Timing the Right to Be Forgotten: A Study into “Time” as a Factor in Deciding About Retention or Erasure of Data
The impact of processing (line) and non-processing (dotted line) over time
For instance in Fig. 7.1, at t0 the curve corresponding to pro-interests is much higher than the curve corresponding to con-interests. This means that at t0 the positive legal impact which the processing provides by promoting certain interests is much higher than the negative legal impact that the same processing causes by diminishing the data subject’s privacy. Therefore, processing at t0 provides a net benefit all things considered. Consequently, a regulation permitting it also has a positive legal impact, all things considered.
We shall here focus on cases where the originally prevailing pro-processing interests are outweighed at a later stage by con-processing interests. This happens in particular when the personal information is distributed online for purposes pertaining to journalism, or more generally to freedom of expression. In such cases, there is generally a continuous diminution in the importance of the distribution of information with regard to both pro- and con- processing interests, up to the tipping point. This is because public interest, more aptly called public intrigue here, is quite fleeting, and thus the public meaning and use of information is equally fleeting. Entering any number of momentary Internet snafus (e.g., Alexandra Wallace, Caitlin Davis, Justine Sacco) reveals spikes in search activity over a matter of weeks and then a sharp drop back to insignificance.73 The ‘newsworthiness’ of content generally protects the public’s right to access the information.74 Like data freshly created (e.g., current address, purchases, body measurements) this information is relatively current, contextualized, and new before it becomes outdated, uncontextualized, and condensed or aggregated. Older personal facts are generally less meaningful for both the public and the data subjects and are thus also less used. In particular, older information about a person usually gives a less relevant clue to what and who a person “is” now, and therefore should in general be less meaningful, both for those who want to know about that person and for the person herself. There are, obviously, deviations from these general trends where certain past information suddenly may become more important to the public and/or more damaging to the data subject.75 For instance, when data subjects apply to elective political positions, their data concerning any past criminal or inappropriate behavior becomes more meaningful to the public. Here, we shall just consider the more common case when there is a continuous decrease in the importance of impacts on both pro- and con-processing interests. Consider for instance those cases where personal information related to crimes or bankruptcies is distributed and remains accessible after such events took place.76 This information is most relevant to the public for a short time after its publication because of its actuality, and then progressively loses its meaning and is used less, but continues to have a significant impact on the interests of the concerned person also because it may affect how that person is publicly perceived. In such cases, usually both impacts on freedom of expression and on privacy decrease as time goes by, but the diminution of the impact on freedom of expression proceeds at a quicker pace. Thus while at the beginning the benefit to the public would outweigh the loss to privacy, at a certain point in time, i.e., the reversal time, there is a change: the loss in privacy outweighs the benefit in freedom of expression. This is the point in time where, arguably, the data should be forgotten. In this typical context the pro-processing interests prevail over a RTBF up until a certain point in time, and after that point privacy takes the lead, as shown in Fig. 7.1.
Figure 7.2 clarifies this point by representing directly the difference between the differential advantage resulting from the favorable impact on publicity-interests (the publicity-related gain) and the differential disadvantage resulting from the unfavorable impact on privacy-interests (the privacy-related loss) obtained by processing the information. The balance is positive before the reversal-time, it is 0 at that point and then it becomes negative.
Net social value of data processing over time
7.5.2 An Increase in Pro-processing Interests Over Time
The pro-processing interests are not always declining in time, because the benefits are not always immediate. Public interest in that which is newsworthy may be fleeting, but public interest in history, social science, and cultural preservation last far longer. Think for instance of historical interests or when there is revived interest in the specifics of the content (e.g., an individual decides to run for office), as well as immediate interests met remotely when information is combined, aggregated, or reflected upon revealing previously unknown insights into the past or future.77 The difficulty is that ‘history’ may be hard to recognize immediately, the interest very likely grows over time with regard to certain data subjects instead of declines.
However, we may be able to cope with such long-term interests in different ways. Because there is a significant difference between individuals like employers or first dates searching an individual and public interest, the meaning of the information in a context can differ; the employer is looking for a specific person while the public interest generally (not always) will be focused on a certain event in its context. Wikipedia’s Biographies of Living Persons Policy draws a distinction between general public interest in the individual or the event or topic of an entry. It reads:
Caution should be applied when identifying individuals who are discussed primarily in terms of a single event. When the name of a private individual has not been widely disseminated or has been intentionally concealed, such as in certain court cases or occupations, it is often preferable to omit it, especially when doing so does not result in a significant loss of context… Consider whether the inclusion of names of private living individuals who are not directly involved in an article’s topic adds significant value.78
Based on this policy, the Star Wars Kid is not named in the entry on the Star Wars Kid.79 Wikipedia also has a deletion policy that results in five thousand pages being deleted each day, one reasoning being a lack of ‘notability,’ which requires significant coverage, reliability, sources, independence from the subject, and a presumption that the subject is suitable for inclusion.80 According to the policy, articles with unclear notability should not be deleted, but those that are clearly not notable should be and useful material preserved on the talk pages,81 which are not indexed by Google.82 Like Wikipedia, the right to be forgotten could (but does not) ask the difference between public interest and private searches in order to determine the right course of action when a user seeks to have personal information erased, as opposed to quick deletion or automatic public interest preservation. In some cases public interests may be served just as well by content that is anonymized (interference with the memory process on the level of encoding), as was done with the Star Wars Kid entry on Wikipedia. Moreover, preservation efforts could seek to conserve that personal data that may continue to serve remote needs while offering limited search access where appropriate and in this way enabling a form of “forgetting” on the retrieval level.83
7.5.3 Carrots and Sticks
To determine how to regulate “digital forgetting”, it is not sufficient to consider the interests at stake. We also have to consider the motivations of the parties involved. Let us now focus on cases concerning the publication of publicly relevant information on online platforms.
A simplified representation is provided in the upper part of Fig. 7.3 where a linear relationship is assumed between the represented interests and time. In part A of Fig. 7.3, the pro-processing curve starts at the higher level, but decreases more rapidly than the con-processing curve, so that at a switch point the two lines cross: from that point on, the damage to con-processing interests is no longer compensated by the benefit to pro-processing ones. Subsequently, processing provides a negative legal trade-off, which apparently justifies its impermissibility, and the provision of sanctions upon the processing parties, i.e., publisher/uploader of the information and the host provider who is storing it in his virtual repository (server/website/forum).
Impacts of data processing related to interests over time
Figure 7.3 also contains a representation of the motivation uploader (part B) and of the host provider (part C), both of which are also decreasing, but remain positive (assuming that no sanctions are provided).
The meaning that data has for uploaders can differ and thus their motivation to upload. The uploader’s motivation includes the economic gains the uploader expects from distributing the information (as is the case for newspapers and websites or host providers getting subscriptions or advertising), but also includes the moral and social importance one attributes to providing such information. Abstracting from different individual attitudes, we may assume that motivation for distributing information is measured by the maximum personal loss one would be ready to sustain for not distributing it, regardless of the grounds that explain this attitude.
Consider, for instance, the situation of a person who has to decide whether to upload on a blog information concerning a political or economic scandal, knowing that this may cause him some personal advantage (e.g., reputation, some chances of having a political role in the future, possible some financial gain resulting from the fact of attracting people to the blog) but larger personal losses (e.g., losing possible contracts, missing career advancements, even putting at risk one’s life or freedom, etc.). Knowing also that this information would be highly beneficial to the public, contributing to curb the plight of corruption, while damaging the data subject, the motivation of such a person would likely be measured neither by the mere trade-off of personal gains and losses, nor by adding to this trade-off the full amount of the expected (net) public benefit. It would rather be measured by adding to the trade-off of personal gains and losses a quantity expressing the limited extent up to which the person internalizes the moral/social merit of his action, i.e., a quantity that indicates what additional personal loss he would be ready to sustain to accomplish that action.
Consider for instance a piece of news being published in an online journal, and assume that after a certain point in time the legal balance becomes negative. At that point in time, the publisher will still have an interest in keeping the news online, since it may still attract readers and thus produce revenue. Thus, if there were no law in place (abstracting from the possibility that the data subject uses private sanctions of incentives), the publisher would probably continue to distribute the information even when the legal trade-off has become negative.
The motivation is assumed to be similar to that of uploaders, while being generally lower, since providers host huge amounts of materials and have a small interest in continuing to distribute a specific single piece of information. Providers have a strong interest in having a legal discipline that does not make them liable for the distribution of illegal information. However, if the battle for a general exemption were lost, they would prefer to comply with removal requests, rather than be subject to sanctions in individual cases.
We may assume that sanctions for failure to remove the data may include the compensation of the damage to the data subject, as requested by Art. 23 DPD. This compensation, according to national regulations, such as the Italian one, may also include non-economic damage. In addition, the sanctions may include administrative or criminal fines, as established by national legislation and required by the GDPR.
If such sanctions were always to be imposed upon a processing only after the point in time where the balance between pro- and con-processing interests is reversed, and the processing party knew exactly where this point is located, such a discipline would induce the behavior that maximizes the achievement of legal values. Before the reversal-time uploaders and providers would leave the material online, since they could enjoy the benefits resulting from the distribution of the information without encountering any legal sanction. After that point, they would take it down, since continuing to distribute the information would expose them to the obligation to compensate damages of the data subject, and to any further sanction established by data protection law.
This analysis however, does not consider that processing parties may be uncertain as to whether distributing certain information at a certain point in time provides a positive or a negative balance between publicity and privacy interests, being therefore lawful or rather unlawful. Or in any case, they may be uncertain as to how the competent decision maker will judge the issue. This uncertainty will very likely lead to premature withdrawal of the material by the parties involved in the distribution, i.e., at times when publicity interests still outweigh privacy interests. This anticipation will be larger when the uncertainty is greater or the motivation to distribute the material is smaller. If we assume, as it seems reasonable that uploaders have a stronger motivation to keep the material on line than providers, the expectation of a sanction will have a stronger anticipatory effect on providers than on uploaders, as Fig. 7.3 shows. Thus, uploaders and host providers would engage in premature self-censorship by honoring removal requests at times when the benefits of keeping the information on line still exceed the damage to the privacy interests of the concerned data subjects. Note that to have this effect the sanction does not need to be extremely severe: it suffices that the sanction, discounted by the probability of not being punished, overrides the motivation of the uploader. Also a punishment limited to damages (in particular when also moral damages are included) may have such a result. Hence, sanctioning the continued distribution from the point in time when the con-processing legal interests outweigh the pro-processing ones is likely to lead to anticipatory removal. Anticipatory removal would also happen when an unfulfilled request by the data subject was needed to trigger the sanction: anticipated requests would lead to anticipatory removals.
Consequently, a takedown system, where a user can simply request data be removed, requires the data controller to perform this assessment for themselves, which may lead to valuable information being removed, because there is so little guidance on how time should be incorporated into the removal equation.84 While a RTBF that adheres to a life cycle approach is better than one that does not, data controllers may not be the appropriate source for establishing a standard for interpreting exceptions. In order for the RTBF to account for the interests of the data subject, the data controller, and the public, more guidance that recognizes the digital life cycle (ephemerality of digital content and public interest, as well as the value to remote and immediate users) would certainly bolster the legitimacy and strength of the RTBF.
In a world where you are what Google says you are and digital dossiers impact automated opportunities beyond view, the RTBF plays an important role in user participation. The complication is that information removal can be just as dangerous as information storage. Digital information sources, and especially the Web, function as very large external transactive memories. Acknowledging the growing wish of individuals to counter the ‘remembering-by-default’ of this memory requires the implementation of a form of digital ‘forgetting’. However, because it is an external transactive memory, data controllers and data subjects are not the only parties to be considered, but also the interests of others: the public. Balancing these interests is difficult. We can, however, gain guidance and inspiration from the human memory process in which the factors ‘meaning’, ‘use’ and ‘time’ play important roles. ‘Time’ is a factor that generally supports ‘forgetting’ when the passed time increases, while ‘meaning’ and ‘use’ generally oppose forgetting when the meaning information and/or the frequency with which it is used increases. This makes ‘time’ a crucial element to acknowledge in relation to the RTBF. ‘Meaning’ and ‘use’ are often in some form or the other recognized by law as being important factors to retain data. For instance, the exceptions mentioned in art. 17 (3) GDPR, inter alia the freedom of expression, scientific and historical interests, are of such importance to the public that they oppose the ‘forgetting’ of the information.
However, beyond this general expression of the societal value of data retention in view of time, the exact role that time plays in current privacy or data protection law is not clear. Generally, ‘time’ can play two parts in law. On the one hand, ‘time’ can play a role as a weight in a balance of interests, as a factor adding or removing weight to the request to ‘forget’ personal information or its opposing interest, resulting in tipping the balance in either direction. On the other hand, ‘time’ can play a role as the marker of a discrete moment where the grounds for retention no longer hold and ‘forgetting’ of the data should follow.
In Sect. 7.4 the important points in time in data processing are identified, where time functions as a marker of a discrete moment in the information process. The identification of these points show that the ‘time’-cycle of data processing highly depends on the use of the data; the conditions under which the data are acquired, the purposes for which they are collected, and whether they are necessary. The analysis of the specific points in data processing shows the importance of the point in time with regard to the use of information in data processing for the invoking of a RTBF. Generally at the stages in the process where the information loses relevance for its use (at least for the initial purpose for which it was collected), the chance for a successful appeal on a RTBF is increased.
The role of time as a factor in a balance of interests is more complex. Important for this balancing is to recognize that information has a lifecycle and its value (also to the different interested parties) changes over time. Data is generally created to meet the current state of affairs in the world and has the most meaning and value in that context. The ‘newsworthiness’ of content is thus often fleeting, and information can easily become outdated, uncontextualized, and condensed or aggregated. Next to immediate needs, information can serve remote needs as it is combined, aggregated, or reflected upon revealing previously unknown insights into the past or future. Despite the fact that these information needs are important, there is very likely a point in time where the added value of personal data retention has diminished so far that the interests of the individual to be ‘forgotten’ prevail.