[i] This is often what people have in mind when they argue against the big data divide. In addition, the legislator (e.g., the GDPR) works hard to ensure that people are empowered to gain informational self-determination as a means to diminish the big data divide.

At first sight, this interpretation of the goal is in line with the guiding values. First, autonomy is preserved, as no one is forced to become an “expert in their own data” and measures are supported that consider human competence variability (at least to a certain degree). Second, the reference to equality is restricted to its welfare aspect. For example, the open data movement assumes that freely usable data leads to more transparency and collaboration. This, however, puts a burden on the argument: measures to decrease the digital divide (such as the right to data portability) and reduce transparency asymmetry (such as regulations mandating the provision of details on data processing) must be shown to have these positive effects.

This is the place where the structural aspects of the big data divide come into place. The content and process dimensions of big data are not fixed entities but rather result from scientific enquiry: finding out what big data can reveal (content), and which methods are needed to capture that information, is an expensive endeavour. Those who already have a skilled workforce or other resources (financial, cognitive, etc.) will disproportionally benefit from any measures that decrease the big data divide. Big tech companies will be better able to crush start-ups, and individuals with greater cognitive abilities will have more opportunities to outperform those with lesser cognitive abilities.

This certainly does not mean that measures guided by the values of equality/welfare and autonomy are problematic per se. But the effect of those measures has to be considered, taking into account the social fact that the resources to handle big data and generate knowledge from it are not equally distributed—and there are good reasons not to enforce equalisation in that respect (clarifying this point, however, goes beyond the scope of this contribution).

Consequently, the ethical debate on the big data divide, including transparency asymmetry, should not focus on measures to decrease the divide and asymmetry. Instead, it should identify realistic harms that could result from the big data divide and transparency asymmetry and develop (legal) safeguards for those who are disadvantaged by the divide. This shift in focus will need an additional guiding value—nonmaleficence or harm reduction. Current data protection legislation that relies primarily on autonomy (and privacy) is poorly structured in that respect. It is insufficient to handle the negative consequences of the big data divide, which is to a certain extent inevitable in a society that values freedom and diversity.

[i] In the following, we will abstain from discussing the conceptual problems with the notion of “my data” mentioned above.

Recommendations

Big data divide is an inevitable consequence of a society that values freedom and diversity; eliminating this divide is a wrong political goal. Instead, legislation should identify realistic harms that could result from big data divide and develop legal safeguards for those who are disadvantaged by the divide. In doing so, values such as nonmaleficence and fairness may be more relevant than autonomy and privacy.