Ends of responsible innovation

Beginning in late 2017, there has been increasing talk about ‘the end of RRI,’ that is, the presumed termination of the European Commission’s program for Responsible Research and Innovation. As a sign of different things to come, budget and strategy documents for the 2021–2017 Horizon Europe program released this summer refer to ‘research and innovation,’ but do not include ‘responsible’ in this formulation. Trends often define funding cycles, and the openness of the RRI concept is partly what explains its rapid political capture (Rip 2016). At the same time, the wave of national and regional policies in the early 2000’s that emphasized public engagement and socio-technical integration points to something greater than the individual programs for responsible innovation that it brought about. In other words, deep structural ambivalence to science and expertise mean that responsible innovation as a necessary aspiration is here to stay. In several respects, a shift toward more decentralized and diverse forms of responsible innovation policy may allow learning and insight, rather than budgetary categories, to guide incremental initiatives (such as described by Van Oudheusden, Turcanu, and Molyneux-Hodgson 2018) and help them continue to take root in national and local settings.

To that end, the Journal of Responsible Innovation will be particularly interested in taking stock of what can be learned from the European Commission’s RRI experiment, what was newly gained, what may need to be recovered, and what futures of responsible innovation should guide our conceptual, strategic, and methodological efforts moving forward. Stay tuned.

Meanwhile, the year 2018 has been an exciting one for JRI. The journal saw its most prolific output yet, augmenting its three regularly scheduled issues—which included, among other offerings, a special issue on brain science (Bowman et al. 2018), new reflective engagement approaches (Felt, Fochler, and Sigl 2018), and empirical demonstrations of the utility of such engagements (Flipse and van de Loo 2018)—with a special supplementary issue on gene drives (Delborne et al. 2018). This, the final issue of the year, contains various discussions over the ‘ends’ of responsible innovation, including its aims, boundaries, and conditions.

In the opening research article, Cohen, Stilgoe, and Cavoli (2018) report on the UK’s first systematic stakeholder engagement on self-driving cars. Taking the massive investments in and rapid development of the ‘emerging sociotechnical system’ (p. 258) around self-driving cars as an occasion to revisit scholarly understandings of deliberation, anticipation, and responsiveness in relation to ongoing experimentation, they begin with reflections on the limited ways in which disruption and ambivalence were handled in the history of automotive governance. Noting some of the well-known difficulties that public and stakeholder engagements face and pointing to the need to situate them within the broader sociotechnical system they are meant to inform, Cohen et al. encourage us to ‘consider how deliberative processes can reach upwards rather than going round in circles’ (p. 260). For instance, they observe several encouraging themes that emerged from their thoughtfully designed exercise and generate a set of critical questions for the powerful actors and institutions who are positioned to develop the rules that establish autonomous vehicle governance processes, and therefore possibilities.

Where Cohen et al. would link anticipation with experimentation, Mertens (2018) would moderate it with ‘ongoing monitoring and adjusting’ (p. 287). In the second research article, she argues that three fundamental assumptions responsible innovation frameworks make—about novelty, malleability, and timing—fail to describe what she observes in clinical settings, where the lines between research and care are blurred and where experimentation is tightly ‘restricted and controlled’ (p. 287). While these assumptions may be justified in cases that are ‘capable of changing the status quo’ (p. 283), they fail to apply to the same degree in more continuous and ‘liminal’ cases of technological innovation. Mertens seeks to move us beyond a focus on ‘the typical emerging technology’ (p. 282) with its over-determined boundaries and ‘away from anticipation of the unforeseeable unknown, back to observation’ (p. 292). In advising a return to the known and discernible, she foreshadows what Nordmann (2018), below, describes as ‘the mundane alternative’ (p. 332) to the conceit of mastery over time. She also provides a reflexive entry point for the third research article, since ‘RI researchers often change the practice they observe simply by being present’ (p. 293).

While Mertens questions key assumptions about the contexts in which responsible innovation is envisioned to operate, de Boer, Hoek, and Kudina (2018) probe the normative capacities of one of its best-known institutional roots, technology assessment (TA). Taking up the debate¹ about TA’s ‘normative deficit’ (p. 302) and its lack of attention to ‘qualitative ethical concerns’ (p. 302), their research article asks whether the approach of technological mediation (Verbeek 2011) can, as it has been called upon to do, ‘provide TA with a normative basis through which our relation with (emerging) technologies can be evaluated’ (p. 311). After explicating the distinct normative stance of the approach, which refuses to base its evaluations on external criteria, de Boer et al. argue that, if TA were to embrace ‘the inherent normativities arising in our relation with technologies’ (p. 254), that it would undergo a radical transformation since it would need to move away from ‘assessment’ and embrace a more reflexive ‘ethics of accompaniment’ (p. 310).

The discussion paper and three responses in this issue continue the theme of ends—in this case debating how to frame goals, address practical conditions, and utilize conceptual boundaries of responsible innovation. In their opening provocation, Ribeiro et al. (2018) seek to dislocate Collingridge’s (1980) dilemma of social control—which asserts that, while early periods of technoscience tend to afford considerable agency yet limited knowledge, the inverse tends to hold at later points—from its place as a central motivating and organizing concept for governance approaches to science, technology, and innovation. They propose a shift ‘from “social control” to “societal alignment”’ (p. 323) and note that societal alignment is problematic not only because of the Collingridge dilemma, but also due to ‘difficulties in democratizing science, technology, and innovation, addressing divergent stakeholder perspectives, and ensuring a closer correspondence between their benefits and the needs of diverse publics’ (p. 318).

While each of the three responses concur that alignment is an objective of responsible innovation, each also takes Ribeiro et al. to task regarding fundamental issues associated with how alignment ought to be conceptualized and pursured. In the first response, Nordmann (2018) appreciates the attempt of Ribeiro et al. to move beyond the Collingridge dilemma, given the diabolical interpretation of it as a problem for TA practitioners and social scientists ‘to manage or even solve’ (p. 336). Taking it this way induces the conceit that such management is possible, which distracts us from more tractable problems at hand and leads us into ‘the hubris of technical control’ (p. 336). While any attempt to refocus our gaze on the mundane tasks of tending to the known and given world is welcome, Nordmann finds that the proposed dilemma of societal alignment fails to offer a ‘genuine dilemma’ (p. 332). This is problematic, since Ribeiro et al. ‘do not begin from a principled critique’ of the Collingridge dilemma and instead appear to ‘share some of its problematic aspirations’ (p. 335).

Kuzma and Roberts (2018), in their response, make clear that responsible innovation is for them a ‘set of processes to achieve societal alignment’ (p. 255). Yet, rather than developing overarching frameworks for alignment, Kuzma and Roberts urge us to be ‘less optimistic’ (p. 338) and to pay attention to the ‘significant misalignment of public goals with research, technology and innovation systems’ (emphasis added) (p. 339) that show up in their current research and that are likely to persist. In cataloguing some of the most salient barriers, biases, and limitations that confront attempts aimed at societal alignment, Kuzma and Roberts encourage us to continue ‘documenting, understanding, and developing collaborative, realistic and practical approaches’ (p. 338) that address the concerns of otherwise unconvinced research and innovation actors.

Lastly, Guston (2018) calls foul on Ribeiro et al.’s use of the term ‘dilemma’ to frame the problem of societal alignment. Unlike the one identified by Collingridge—which is a ‘true dilemma’ (p. 347) since attempts to grasp either of its horns will come up empty handed—no such ‘logical conflict’ (p. 348) characterizes the challenge of societal alignment. The Collingridge dilemma is instructive, Guston argues, because its precise formulation reminds us ‘that social control is illusory’ and therefore ‘should not be pursued or expected to be pursued’ (p. 349). By contrast, working towards societal alignment is something that the responsible innovation community has justifiably devoted considerable ‘energy and commitment’ (p. 350) to addressing, as for instance work on anticipatory governance shows.

This issue of JRI concludes with reviews of two very different approaches to the problem of values in algorithms. Woodson (2018) details the ways in which O’Neil’s Weapons of Math Destruction (2016) convincingly demonstrates and with numerous examples how unexamined values and biases find their way into the design and use algorithms and data models, often to the considerable detriment of poor, minority, and other vulnerable populations. Although he would prefer O’Neil to attend more carefully to complex socio-technical interactions, and notes that the book does not wrestle with fundamental social questions, he nevertheless recommends Weapons as ‘a compelling introduction into the idea that science and mathematics are never value-free’ (p. 363).

Grinbaum (2018), who returns us to self-driving cars and to the instructive character of dilemmas, offers a radically different take on the problem. In explicating the central thesis of his Machina Delatrix (2017), published in Russian and soon to appear in French, he argues in favor of opacity rather than transparency as a value in machine learning on the basis that machines cannot be ‘moral by design’ (p. 353). Rather, the most viable route to ensuring trust in the algorithms that guide the physical pathways of artificially intelligent machines will be rooted not in the ethical acceptability of the pre-determined calculations they make when facing moral dilemmas, but instead in the strictly random practice of ‘choosing a path by chance’ (p. 355).

Notes

1 See, for instance, Grunwald (2014) and the recent exchange between van Van Lente, Swierstra, and Joly (2017) and Delvenne (2017), Nentwich (2017), and Van Est (2017).