The ethics of data self-reporting: important issues and best practices

1.1 Function creep

Data have a tendency to function creep: they often go on to be reused for many other purposes than the ones initially envisaged. The technology used to organise data collection tends to function creep too. This is true both in private and public sector projects. It is worrying because if data are not risk-managed in the best way, disclosures and leaks that lead to individual and group harm are more likely.

In the private sector, the emergence of surveillance capitalism (Zuboff, 2019), and the lively debate discussing it, has highlighted the deceitful and pernicious intrusions of privacy and the subsequent behavioural manipulations that US tech giants have built their empires from: an almost two decades long thread of surreptitious projects to take ever more and diverse data about users; flanked by an assertive and crafted rhetoric on the value of sharing, emancipation, community, and peer2peer entrepreneurship; that increased the ability of these giants to predict, modify, and generate human behaviour to the point that legitimate concerns have been raised as to the effects of big tech services on just about anything social including politics, markets, and mental health. The cost of surveillance capitalism, as Zuboff observes, is human futures. The model has been imitated by companies throughout the sector to the point that both practitioners and public, in different ways, are failing to imagine how tech can be built without relying on the sale of user profiles and behavioural modification. Innovative organisations focusing on privacy-first products are struggling to be seen and break through the current market chokehold. For this reason, the offer by Google and Apple to collaborate in developing an interoperable framework for contact tracing met mixed reception. In a time of great emergency, it was congenial for national COVID-19 response efforts to develop their apps on top of this tech giant-concerted framework. The UK tried to independently develop the same functionality for its NHS contact tracking app, only to backtrack. The tech giants were praised for munificence and sense of civic responsibility. The framework also employs gold standard privacy-preserving techniques approved by privacy experts. It was easy to see, however, how self-serving the move could be to white-wash the giants’ reputations with the very same broad framework they had been tarnished with – individual-level, comprehensive digital surveillance (Kitchin, 2020). The offer to help fits a history of manipulative attitudes to the public discourse.

The legitimisation of surveillance technology, methods, and frameworks as a response to the pandemic emergency has been such that even organisations who have been known to the public for nothing but scandals have had a chance at ‘covidwashing’. For instance, Israel’s contact-tracking app was developed by NSO Group, a secretive spyware organisation notorious for its services to authoritarian governments across the world (Kitchin, 2020). Similarly, Palantir and Experian, other actors of the digital economy of questionable ethics reputation, have also taken part. Observing how easily techniques from the most controversial commercial surveillance practices can be transferred all the way to state infrastructure for pandemic response should make us ask what we are dealing with. For companies such as Palantir, the opportunity was not only reputational, but also an opportunity to turn private technology into state infrastructure, with the commercial benefits that can ensue in the long run (Tempini et al., 2022). The means are the translation, adaptation, and re-deployment of technology, methods, and resources used in population surveillance. The aim is to become indispensable infrastructure of public health response and prepare state administration processes for further penetration through new drives of technologically, and organisationally, compatible systems provision. Case in point, after providing an analytics infrastructure crucial to the coordination of the covid response, Palantir now looks set to win an unprecedentedly valuable and wide-ranging contract for the provision of future NHS digital capabilities (Financial Times, 2022) extending from the infrastructure already provided during the pandemic. The strategy is right out of the infrastructure studies playbook.

Despite the scandals revealed by a number of whistleblowers and leaks (e.g., Snowden on the PRISM program (Wikipedia, 2022a) reaction over recent years has been mixed and, so far, shy of introducing game-changing protections; such is the attraction that the public sector feels in respect to these new sources. The single-most impactful law recently issues on the topic, the European Union (EU) (UK) General Data Protection Regulation (GDPR), has had an uncertain track record as a sensitivising device, as broad patterns of use have not shifted. It provided great flexibility in implementation. In it, the opportunity for technology providers’ is implementation fragmentation. If every provider implements the regulation in a slightly different way (as in the myriad different ways to do ‘cookie’ language, explanation, and notification interface), user burden in expressing and articulating privacy preferences is highest. It is possible that it is rather a string of publicised scandals, such as the Facebook/Cambridge Analytica one, involving the threat of election manipulation and disinformation warfare, that has finally started to bite.

Internet researchers, in the Internet and beyond, should also question and account for the ways in which they might be benefiting from such widely criticised ways to generate data, and the ways in which their own practice can relate to function creep issues: “Similarly, harvesting sensitive information from public–private environments such as social media may raise ethical issues, especially for research involving vulnerable populations who may have limited understanding of the implications of disclosing personal information on these platforms” (Clark et al., 2019).

In the public sector, large trends in networking and ubiquitous computing have brought about an explosion of digital watching as made possible by a highly granular fabric of recording devices; and a network of interoperable databases that are increasingly interconnected for fast, increasingly real-time, networked access. From CCTV cameras, which the UK has led the way in ubiquitous deployment and acceptance, to a panoply of digital listening devices, huge amounts of data can be made available to various agencies from the law enforcement to public administration. The value of data and the pressures to fuel the growth of a competitive digital economy have created incentives for governments to enable the regulated access to previously unavailable data even down to the individual level. The unveiling in Spring 2021 of plans by the UK Government for the regulated access and re-use of NHS England and Wales data (see General Practice Data for Planning and Research (GPDPR) reporting; (Machirori and Patel, 2021)) is the latest of a series of attempts to inject these data in the digital and research sector economies (Vezyridis and Timmons, 2017) and signals the determination on the part of policy makers to find the ‘right’ conditions that will make this digestible to the public. Neoliberal politics and New Public Management ideology of public administration have captured the imagination of policy makers in matters of technology innovation and management. They have made for social, political and financial pressures on state institutions to try to generate value out of any viable public asset. But the protracted occupation by private sector of public infrastructure provision has the effect of stifling imagination and discussion of alternative ways to develop public technology (Sharon, 2020; Tempini et al., 2022). For a long time, scholarship has further pointed out how technology infrastructure and standards have an inertia that makes radical revision as difficult as function creep is easy (Hanseth and Lyytinen, 2010). Once important social and organisational interdependencies have developed over information and communications technology (ICT) that saves as infrastructure, it is very difficult to unseat it. From the care.data debacle (REF) to the latest GPDPR move, NHS data have been one of the fields in which the pressure to data function creep is highest, raising the appeal for government to try to avoid the turbulence of inevitable questioning by trying to get plans pass scrutiny as quickly and silently as possible. As the Ada Lovelace Institute observes, this fits a pattern of a “decide, announce and defend” approach to the public focused on its persuasion instead of its involvement (Machirori and Patel, 2021; Ada Lovelace Institute, 2020). It is a self-defeating strategy, as each time, the plans have been halted by public debate that is only in part focused on the merits and demerits of the envisioned ways of using data and doing research with them; and is focused, in the rest, on public issues of governance, accountability, and trust. The GDPR’s roll out has now been indefinitely frozen (O’Donovan, 2021), as the size of deliberate opt-outs from a distrusting public was concerning. The pandemic has offered a few tell-tales on the issues and risks of data function creep. The UK government attempted to extend regulations of pandemic emergency that loosened constraints on the access to sensitive health data (Tempini et al., 2022; O’Donovan, 2021), but its failure to attend to public trust led to new failure (Wilson, 2021; Bharti et al., 2021).

The public deserves much better, more responsible approaches to data governance accountability and control of function creep. The public should be involved in the deliberation and articulation of values and aims embedded in projects and infrastructures using data about it (Bharti et al., 2021). Projects should be paramountly concerned with gaining public trust. At a minimum, Kitchin recommends that: “Citizens should know precisely what the app seeks to achieve and what will happen with their data. There should also be safeguards to stop control creep and the technology being repurposed for general or national security, predictive policing or other governance or commercial purposes” (Kitchin, 2020). And project managers should be concerned about the cumulative effects that straining public trust with exploitative or unaccountable data management can have in the long-term. In pandemic times, such risks are accelerated as function creep can ensue very quickly. An example is the COVID-19 Symptom Study, data self-reporting app ZOE (ZOE, 2022), launched by a private company and academics at the start of the first wave. Its funding was quickly secured, for the ensuing two years, by the government. Only few months later, when the research team started publishing promising results over the power of self-report symptom data to predict infections, authors observed the trend for the ZOE data to be “increasingly being linked to the public health response within the National Health Service (NHS) in the United Kingdom” (Drew et al., 2020). In the immediate, this could be seen as a good example of collective mobilisation and ingenuity in the face of exceptional challenge. But it poses concerns over the risks this shift could mean for the livelihood of individuals (See Box 1).

These are not only ethical concerns. They often become epistemological if people start to take countermeasures which in turn affecting the quality and reliability of the data (Bharti et al., 2021). As about any health researcher knows, opt-outs are not randomly distributed. Just as many other individual preferences and social facts, they introduce bias in experimental and observational research (Teira 2013). When sizeable, they quickly lead to issues statistical representation of the remaining sample. Thus, Bharti et al. (2021) stress how public involvement in the articulation of values and aims of a data project is not only politically beneficial but also leads to better outcomes. The GPDPR is a case in point, as the value of the proposition was tarnished by huge spikes in drop-out rates (O’Donovan, 2021). And as the controversies around the ‘pingdemic’ demonstrated, allowing an app to report information can give a third party the power to interfere with our everyday life at a latter point, on terms one might not be happy with. Perhaps ironically, many deleted the contact-tracing app once it started doing what it was supposed to do and ‘pinged’ them out of circulation. Contact-tracing apps are designed to discipline and reshape spatial movements and social connections (Kitchin, 2020), but they have an uneven impact. Essential workers are likely to face higher costs of ‘protecting others’ policies, if the government does not provide for any other forms of support. The complex dynamics of social systems (Wilson, 2021) mean that these workers will have very high incentives to flout unfair rules. Kitchin notes that data function creep can also be indirect. Location data that is automatically generated, for public health apps, by a smartphone operating system’s location; services can be shared by the operating system with other apps that also use location data; with the result of deepening the surveillance of a user within the ecosystem of data brokers and surveillance capitalism.

1.2 Consent and privacy

Research with Internet data has been raising ethical concerns related to consent and the distribution of harm and benefits for quite some time. Data science and artificial intelligence (AI) methods often rely on the use of large amounts of real-world data to train machines, and these data have often been taken where they are most easily available, e.g., much of the Internet. Internet users directly generate data as they post, record, and interact over platforms, but data are also continuously generated by the underpinning technological infrastructures as a matter of mere technical operation. The whole array of data generation and storage instances is impossible to keep track of (Clark et al., 2019). Intense debates have been generated over the ethical standards that new forms of digital research should adhere to (Petermann et al., 2022); including questions over the double standards that public- vs private-sponsored research follow. Questions of consent will be important for self-report data collection initiatives. If data could be reused multiple times a whole host of questions are raised as to what could be the appropriate mechanisms for granting consent for uses other than those originally envisaged at the point of data generation.

Calls for ‘blanket’ consent that pre-emptively authorise a broad spectrum of re-use (and the milder version of ‘broad’ consent) have been heavily criticised for their vulnerability to misuse, capture and manipulation. Also heavily undermined have been individualistic arguments that envision each research participant deciding on the use of their own data as the most desirable ideal of data governance – most impractical but also unethical to dump all the responsibility onto isolated individuals. Organisational arrangements that require more complex governance procedures and accountability structures have been better received.

Self-report data, especially when health-related, are often very sensitive, giving rise to the potential for direct and indirect harm from misuse, manipulation, and profiling. It is thus important for contributors to be able to keep track and access contextually relevant information as to what is done with their data, and what is envisioned for the future. They should also be made aware that parties interested in accessing and using their data, because ethical standards and guidelines as to what are admissible re-uses of individual data change over time and across research domains and institutions, might not always think that contributors have a right to informed consent when:

Given the higher exposure to diverse and distributed expectations and engagements, high standards should be required of initiatives relying on self-report data. To be meaningful, question of consent should be renewed at major project milestones or updates. Participants should be actively invited to consider whether changing situations fit them and should not be expected to keep track and make sense of change by themselves. For instance, if the governance of a project changes due to change in ownership or management, self-report data should not be considered a conventional asset that can be sold on. Instead, they should be treated as objects that bind different people together in an economy of relational ethics (Prainsack, 2019a, 2019b; Birhane, 2021). Data generation and access should be minimised as much as possible. Many questions require functioning ethical oversight.

1.3 Contribution and participation

Self-report data collection initiatives can achieve impressive feats thanks to the economics of crowdsourcing, where a large number of contributors each invest only a small portion of their time to a great cumulative effect. However, it is not easy to run these initiatives over a long time, and how to do this will be particularly crucial for those initiatives whose data collection becomes more valuable the more it has taken place regularly and over a long period of time, e.g., data about health events and symptoms used for the understanding the spread of disease.

The emergency of the pandemic, with the emotional response and mobilisation it generated, along with the lockdowns and the way many people redirected their attention and free time to activities that can be completed at home, created one of the easiest scenarios for a self-report data collection project to succeed. Many people wanted to understand and learn about COVID-19 and needed reassurance and explanations on their daily experience. Many wanted to help. It was easier to accept intrusions of privacy, the boundaries and consequences of which were not fully clear. And still, even in the times of pandemic, several challenges arose to the motivation and distribution of participation. The UK ‘pingdemic’, and the consequent migration of many users deleting the contact tracing app from their phone, spoke of an unresolved tension on expectations and boundaries of the data reporting that contributors are voluntarily enabling; a tension that was only exacerbated by the experimental nature of the methods involved.

A rich literature on participation in data self-reporting offers many points of concern to keep in mind for future self-reporting projects. The seminal ‘ladder of participation’ published by Sherry R. Arnstein (Arnstein, 1969) has since served to highlight how nuanced the concept of participation can be, open to manipulation and, as a result, how appearances can be misleading. An extensive literature has followed to question the concept of participation further. Self-reporting projects distributed over the web are often opaque as to their inner, complex workings, while very public about the moral economy that they want to draw on to mobilise support, with frequent calls to share for the common good, for altruism, emancipation, empowerment of oneself and their kin. On close scrutiny, participatory practices can turn out to be empty, tokenistic, or extractive ‘crowdsourcing’. Kelty and colleagues (Kelty et al., 2015) point out that participation on the web would be better understood as something that is not a linear spectrum along one axis, but rather as something multi-dimensional. In this way, the limited openness to participation of most web projects is easier to observe, with most projects restricting participation to one or few dimensions.

It is crucial that self-report data collection projects respect the effort and investment put in by contributors to make it possible by relating to them as invaluable partners and project stakeholders. Extractive models popular in the Silicon Valley, that promote the value of sharing and promise empowerment and emancipation, only to extract information from individuals while excluding them from benefits (H. R. Ekbia, 2016) and governance of a project, will come under fire and will be increasingly unpalatable, given the recent mood change in public sphere discussions on these matters. Participation in health research involves a particular kind of labour that involves turning one’s body into readable and available to observation and cognition (Brives, 2013; Milne, 2018; Cooper, 2012); and even in the case of unambiguously commercial projects where crowds are called on to contribute cognitive labour in exchange for financial retribution, as in Amazon’s Mechanical Turk, strong arguments have been raised that point to the exploitation that these models build on (H. Ekbia and Nardi, 2014; Irani and Silberman, 2013; Nardi, 2015).

It is beyond dispute that many projects have exploited contributors who are asked to volunteer time and information while they are kept out of any relevant sense of ownership of the results. In light of much literature, it is clear that it is difficult to develop a self-report data collection project without controversy. The asymmetries of projects with a very large, distributed base of contributors and a very small team of managers and developers will make for very sensitive politics of contribution and participation. Managers of data self-reporting projects should avoid exploiting a common double standard where the efforts of the contributor base are celebrated and recognised with the language of empowerment and ‘bottom-up’, or ‘patient-led’, research while at the same time, individual contributors are excluded from formal recognition when the research is published in peer-review outlets. Only a few projects have tried to recognise the individual contributors that a crowd is made of through authorship credits, and this not always be possible nor desirable. There are many other ways to formally recognise the contribution of the public in these projects other than scientific credit. Direct involvement in governance and management can be an alternative, as well as having formal and public ways to gather and respond to request, motions and value demands.

It is important, and even more so once out of pandemic time, for self-report data collection projects to consider very seriously how they can best and actively engage in transparent communication about all aspects of the project, including decision structures and finalities (See Box 2). The projects that do not cause controversy are likely to be those that engage in a sustained dialogue with the base of contributors; explain key changes in governance and direction of the project; demonstrate accountability to the contributor base at least as keenly as they strive towards other stakeholders such as sponsors, third party researchers, or regulators.

1.4 Self-report and self-harm

Data can be used for purposes other than initially envisaged in many different ways. Contributors themselves, once engaged and active in the data collection, can repurpose or imitate the methods and inclusive rhetoric of the projects they joined, and extend aims of the data collection and networking taking place on a self-report data collection system to organise research on matters of their own interest.

There is a long history of this this kind of initiatives in health research, with perhaps the most historically significant one that captured by the pioneering work by Steven Epstein (Epstein, 1996), describing how in the midst of the US AIDS epidemic, the AIDS Coalition to Unleash Power (ACT-UP) (Wikipedia, 2022b) activists aiming to influence research on HIV and extend access to experimental therapies harnessed existing networks of civil rights activists to coordinate independent research, protest and sabotage.

Since the rise of the web, techniques and methods for organising independent research that exploits available resources, data, and technology have grown more sophisticated. Examples have multiplied as we saw the emergence of self-report data collection beyond the domain of scientific research and into the domains of civic and political action, where data have been used by activists to mobilise evidence in support of their claims (Milan and Velden, 2016; Bruno, Didier, and Vitale, 2014); as well as artistic performances and other individual applications of data collection on the conduct of one’s life, in the culture of the Quantified Self movement (Sharon, 2017). In the field of health research, the self-organisation of data collection through web networks and technology has been called ‘patient-led research’ (Vayena and Tasioulas, 2013) in the wake of some celebrated examples (Paul Wicks et al., 2008; Paul Wicks et al., 2011) that excited commentators about the social web’s potential in scientific production.

Exciting as it might seem for ordinary individuals to take knowledge matters into their own hands, there are a number of thorny ethical issues associated with distributed contributors’ ability to organise data collection on the side (Vayena et al., 2015; Tempini and Teira, 2019; P. Wicks, Vaughan, and Heywood, 2014; Ledford, 2018). It is clear that people enrolled on these initiatives often interpret the data thus generated in order to make life decisions with potentially large consequences. For instance, patients might make changes in their treatment course, or take uncontrolled chemicals. And when contributors are many, different aims, expectations, levels of engagements and degrees of literacy on the subject matter are at play. Some will be much more vulnerable than others. They have many questions that they might seek to satisfy with what they find, whether the answer is provided through state-of-the-art methods and project level supervision or not.

These projects complete lack of ethical supervision, let alone of the sort that has long been required of publicly funded university research. No one will have a view as to who these people might be and how they should be supported. What these initiatives risk creating then is something short of well-developed solidarity, but rather, an ephemeral alliance and confluence of interests between individuals who might then be find themselves to face the consequences alone. The regulatory trend towards allowing patients more freedom to experiment with treatment courses means that the risk calculus is becoming more complex just as patients gain more freedom to do it themselves (Carrieri, Peccatori, and Boniolo, 2018; Navarro, Tempini, and Teira, 2021; Tempini and Teira, 2020).

Pandemic time is only likely to exacerbate the problem. With COVID-19 we have seen the emergence in social media and public sphere of a number of theories, since debunked, as to the causes of illness and more or less implausible treatments that are within easy reach of each citizen with a curious mind and misplaced scepticism. Many people are willing to shape actions and medical decisions based on information they find more or less casually. They are mostly on their own evaluating risks and kinds of harms that could arise. Spontaneous and ephemeral instances of self-report data collection are likely to compound this kind of issue, by buying credibility, and time, to implausible theories and methodologies. In several examples, patients eventually self-harm (Paul Wicks, Heywood, and Vaughan, 2012).

Self-harm is not the only relevant ethical issue. Another important issue is the potential for contributor-led initiatives to interfere with other, better designed clinical research, such as clinical trials, when participants decide to tamper with the experimental protocol by carrying out their actions on the side from it. The result could be the creation of noise, the jeopardisation or slowing down of the best research. These issues can create a rather uncomfortable situation for the managers and developers of the technology platforms these initiatives are relying on. An issue for them will also be how liabilities and best courses of action can be worryingly unclear.

1.5 Making the data worth it

Given that contributors can invest a significant amount of resources labouring to collect data (Milne, 2018; Cooper, 2012), develop expectations, and can face risk of undesirable consequences of participating in the data collection, it is really important that data are valued and used. As it has been made obvious by great amounts of literature, much of the value that has been ascribed to new kinds of data-intensive data collection, including the collection of self-report data, is due to the assumed possibility of reusing the same data multiple times for different purposes. Given that data can acquire new value when they are linked or juxtaposed with other data and questions, their value could be renewed as many times as people having access to them believe it is worth to do so.

While the investment on the part of an individual contributor can be variable and small, the cumulative investment asked of society, or a relevant group within it, can be considerable; the more so the more contributors a project is able to attract. A self-report data collection project, therefore, will face demands of accountability as to the ways in which the data have been put to use; and whether the amount of effort it commanded was worth the good it generates.

If little is done with datasets that people might have a reasonable expectation more could have been done with, a tension can arise. Scholarship on data self-reporting has thus questioned (Prainsack, 2017; Tempini and Del Savio, 2019; Sharon, 2016) specific projects that harness the rhetoric of empowerment and participation but might not live up to expectations (see Box 3); and asked if those who directly contributing are then excluded from data governance decisions of the data, the incentive to maximise the use of the data might be weaker (Birhane, 2021; Prainsack, 2019b; Ernst Hafen, 2019; Tempini and Del Savio, 2019). Expert practitioners involved in health data self-reporting studies often stress the duty that a project has to make justice to the participant’s burden.

There are various reasons why a dataset could not be widely re-used. For instance, in the case of health data, their sensitive nature and the high risk of misinterpretation require not everybody should be given access. Also, the data that is actually collected might not be as good and reliable as initially expected by the researchers who design the exercise. Much depends on the everyday practical circumstances of a self-selected group of contributors, with different expectations, hopes and levels of literacy.

Health data can be particularly expensive to keep and govern. They require complex infrastructures to protect the data from unauthorised access. This means keeping up with continuously changing security standards and a quickly evolving risk and threat landscape; while at the same time, maintaining the knowledge that is required to ensure that the data remain valuable, of high quality, and that their peculiarities and qualities are well understood by those who directly re-use them (Demir and Murtagh, 2013; Tempini and Leonelli, 2018). In this respect, self-report data pose more challenges because they are often made available in a structural lack contextual of information and awareness of the specific situation in which they were generated by unknown, distributed volunteer contributors. Also, many data self-reporting projects, especially when funded through public research funds, can ran into issues of long-term maintenance and sustainability; it is important that plans are made early on as to how the data will be managed in the long run and how continuous funding and support for data governance functions should be secured.

At any rate, data self-reporting projects should attend to questions of contributor benefit; reflecting on how to feed back relevant and valuable information learned from the research done with the data, a task that becomes more difficult the wider and more diverse the contributor base is; actively rejecting forces pulling the relationship with contributors towards an extractive state, were contributors keep being asked to contribute data while a clear sense of benefit and utilisation is being lost. This attitude requires active vigilance on the part of project managers and a willingness to challenge abnormal power asymmetries, especially when the project is driven by private entities. Powerful tech giants have entrenched their economic might, to the detriment of the public good, on extractive relationships where ever more data is continuously collected in order to predict and generate behaviour.

2.1 Data trusts and data cooperatives

Managers and developers of data self-reporting projects should consider what measures they can take to ensure data governance is accountable, inclusive, competent. Projects where data are extracted from a contributor base to then be shifted and shared according to the judgement of a self-selected few are more likely to enter controversy or make poor choices from an ethical point of view. These are the governance approaches privileged by private companies and tech monopolists who intend to control the way in which commercial value can be created by contributed data because they concentrate decision power in the hands of allegiant few.

Arguments in favour of individualised, distributed control of data that resonate with recent visions of decentralised web and organisational governance are problematic because of the burden and risks they put each individual deciding for themselves under. They imagine technological frameworks could make it possible for the data to be locked and unlocked by the individual they refer to, thus giving, in principle, total control over the data back to them. The real world might differ. Few have the expertise and time necessary to scrutinise proposal to use their data that might be put before them. Most are likely to take decisions that are against their best interests or principles because of poor or rushed judgement.

In recent years there has been a great deal of innovation in data governance frameworks that deliver a form of collective control on the projects, uses and aims the data are put to. These are the data trusts (Delacroix and Lawrence, 2019) and data cooperative (Ernst Hafen, 2019; E. Hafen, Kossmann, and Brand, 2014) approaches. They have been applied with particular promise in domains such as health data, where data are sensitive, requiring tight scrutiny, and highly valuable and desired by researchers and business alike.

These approaches reject the assumption that each individual is capable and best positioned to decide what projects are worth giving access to their data. They instead elaborate on forms of delegation whereby a number of research participants are appointed as delegates or trustees with data governance responsibilities and are charged with day-to-day decision making and governance. Depending on the setup, members can exercise their own decision over whether to participate in one project or another more or less often. These approaches can be used to take over and collectively govern and mobilise existing datasets that were already generated for other purposes, for instance, receiving healthcare; but can also be used to start and coordinate the collection of new data from the outset, in a ‘bottom-up’ fashion where the project is managed by participants. They are intended to afford better inclusivity, accountability and competent data management. Different individuals should be able to seek different levels of involvement according to their interest.

Not all self-report collection projects need to be governed through a bottom-up approach such as data cooperatives or trusts and in many cases, data governance might be better managed through review boards composed of independent experts (see for instance, section 1.3) who might or might not be participating in the project themselves. The value of representative bottom-up data governance approaches is still in the potential for the project management to better articulate finalities, values and directions of a project in a way that better reflects the views and aims of its pool of participants, enacting the sort of relational ethics (Birhane, 2021) where experiential expertise is valued at the same time as professional expertise

2.2 Solidarity and harm mitigation

Data governance includes the assessment of risks and benefits of ways of working with data to the best of knowledge and assumptions held by those involved in decision making. Assessing risks and benefits of different ways of using data necessarily involves making many assumptions about future developments and events. The exercise is uncertain, and it is possible even for the most suitable experts to make mistakes. This is true even when data are governed through a collective governance framework such as a data trust or cooperative. Harms and benefits will be very likely to distribute unevenly, and collective governance will not in itself make things right or fair if mistaken assessments lead to harm and benefits that are unfairly distributed in ways that could be mitigated.

As Barbara Prainsack argues (Prainsack, 2019a, 2019b), only when collective governance frameworks are combined with a commitment to the principle of social solidarity they can truly fulfil their revolutionary potential of alternative to the monopolistic extractive models pioneered by tech giants. To achieve this, data self-reporting projects need to accept the possibility of harm and prepare for its mitigation. With her colleagues (Prainsack and Buyx, 2017; McMahon, Buyx, and Prainsack, 2020) she calls for the creation of harm mitigation bodies that will operate harm mitigation measures. These, they suggest, could include issuance of apologies and stipulation of amendments to avoid future harm recurrence and financial support for the gravest of cases.

Through a combination of collective governance (see 1.1) and harm mitigation features, self-reporting projects can aspire at the implementation of a data donation economy that maximises data’s potential to be used for the common good.

Data donation has been used as a buzzword for broadly mimetic purposes by some exploitative projects, but it is not a fuzzy concept. There are a few conditions that must be met for a project to be based on data donation. Barbara Prainsack emphasises that for practices of data reporting and sharing to qualify as based on data donation, they need to exhibit relationality, indirect reciprocity and multiplicity. Relationality requires that the two parties (giver and receiver of data) mutually acknowledge the act of donation; this means that the data receiver (self-report data project manager) honours the work that the data donors (research participants) have done by “systematically considering the needs and interests of data donors and their significant others” (Prainsack 2019a:14); this can be reflected in the ways in which a project is organised to register, reflect and enact the views of its contributors. Relational ethics and bottom-up data governance, for instance, can go some way in this direction. But for the data donation economy to ensure that reciprocity is indirect and thus shared across the membership, harm mitigation measures are seen as a necessary instrument. It is through this combination of arrangements that data multiplicity (Prainsack’s way of pointing out the ability for data to be reused multiple times) can serve the common good. Data that are collected, for instance, through mainstream, status quo, commercial data governance practices (for instance, data collected through a user-generated data platform and used by a Silicon Valley giant for undisclosed commercial purposes) might still have multiple uses, i.e., multiplicity, but the economy of such a project would not be based on donation, because the data are not received as a gift and respected with the ethics gifts command.

2.3 RECs and research ethics governance

A crucially important function that officers of a data trust, data cooperative or data governance organisation form should need to provide is the oversight of ethical risks. At any rate, a data self-reporting project should implement an ethical risk oversight function that is adequate to its size and ambition. A project that aims to create a data resource that can be used time and again and is widely trusted by internal and external stakeholders is highly likely to need a formal oversight structure with clear governance procedures.

The time-tested standard model in this domain are the Research Ethics Committees that have been implemented for decades by universities in the UK, US (aka institutional review boards (IRB)), and the rest of the world. They normally require researchers who intend to embark on new research to submit a research ethics application. This is usually centred on a formal document (e.g., application form) that provides key information about the project aims, methods, resources; and evaluations of ethical risks together with a specification of measures planned to control for them. The application is reviewed by experts serving in the committee who should be able to judge whether the project is responsibly conceived, and the eventual ethical risks are acceptably approached, i.e., control measures are fit for purpose. If the application is rejected, the research cannot start.

Institutions can design, and have designed, research ethics oversight committees and processes differently to adapt to local conditions, but the above tend to be the main features, and this model has been adapted outside academia in the public and private sector as more organisations sought to control risks of their research and development (R&D) activities (including reputational risk, through the accountability that having a formal process affords). This means that there have been no off-the-shelf approaches to research ethics oversight, though there have been various guidelines.

The need for up-to-date guidelines and models has recently become apparent as some late developments in research methodology are challenging the research ethics status quo. A report by the Ada Lovelace Institute, the University of Exeter Institute for Data Science and Artificial Intelligence, and the Alan Turing Institute (Petermann et al. 2022) highlights the new challenges and the possible solutions that institutions should consider keeping their research ethics oversight function suitable (also, among others, explored by Ferretti et al., 2021; Jordan SR, 2019; Basl and Sandler, 2019; Clark et al., 2019). While for a long time RECs have been focusing on the ethical harms and risks that research could pose for the participants in a study (resulting from the origins of research ethics in medical research and the design of clinical experiments), new kinds of research employing cutting edge data science and artificial intelligence methods are bringing up emerging ethical issues. Here, the new innovations engendered by data self-reporting projects coordinated over the web are part parcel of these changes that need to be confronted.

Of growing concern have been the long term consequences of research that might be developing methods and technology that can be easily re-deployed (recall ‘function creep’) across institutions, jurisdictions, countries, populations, and how they might have disparately unequal social impacts; along with a redefinition of issues of privacy, confidentiality, and consent at a time when researchers can source individual level data from the web or other easily available resources, without asking the individuals these data originated from.

Issues are complex and their governance cannot be a box-ticking exercise. Methods that can be seen as gold standard in tacking some problems still require close and situated scrutiny. For instance, anonymisation can be a strong solution for granting privacy protection to individuals whose data widely available on the web is collected without their explicit informed consent. However, it is widely understood that the strength of anonymisation is always relative to the way in which the various other data sources a researcher can have access to, can lead to re-identification. Even when anonymisation is considered satisfactory for the purpose of privacy protection, in itself it offers nothing to protect individuals from the eventual unfair impacts of the outcomes of research once they are operationalised and deployed in new systems shaping individual movements, interactions, behaviours and access to services.

In recent years and in response to the emerging ethical challenges posed by the latest innovations in data science and AI, there has been growing demand for the active consideration of broader downstream impacts of such research. Research communities and academic societies have started to require researchers to submit research ethics statements on the broad societal impacts of their endeavours (Petermann et al., 2022). Leading commercial technology organisations have started to experiment with the implementation of internal ethics review processes for their commercial R&D projects, in an effort to improve their reputation and accountability when the public is questioning whether they can be trusted with running the socio-technical infrastructure of much public life. How much can commercial organisations be expected to follow together ethical and business principles, and successfully map, and manage, the broad landscape of social groups, needs and interests, is very much an open question. Criticisms of ‘stakeholder capitalism’ abound. But it can perhaps be argued that when self-report data power so many projects and systems across both the research and commercial worlds, and underpin affairs in the public sphere; and when so much of public life relies on complex infrastructures developed on methods and techniques of recent invention and never fully understood consequence; then, attempts at translating some of the ethical principles, expectations and methods developed to deal with the arising ethical issues of self-report data might be in order. The implementation of ethical oversight functions, which can take many forms, in new contexts is one such opportunity.

Questions around the re-definition of ethical standards and red lines are tightly linked to practical questions over the right training and composition of experts that REC should seek when evaluating projects with strong interdisciplinary components, whose risks are inherently more difficult to assess (Petermann et al., 2022); and complex projects involving multiple staggered undertakings where a technological solution builds from the previous. Questions that are open to debate and that will require careful consideration when implementing research ethics functions in organisations include also the scope, duration and frequency of interactions between research ethics functions and the researchers leading a project. These questions are not merely operational but really shape the quality of understanding and assessment that a research ethics function is able to deliver.

But at any rate, a form of research ethics oversight is needed in data self-reporting projects.

Here again principles of bottom-up participation will be important. The world-leading public health databank SAIL (Ford et al., 2009), which has been managing dozes of highly sensitive, linked health datasets and making them available to hundreds of research projects, involves members of the public, together with domain experts, as reviewing members of research ethics applications.

Bottom-up participation will improve diversity of perspectives and sensitivity to assess bias and unequal impacts, and will be a required condition for a full implementation of collective governance approach involving data management through trust and cooperative; and design and operation of harm mitigation measures. Birhane (Birhane, 2021) emphasises, with the concept of relational ethics, how those communities and groups that are on the receiving end of a project’s impacts should be included in its governance, because they have a key epistemic privilege about its social outcomes – decisions should be taken with and through them, instead of without.