Value sensitive design and the ethics turn

Every design embeds values whether or not anyone names them. Value sensitive design (VSD) is the attempt to name them on purpose, at every stage, rather than to inspect them once the artefact ships.

This unit, the seventh of the ten, argues that the approach's real contribution is procedural. Earlier units settled who decides and what can be shown; this one asks which values the decision embeds. It turns a value trade-off into a decision that has to be written down, argued over and owned. It does not settle which value should win. Its unresolved problems are the indirect stakeholders who cannot be consulted, and the gap between the values a project elicits and the values its artefact embodies.

Naming a value is not resolving it

Friedman and Hendry (2019) give the fullest statement of the approach. It sets out seventeen methods including stakeholder analysis, value scenarios and multilifespan timelines.

The architecture comes from Friedman, Kahn, Borning and Huldtgren (2013). They set out an iterative tripartite methodology of conceptual, empirical and technical investigations. They illustrate it with browser cookies and informed consent, an office video window balancing employees' well-being against the privacy of passers-by, and the UrbanSim planning simulator. Technologies provide value suitabilities: architecture makes some values easier to support than others. Friedman, Hendry and Borning (2017) inventory fourteen methods sorted by purpose.

The record: a strong conceptual arm and a thin technical one

Winkler and Spiekermann (2018) audited what the method has produced. They searched four libraries for work from 1996 to 2016. They identified 229 contributions, 219 of them accessible. Of those, 106 used no part of the tripartite method, 63 conceptualised values without designing anything, 22 contributed a technique and 11 proposed extensions. Only 17 projects developed technology with the complete tripartite methodology. Sixteen began conceptually, nine ran an empirical investigation, and only four reported iteration between them. Technical investigations, where a value would be designed in, relied mostly on the authors' background knowledge.

Gerdes and Frandsen (2023) extend the audit to 2023. From 2,232 records and 547 further records they included 54 studies, 22 theoretical and 32 applied. They record that the absence of technical analyses, reflections and solutions is a problem for the framework.

Elicitation is not justification

Manders-Huits (2011) stated the central objection early, and it has not been answered. She lists five failures. There is no clear methodology for identifying stakeholders. Empirical and conceptual research are integrated obscurely. There is a risk of the naturalistic fallacy. The concept of values and their realisation is undetermined. There is no ethical theory for value trade-offs. Normative evaluation of a technology, she concludes, requires an explicit and justified ethical starting point.

Jacobs and Huldtgren (2018) sharpen the charge. VSD commits to no ethical theory, so they argue that it cannot distinguish genuine moral values from stakeholder preferences. This is the sharpest attack on the first half of the thesis, because it separates the two. VSD can make a trade-off explicit and owned. It cannot say whether the value that won was the right one, or whether it was a moral value rather than the loudest preference in the room.

The hardest category is the indirect stakeholder, who cannot be consulted

Friedman and Hendry (2019) list the indirect stakeholder among the method's open questions. The office video-window case in Friedman, Kahn, Borning and Huldtgren (2013) is the worked example. People who pass an office screen are affected without ever using it, including the older passer-by who reads it slowly. The remedy proposed is architectural: keep the technology flexible enough to accommodate unanticipated values later.

Shilton (2018) shows how badly this scales to infrastructure. She observed a team developing internet protocols, and she found two barriers to values reflection. It is difficult to say who the stakeholders of a protocol are, since they include users on a slow or metered connection whom the architects never meet. Local and global viewpoints pull against each other. The architects also treated interoperability, a core value of their work, as equivalent to values neutrality. Her earlier ethnography (Shilton, 2013) reaches the same conclusion from the other side. She studied a laboratory building location-tracking software for mobile phones. She identifies values levers, practices that open conversations about social values and produce agreement on them as design criteria. Reflection, on this evidence, is an artefact of working conditions, not a step in a method.

Le Dantec, Poole and Wyche (2009) explain why a form cannot fix it. They identify three limitations: in value classifications, in guidance on empirical tools, and in the order imposed on the design process. They respond with three case studies supporting value discovery. The values are found in lived experience, in local terms, before anyone designs for them.

Values elicited from people need not survive into what is built

Sadek, Calvo and Mougenot (2024) turn the question into criteria. They review socio-technical design processes for AI, and they assess each against comprehensiveness, guidance offered, and methodological value-sensitivity. A process is value-sensitive only if it can do something with a value after eliciting it.

Davis and Nathan (2015) were trained in the framework's first doctoral cohort, and they arrive at a similar place from inside the tradition. They argue for a condensed VSD with heuristics for those who may not follow its tenets. They question the grounds on which usability is excluded from moral values while calmness and courtesy are admitted. They report that no published VSD project before 2006 was free of close links to the founders. Zhu and colleagues (2018) built a recruitment algorithm for Wikipedia's WikiProjects with contributors' values designed in, then deployed and evaluated it. Invitations raised experienced editors' activity inside a project without reducing their edits elsewhere. What was evaluated was editors' activity and their acceptance of the tool, not whether the values the design was built around survived in the community.

Contested question: is the missing ethical theory a defect or a design choice?

The first position holds that thinness is the virtue. On this view, Friedman, Kahn, Borning and Huldtgren (2013) offer a design framework rather than a moral theory. Refusing to rule on values is what lets it travel across domains and leave the decision with the people affected. Davis and Nathan (2015) support the position by condensing the framework and offering heuristics to those who do not adopt its tenets. Those heuristics would be unnecessary if it were a doctrine.

The second position holds that the omission is a defect. Manders-Huits (2011) and Jacobs and Huldtgren (2018) agree that without an ethical commitment the method cannot separate a moral value from a preference. That makes its output arbitrary where the stakes are highest.

The third position denies the terms of the disagreement. Martin and colleagues (2023) argue that mainstream value-sensitive approaches rest on narrow understandings of personhood and social dynamics. Those understandings are biased towards Western, educated, industrialised, rich and democratic cultures, and evidence contradicts them. The authors also argue that values are poorer predictors of behaviour than norms, so eliciting them cannot produce the behaviour the framework promises. Their remedy, norm sensitive design, moves the unit of analysis from values to norms. This is the strongest counter-argument available to the thesis. It does not dispute that a trade-off can be made explicit and owned. It denies that the material being made explicit can predict what a system will do to people.

What would settle it is a study no source here has run. It would run one brief twice, with and without an explicit ethical commitment. Independent coders would measure whether the trade-offs were documented and owned. They would also measure whether the values embodied in the artefact match those elicited. Winkler and Spiekermann (2018) give reason to expect a null result on the second measure, since only four of seventeen projects reported any iteration. Iteration is the mechanism that would close the gap.

The three positions rest on different evidence. The thinness-is-a-virtue position rests on the framework's own statement and on a condensation by two of its practitioners, both normative. The defect position rests on two critical essays, which argue what the method cannot do rather than measuring what it fails to do. The denial position rests on a conceptual critique and a call for cross-cultural evidence not yet gathered. The audits are the only counting here, and they are agnostic on the ethics question, since they record what VSD projects did, not what a better-justified project would do. Until the two-condition trial exists, the designer should write the ethical starting point down before eliciting, keep elicitation and justification in separate sections, and report whether the elicited value survived into the artefact. The strongest objection to the unit's thesis is that a written and owned trade-off can launder a stakeholder preference into a moral value, so the record confers legitimacy without supplying justification.

Where the argument is heading

Named work points in several directions. Sadek, Calvo and Mougenot (2024) convert value-sensitivity into three evaluable criteria. Gerdes and Frandsen (2023) reconstruct the technical arm across three decades and hand the framework a development agenda. Martin and colleagues (2023) ask for cross-cultural evidence about norms rather than values. Friedman, Hendry and Borning (2017) remain the inventory that later work measures itself against. Shilton (2018) adds the possibility that the binding constraint is occupational rather than conceptual, since infrastructure developers equated neutrality with interoperability.

The sources this unit rests on

• Friedman, B. and Hendry, D. G. (2019). Value Sensitive Design: Shaping Technology with Moral Imagination. MIT Press. https://mitpress.mit.edu/9780262039536/value-sensitive-design/ Friedman and Hendry give the fullest statement of value sensitive design, setting out seventeen methods including stakeholder analysis and value scenarios, with practice reports from ten domains. It is the approach's own account. Not an evaluation.
• Friedman, B., Kahn, P. H., Borning, A. and Huldtgren, A. (2013). Value sensitive design and information systems. In Early Engagement and New Technologies: Opening Up the Laboratory. Springer. doi:10.1007/978-94-007-7844-3_4 Friedman, Kahn, Borning and Huldtgren set out the iterative tripartite methodology of conceptual, empirical and technical investigations, illustrated with browser cookies and UrbanSim. It gives the unit its indirect-stakeholder example.
• Shilton, K. (2013). Values levers: building ethics into design. Science, Technology and Human Values. doi:10.1177/0162243912436985 Shilton studied a location-tracking system for mobile phones and identifies values levers, practices that open conversations about social values and produce agreement on them as design criteria. It is an ethnography of one site.
• Shilton, K. (2018). Engaging values despite neutrality: challenges and approaches to values reflection during the design of internet infrastructure. Science, Technology and Human Values. doi:10.1177/0162243917714869 Shilton observed a team developing internet protocols and found two barriers to values reflection: it is hard to say who a protocol's stakeholders are, and local and global viewpoints pull against each other. Participant observation of one team.
• Manders-Huits, N. (2011). What values in design? The challenge of incorporating moral values into design. Science and Engineering Ethics. doi:10.1007/s11948-010-9198-2 Manders-Huits argues value sensitive design has no methodology for deciding which values are morally relevant and conflates stakeholder preferences with moral values. It is a philosophical critique stating the unit's central objection. Conceptual critique.
• Jacobs, N. and Huldtgren, A. (2018). Why value sensitive design needs ethical commitments. Ethics and Information Technology. doi:10.1007/s10676-018-9467-3 Jacobs and Huldtgren argue that without an ethical theory value sensitive design cannot distinguish moral values from stakeholder preferences, and propose a mid-level theory. It is a critical essay on the method.
• Winkler, T. and Spiekermann, S. (2018). Twenty years of value sensitive design: a review of methodological practices in VSD projects. Ethics and Information Technology. doi:10.1007/s10676-018-9476-2 Winkler and Spiekermann audited twenty years of value sensitive design projects. Of 229 contributions, only 17 developed technology with the complete tripartite methodology, and only four reported iteration between the investigations.
• Davis, J. and Nathan, L. P. (2015). Value sensitive design: applications, adaptations, and critiques. In Handbook of Ethics, Values, and Technological Design. Springer. doi:10.1007/978-94-007-6970-0_3 Davis and Nathan survey value sensitive design's applications and critiques and report that no published project before 2006 was free of close links to the founders. It is a handbook review by practitioners. A review, not an evaluation.
• Le Dantec, C. A., Poole, E. S. and Wyche, S. P. (2009). Values as lived experience: evolving value sensitive design in support of value discovery. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. doi:10.1145/1518701.1518875 Le Dantec, Poole and Wyche identify three limitations of value sensitive design, in value classifications, in guidance on empirical tools and in the order imposed on the design process, then answer with three case studies. Only three cases.
• Friedman, B., Hendry, D. G. and Borning, A. (2017). A survey of value sensitive design methods. Foundations and Trends in Human-Computer Interaction. doi:10.1561/1100000015 Friedman, Hendry and Borning survey fourteen value sensitive design methods and describe what each is for and how it is used, grouped by purpose. It is the field's descriptive methodological inventory. It leaves evaluation open.
• Zhu, H., Yu, B., Halfaker, A. and Terveen, L. (2018). Value-sensitive algorithm design: method, case study, and lessons. Proceedings of the ACM on Human-Computer Interaction. doi:10.1145/3274463 Zhu and colleagues build a recruitment algorithm for Wikipedia's WikiProjects with contributors' values designed in, then deploy and evaluate it, measuring editors' activity and acceptance rather than whether the values survived. Field study.
• Sadek, M., Calvo, R. A. and Mougenot, C. (2024). Designing value-sensitive AI: a critical review and recommendations for socio-technical design processes. AI and Ethics. doi:10.1007/s43681-023-00373-7 Sadek, Calvo and Mougenot critically review how socio-technical design processes for AI create value-sensitive AI and find few provide guidance, elicitation, or means of embedding and tracking values. They set three criteria.
• Gerdes, A. and Frandsen, T. F. (2023). A systematic review of almost three decades of value sensitive design (VSD): what happened to the technical investigations? Ethics and Information Technology. doi:10.1007/s10676-023-09700-2 Gerdes and Frandsen systematically review almost three decades of value sensitive design studies, including 54 studies, and find the absence of technical analyses a problem for the framework. It measures no effect.
• Martin, D. A., Clancy, R. F., Zhu, Q. and Bombaerts, G. (2023). Why do we need norm sensitive design? A WEIRD critique of value sensitive approaches to design. Global Philosophy. doi:10.1007/s10516-023-09689-9 Martin, Clancy, Zhu and Bombaerts argue mainstream value-sensitive approaches rest on narrow, WEIRD assumptions about personhood and social dynamics, and that norms predict behaviour better than values. They propose norm sensitive design. Conceptual critique.