Mobility Data Mining: from Technical to Ethical

Geoffrey Rockwell’s presentation started with an example from the early days of the pandemic when Tectonix, a geospatial analysis company, tweeted a video of a visualization showing where people active during the March 2020 spring break on a beach in Ft. Lauderdale returned to after their break. Tectonix was promoting how their analysis and visualization technology combined with mobility data from X-Mode could help efforts to control the pandemic by tracking the irresponsible youth who come to Florida to party on the beach instead of social distancing.

This technology demonstration on the one had appalled some on Twitter who were surprised by how their privacy wasn’t respected, but it also coincided with stories that the Trump administration was in conversation with various companies to see if location data could be used in anonymized form to track the novel coronavirus. An official quoted in The Washington Post article that broke the story said that mobility data could, “help public health officials, researchers, and scientists improve their understanding of the spread of COVID-19 and transmission of the disease” [1]. The point of this introductory case study was that there are significant ethical issues around how mobility data is gathered, used, aggregated, and shared. The variety of ways mobility data can be gathered, the inferences possible, and the number of players who are commercializing such data raise difficult problems around principles, privacy, the application of ethical frameworks and how to develop a culture of ethical use. In the rest of the talk, Rockwell surveyed these issues in order to provide a common ground for discussion.

Why bother with ethics? Ethics is often seen as a drag on innovation. Rockwell reviewed some reasons why addressing ethics is becoming important in both commercial and academic research and development [2].

What ethical principles apply? A common starting point in data ethics is to identify common principles that should guide researchers, developers and users. Rockwell surveyed some applicable sets of big data principles like the 2017 Big Data Guidelines from the Information and Privacy Commissioner of Ontario and principles from the Harvard Business School [3]. Researchers, however, often come to different conclusions, even if they start from similar principles. For example, on the issue of using mobility data for public health we found positions like: Ultimately, if residual risk is acceptable, analysis of mobility data can be justified if it can yield actionable insights that benefit public health [4].

From these discussions spurred by the current pandemic, three major location data issues emerged: (1) anonymized data sometimes are not anonymous, (2) location data are often not representative and can exacerbate inequality, and (3) location data are a key part of the extension of the surveillance state.

Rockwell concluded the case study by asking whether this moment of visibility for location data collection could provide an opportunity to push for new media literacies [5].

Privacy. Rockwell reviewed some of the definitions of privacy and how they might apply to mobility data [6, 7, 8].

Ethical Approaches. Finally, Rockwell reviewed some of the common ethical approaches that can be used to think through situations:

• $\mathrm{■}$

  Duty-Based Ethics where rules and guidelines are used to establish the morality of actions rather than their consequences;

• $\mathrm{■}$

  Consequentialist Ethics such as Utilitarianism which judge the morality of actions based on their consequences;

• $\mathrm{■}$

  Ethics of Care which emphasizes the relationships in actions and care for others.

Rockwell concluded by asking how we can develop a culture of ethics in data disciplines.

Some general resources mentioned in the talk include also [9]. ⁴⁴4The slides of this tutorial are available at:
http://www.master-project-h2020.eu/dagstuhl-materials/.

Peter Kraus gave a short presentation of how the European Data Protection Board (EDPB) works as a European Body, composed of Members from the different Member States of the European Economic Area and the European Data Protection Supervisor. The EDPB, being tasked with ensuring the consistent application of Europe’s General Data Protection Regulation (GDPR), has issued Opinions, Binding Decisions, Guidelines, Best Practices and Recommendations. He also gave a short overview of the EDPB Strategy and the transnational enforcement done by the Member State authorities.

Concerning the topic of mobility, Mr Kraus highlighted the work done by the EDPB regarding contact tracing during the COVID-19 pandemic and presented the EDPB’s Guidelines on connected vehicles [1]. These Guidelines analyse the different legal instruments that are relevant from a data protection perspective and their interaction: the GDPR and the e-Privacy Directive. The e-Privacy Directive, as a so called lex specialis, provides additional protection to electronic communications, whereas the GDPR sets the general framework.

Following the presentation by Mr Rockwell and the mentioning of possibly applicable ethical principles, Mr Kraus highlighted that the GDPR is known as a principles-based legislation, which in its Article 5 contains the legal principles of data protection, which in many respects are similar to the ethical principles mentioned for ethical use of mobility data. This includes, in particular, the principles of fairness, transparency and data minimisation. These principles are highlighted and operationalised by Article 25 of the GDPR – Data Protection by Design and by Default.

One of the possible legal bases under the GDPR to process and to use personal data, as well as one of the requirements of the e-Privacy directive to access personal data on an end device, is consent. However, consent has a list of properties for it to be considered valid consent in the legal context. It needs to be informed, unambiguous, specific and freely given. This relates back to the principles of data protection. Further, the GDPR requires that consent can be withdrawn and that this withdrawing must be as easy to do as it is to give consent. Therefore, when using consent as a legal basis, consideration should be given to what happens when the consent is withdrawn.

One way to avoid being in scope of data protection law altogether is to effectively anonymise personal data. However, as already presented by Mr Gambs, anonymisation is hard to do in practice. Encryption and hashing are frequently mistaken for anonymisation processes, whereas from a legal data-protection perspective, they merely amount to pseudonymisation. One has to be aware of re-identification attacks that may turn a dataset previously deemed to be anonymised into one that is not anonymised at all. Lastly, it was highlighted, that anonymisation is particularly difficult for location data due to the additional context that can be derived from a specific location.

Finally, a few examples that are provided in the Guidelines on connected vehicles were given, which showcase how the legal framework can be applied to use cases such as contact tracing or usage based insurance services.

In this working group, the discussion focused on multiple aspects of data anonymity, which is extremely hard to reach [3] for mobility data [1], especially considering the fact that mobility patterns are usually highly predictable. As experts, can we recommend (and develop) ways to use technology still but with some evasions of location tracking (e.g., download offline maps and navigate with GPS), and do our best to increase people’s awareness, especially considering how we are exposed and easily manipulable [2].

Moreover, there is a need to consider the danger of predatory data collection. Sometimes, the choices are false choices (you “must” download an app due to peer pressure or to participate in ordinary social life). Therefore, we should encourage the definition of different levels of consent. Another possibility is to implement tools to help people to analyze the permissions required from mobile phone apps⁸⁸8https://privacyflag.eu or from web pages⁹⁹9https://github.com/chatziko/location-guard.

However, as said in Section 4.1, it is worth noting that the choice to participate in a project or to share something affects the final goal: the goodness of the model depends on the number of individuals that participate. And this has an impact on the individuals themselves. In the literature a scenario that evaluates this trade-off is missing: what is the impact (in terms of privacy loss) if you participate? And what is the impact (in terms of service quality) if you do not participate? Another challenge is that this must be computed and evaluated in real-time. Moreover, in the mobility context, it is not enough to have a certain percentage of the population, but we need to evaluate also the geographical area; it seems that the only possibility is to simulate this computation.

The discussion ended up with a possible trend of personal data management, providing decentralized solutions. There are solutions actually existing, such as personal data cloud, where you locally store data, but to be usable, they need suitable infrastructures and ad hoc protocols. This is not implementable today because there is no pressure neither from governments and surely not from companies. We need to ask ourselves if we have the power to influence choices, both at individual and at companies levels.

A follow-up article on the concerns of surveillance and the best practice we can adopt to contrast them has already been published [4].

The question of data ethics and technology ethics has been a hot topic for some time. Over the past decade following an explosion of the availability of various and voluminous data sets often referred to as big data, a common laminate has been that technology is often developing faster than the regulations that might keep the implications of those technologies and data uses at bay in step with the innovations in the field. The mobility data analysis field is not excluded and efforts are underway to engage with and “update” ethics thinking to new data and technology realities. Those efforts raise a number of questions. In particular, we discussed the questions described in the following Sections 4.3.1–4.3.3. As a result of the discussion, we discussed existing guidelines and drew up a tentative set of guidelines tailored to the needs of a graduate student doing a mobility data project, provided here in Section 4.3.4.

Debates about the ethics of mobility analysis tend to focus on the data and what it reveals. They are often framed in light of privacy and the problems with guaranteeing privacy given the specificity of spatio-temporal data compared to other categorical data. To take a step back from this perspective, we explore what kinds of ethics questions surface if we think about the ethics of mobility data analysis beyond the data. Engaging in such an exercise is helpful to consider mobility data and its analysis as part ofmobility information infrastructures. These infrastructures include broader institutional infrastructures that make mobility research possible (think servers, data specialists, multi-origin data sources) and political and economic infrastructures that influence research priorities.

As with many other fields, mobility data analysis used to grapple with relative data scarcity. New data collection and processing capabilities have transformed how mobility research is done. More and more actors interested in technological innovation and monetising data are entering the field. Such changes bring opportunities and concerns. These require a broader engagement with ethics – with what it means to do good. The lure of innovation itself can be a concern as it often trumps a real need for innovation. Use cases are frequently found after the fact. Given economic interests that undergird the field, it is not immune to a mentality of “move fast and break things” that leaves little room for considering ethics and long-term implications of new mobility data ecosystems. The concerns raised in our discussion centred around questions of consent, equitable access, reproduction of power asymmetries, and fundamental questions about who should benefit from mobility research how.

Our discussions involved thinking about issues linked to a digital divide, both in terms of data produced and the mobility tools that data is used to create. New ideas about mobility as a service, for example, are frequently thought about in terms of creating sustainability and efficiency. Still, social aspects require considering a diversity of stakeholders who might be disadvantaged by hyper-efficient and environmentally sustainable systems. City building always tends to happen both as a byproduct of natural growth and the engineering of cities – cities are arenas where there is an interplay between policy and new knowledge. At the level of ethics, this raises the question of what kind of city we want to live in. It seems complicated to imagine a truly ethically ’optimised’ mobility data application without concerted efforts engaging with that question.

The question of third-use scenarios and the pivoting of services that might have been conceptualised with good intentions is another concern that looms large in the field. Beyond this, there is an increasing growth of hybrid applications that take advantage of mobility data information infrastructures. One prominent example is how developers might draw on mobility choice data for identifying investment areas – accelerating problems with already tight housing markets in many urban areas across the globe.

Questions of access and equity also matter for how mobility data becomes bound up in platforms and projections of how mobility should work in the future. For example, while visions for developing so-called EU “mobility data spaces” are framed as openly accessible platforms, de facto who registers to those spaces is limited, as technology knowledge and resources required to access such platforms are pretty high. This might perpetuate inequities as it consolidates the prominent role of large corporations and creates dependencies that may not necessarily be just or desirable.

There are different technological solutions envisaged to bridge these different access inequalities. For example digital identity management may offer inclusion opportunities also to people without legally recognised identification [1]. At the same time, a drive towards digital identity verification is often under scrutiny for their surveillance potential. In terms of facilitating access to data, debates hone in on interoperability standards or so-called mobility data marketplaces.

Privacy remains a significant value. Privacy and how its protection is interpreted requires critical evaluation. Issues that arise are those linked to reproducing prevailing power inequalities. For example, in debates on who ought to access and share what kinds of mobility data, customer data is often excluded as too sensitive. In effect, this means that this data stays with those who can harvest it. It also means that those with access to more detailed customer data have a competitive advantage over others as “innovative” mobility platforms grow. While it remains unclear how to counter this type of development, it is important to re-emphasise that debates about a multiplicity of principles – with privacy being one – have to continue to ensure that mobility research does not consolidate undesirable social patterns.

At the same time, innovations in the field are also shaking up who the relevant players are. Some startups – like, for example, Citymapper – might perform better at facilitating mobility for their subscribers than the transport operators themselves. This eventually also gave them a stake in the mobility debates, even though their practices might not fall under the same regulatory frameworks as public service providers. This is not to say that such changes are necessarily ethically suspect – various business models are emerging, some of which sincerely have social and ecological sustainability at their heart. Examples here might be the establishment of data trusts in some Canadian cities that conceptualise mobility data as a public good. Similarly, the ideals of Gaya X at the EU level are pointing us to a rethinking of priorities. Still, a prevailing pattern is strong incentives to cut corners to stay competitive.

In this group, we considered issues around the dignity and rights of non-humans when it comes to location data. We track, for example, land animals (individuals and herds) and fish, for all sorts of reasons. The location data can be used to study them or to hunt/fish them. This raises a number of ethical questions.