On the empirical front, it has been argued that VSD cannot properly account for the diversity of values and how they play out in different ways across cultures and contexts (Davis & Nathan, 2015, p.21). As Albrechtslund (2007) notes, even though VSD draws on ethical theory and claims to take universal values into account, it does not make clear what and whose theories and values it includes. As such, VSD runs the risk of perpetuating the “belief that a particular group, culture, or religion is the keeper of those values, and needs to impose them on others – with sometimes tragic consequences” (Borning & Muller, 2012, p. 1127). Furthermore, VSD frameworks offer little light on empirical matters related to the emergence or intermediation of values (Johri & Nair, 2011, pp. 297-8). In this section, we continue the critical discussion by questioning value-sensitive approaches from a non-WEIRD cultural perspective based on current empirical research.
3.1 Primacy of WEIRD values
When it comes to the values typically included in VSD, Friedman and Hendry (2019) list a collection of twelve “human values with ethical import often implicated in system design,” which includes human welfare, ownership and property, privacy, freedom from bias, universal usability, trust, autonomy, informed consent, accountability, courtesy, identity, calmness, and environmental sustainability.
While VSD frameworks might seem well positioned to account for societal preferences, they may do so in a misguided manner. Davis and Nathan (2015, p.12) argue that VSD tacitly suggests that “designers must attend to values supported by theories of the right, which are obligatory, and may attend to values supported by theories of the good, which are discretionary.” Nevertheless, these theories (i.e. deontology, consequentialism) have developed in the cultural space of Europe and Ancient Greece. As Luegenbiehl (2009) points out, “foundational ethical principles as the major source for ethical decision-making are a specific product of the Western philosophical tradition.”
Insofar as technologies are used by different people in different parts of the world, such theories may not be relevant for designers in specific regions looking to take into account localized user groups. Prominent studies by Hofstede (1991; 2003) and Leung et al. (1999) have shown that values vary by culture and country. More recently, the large-scale study The Moral Machine launched by MIT asked people across the globe to engage in a simulation game that requires them to make decisions based on the moral dilemmas potentially faced by autonomous vehicles of who should live and who should die (Awad et al., 2018). It gathered 40 million decisions in ten languages from people in 233 countries and territories. Thus, the study provides insights for approaches one might take in designing self-driving cars, the most important of which is that Western values are unrepresentative of global populations.
The study shows that respondents from different regions and cultures prioritized different values when it came to the age, gender, social status, fitness level of the characters saved. These differences mirror patterns reflecting the cultural and economic variations between countries. The study found that “the more culturally similar a country is to the US, the more similarly its people play the Moral Machine” (Awad et al., 2018, p. 8). The authors consider that “an important obstacle” is the split between individualistic and collectivistic cultures when it comes to the “distinctive value of each individual” (Awad et al., 2018, p. 8). The individualism-collectivism axis is the exact same dimension highlighted by Hofstede’s (1991; 2003) empirically based model of cultural values which showed that people from different countries tend to hold contrasting values.
Saab (2008, p.8) also draws attention to the peril of treating the values promoted by VSD as universal. For example, he points out that the values of privacy and autonomy are part of individualist cultures, while group cohesion and harmony belong to collectivist cultures. As such, Saab (2008) notes that a US designer of an information system might favour the autonomy of the individual and design an interface with “concomitant responsibility for revisions of information content, the ability to personalize the technological interface, and the ability to keep an individual’s information private while allowing open access to other information and knowledge.” Such design would not meet the values of Guatemalan or Singaporean users, for whom “the personalization of the interface is likely to be less important than a single interface that fosters harmony of process among users” and who prefer that the technology “prioritizes the accomplishment of group goals” and “tracking revisions of information content at a group rather than individual level”.
Friedman and Hendry (2019) do not claim that the list of values is exhaustive and leave open at the declarative level the possibility for additions. Yet, Le Dantec et al. (2009) argue that the articulation of VSD methodologies does not support the active refinement of the value classification. This is particularly concerning given the industry, government and academia’s increased interest in AI ethics, which focuses on transparency, explainability, and trustworthiness (Balasubramaniam et al., 2022; Deloitte, 2019; HLEG AI, 2019; European Commission, 2020). These principles rest on narrow conceptions of ethics characteristic of the Western world that emphasise autonomy and individual protections.
Considering DfV, Van den Hoven et al. (2015) include the values of safety, sustainability, responsibility, democracy, or justice. Among these, justice is a culturally problematic value. Bombaerts et al (2020, p.11) draw attention that justice is a WEIRD concept that has been “pushed” into global philosophy. According to Tan (2015), the word justice does not exist in Chinese, thus questioning its universality as a philosophical concept. Furthermore, it is acknowledged that in Confucianism, justice is conjoined with two other concepts related to humanness and propriety, with the latter representing the external realization of justice via rules and norms (Lebow & Zhang, 2022).
These examples show that value-sensitive approaches have been biased towards WEIRD cultures in two ways: first, values representing western philosophical traditions are overrepresented; second, non-western values are adopted via an improper western translation and theorised as universal.
3.2 Enacting values across cultural contexts
We continue our analysis by focusing on the acontextual fallacy of value-sensitive approaches in the discourse about values. Values such as privacy or sustainability have become so commonplace in day-to-day language that their origins, specificities or legitimacy are left unquestioned, leading to acontextual interpretations. Stone et al. (2020, p. 1384) call this the “empty signifier problem”, referring to the unquestioned connotations with which values are being used on a routine basis to accept or reject technological designs.
Furthermore, there is a contextual side to human values, since they are situated and experiential: what a value means and how it is enacted depends partly on a local context and situation (Wong & Nguyen, 2021). Hence, values whose interpretation differs in WEIRD and non-WEIRD cultures require contextual considerations for translating values into behaviours via design. For example, in the interpretation of safety, the anthropological component of the safety culture is considered to reflect the broader national cultural (Yorio et al., 2019). It pertains to shared organizational beliefs, assumptions and values related to safety (Yorio et al., 2019). Cultural tendencies that play out in a national setting are seen to shape the beliefs and work behaviours of employees, as well as their safety-related values (Noort et al., 2016). In a highly global profession, such as the maritime industry, empirical studies found that nationality impacts one’s attitude towards safety (Håvold, 2007; Hansen et al., 2008). More specifically, Hansen et al. (2008) found that seafarers from Southeast Asia, mainly the Philippines, have a lower risk of occupational accidents than those from Western Europe, which is due to different safety attitudes. Moreover, the reduction in human failures in shipping operations is correlated with a high Confucian dynamism, manifest in following rules and procedures or saving face when it comes to safety (Lu et al., 2012). Merritt (2000) found that Hofstede’s dimensions replicate also in aviation. Even though aviation is a tightly regulated industry at the global level, we encounter a meaningful influence of national cultures on attitudes and behaviours (Merritt, 2000).
This variation is also present when considering autonomy and informed consent, which are among the VSD values listed by Friedman and Hendry (2019). In medical ethics, discussions on the universality of values are more advanced and have explored how values are interpreted differently in WEIRD and non-WEIRD cultural contexts. Xu (2004) gives the example of how in the Chinese culture patient autonomy is extended to the entire family and ethical practice requires seeking informed consent from the entire family rather than from the individual alone. Following Hanssen (2004), who argues against the universality of autonomy across cultures, Xu (2004) concludes that WEIRD values need to be translated to other cultures to preserve the best interests of users across the globe.
3.3 Axiology concerns: Values as unreliable predictors and descriptors of behaviours
The third point of critique is an axiological concern of whether values function the way value-sensitive approaches assume they do. A major assumption for pursuing VSD is that values are output-oriented, and as such, reliable predictors of behaviours. We further unpack this assumption, noting that it is based on a WEIRD conceptualization of value.
The supposed nature and assessment of values have arisen in WEIRD cultural contexts (Kulich & Zhang, 2012). Values have been conceived as long-standing beliefs or ideas about which states are worth pursuing (Torelli & Kaikati, 2009), rooted in consequentialist or deontological criteria (Luegenbiehl, 2009). As such, in WEIRD cultures, values are oriented towards the “output” side of human actions, translated into considerations of how one ought to behave (Garfield, 2022, p. 199). This comprises normative specifications of either the actions to be pursued or of the rationale for those actions. In contrast, Buddhism places one’s experience of the world as the principal object of moral evaluation, grounded in a social and psychological account of human nature (Garfield, 2022, p.22; p.169). Ubuntu also acknowledges its own distinct understanding of morality, which is relational rather than rational (Birhane, 2021). Birhane (2021) further claims that WEIRD cultures are inherently rationalistic, having at the core the Bayesian model of prediction to establish normative explanations of behaviours. This is reflected, for example, in the design of advanced distribution management systems, whose WEIRD centric rationalistic value system is prone to the amplification of socially held stereotypes (Mhlambi, 2020).
This is true also of value-sensitive approaches to design, which rest on a WEIRD conception of values as guiding and, therefore, able to predict behaviours. An example relevant for VSD is Verbeek’s (2006) use of the "script" concept to indicate how technologies prescribe human actions via specific values integrated in technological designs. Nevertheless, social psychology research shows that values do not necessarily function as ascribed by VSD and DfV scholars. Specifically, values are unreliable predictors of behaviours when individuals are induced to think concretely about situations (Torelli & Kaikati, 2009) or in the immediate future (Eyal et al, 2009). Eyal et al. (2009) showed that due to their high-level abstract nature, values are aspirational constructs which are more likely to be activated when considering future situations but are less likely to inform immediate actions. As such, a medical check-up, when envisioned in a distant future, is more likely to be represented as a desirable opportunity to improve health, compared to an upcoming check-up which is more likely to be represented in terms of discomfort or duration, which goes to show that the value individuals attach to health is less likely to guide one's behaviour of signing up for a medical check-up in the immediate future (Eyal et al, 2009). High level abstraction also comes into play when individuals deem particular objects or events as attractive or aversive based on the values held but may fail to focus on these values for interpreting their own actions (Torelli & Kaikati, 2009). As Darley & Batson (1973)’s study shows, theological seminary students running late to give a talk failed to help an ailing person on their way, even when the subject of their lecture was the parable of the Good Samaritan, which is centred on the value of helpfulness. The power of values to predict behaviours also varies by culture (Knafo et al., 2009). The values of non-WEIRD (Western Educated Industrialized Rich and Democratic) groups say less about how people will behave (Knafo et al., 2009; Henrich et al., 2010). For example, prosocial behaviours such as helping strangers are more likely to be encountered in less developed economies, as well as in Spanish and Latin American cultures (Knafo et al, 2009).
Additionally, although values are often invoked to understand cultural differences, they are nevertheless poorly suited for this role. Values are only one of many ways to understand culture, and not an especially accurate descriptor. For example, the values of mainland Chinese, Hong Kongese, and Singaporean Chinese, which are typically conceived as all belonging to the same cultural group, are closer to those of people in Zimbabwe, Israel, and Malaysia, respectively, than they are to each other (Smith, 2010). Research based on the World Values Survey (Haerpfer et al., 2020) shows that “socioeconomic development tends to propel societies in a common direction, regardless of their cultural heritage,” with all of the "high-income" societies (as defined by the World Bank) ranking relatively high on the dimensions purporting to secular-rational values and self-expression values, while all of the "low-income" societies fall into the lower left zone of the cultural axis, pertaining to traditional values and survival values.
Thus, VSD is seen to run the risk of committing the naturalistic fallacy of treating selected values in a normative sense. Manders-Huits (2011) points out that while VSD claims that values depend on the stakeholders’ stance, this implies a sociological conception of values rather than an ethical one. Simply because design frameworks address stakeholder preferences or values, does not mean they would be normatively sufficient. In other words, accounting for values is a necessary but not sufficient condition of ethical design. Taebi (2017) puts this problem down to the difference between “ethical” (or normative) and “social” (or descriptive) acceptability. This critique highlights the need to complement the focus on values in design with additional considerations, in order to bridge the gap between, on one hand, individual and localised preferences, often of a WEIRD influence, and on the other hand, behaviours that are manifested by groups and that can be representative across cultures. In the following section, we will explore how a focus on norms can serve the role of guiding design practices towards accounting for WEIRD and non-WEIRD behaviours or practices.