Introduction
Trust is a concept that put generations of social scientists to work trying to define it, although in everyday life we usually do not have to think twice before using the term. Similarly, we often talk about trust in technology, based on experiences of trusting some technologies more than others, or the trust in particular technologies developing over time. Yet, the very existence of trust in technology is brought into question by scholars. This opinion piece aims to provide an overview of trust as a concept and how it applies to technology. In particular, it analyzes how the development of AI, and specifically artificial general intelligence (AGI), changes the societal, scientific and political debate around trust in technology.
The question of what trust really is may well be answered by one of the most famous phrases in the history of the US Supreme court that is often used in lack of a proper definition – “I know it when I see it”. However intuitive the definition of a universal human experience such as trust may seem, to answer the question of how it applies to technology, trust must first be properly defined. The following section attempts to achieve this by considering several elements present in most of the (exceedingly numerous) definitions of trust.
What is trust? (Baby don’t hurt me)
Trust is most commonly defined as the willingness of a party to be vulnerable to the actions of another party, based on certain expectations of the other party’s actions, and irrespective of the ability to monitor or control that other party (Mayer et al., 1995). The key defining feature of trust is vulnerability, meaning that trust can only exist if there is a risk from harm. In other words, trust is conscious exposure of an individual to the risk of (physical or psychological) harm from another.
The many factors that affect trust fall into three categories, relating to the trustor (the one who trusts), the trustee (the one who is given trust), and the context within which the trustor and the trustee exist. Characteristics of the trustee pertain to the trustee’s trustworthiness, including ability, benevolence and integrity; characteristics of the trustor pertain to everything that influences trustor’s trust-giving, such as their dispositions and perceptions; and contextual factors, which affect both the trustor and the trustee, relate to the broader environmental, social and institutional context.
The core element of the trust definition – namely, vulnerability – implies the necessity of trust involving (at least) two actors. It follows, thus, that trust is relational. This relational quality of trust often generates reciprocity. However, reciprocity, although phenomenologically common in interpersonal trust relationships, is not necessarily a defining characteristic of trust. In addition to reciprocity being omitted from most definitions of trust, illustrative conversational expressions such as blind trust or breach of trust indicate that reciprocity is likely a common side effect of trust, rather than its defining characteristic. In some instances, trust can be relatively one-sided, or even occur without conscious awareness of the trustee.
Thus, while reciprocity is not necessary for the existence of trust, a relational element, i.e. relating to an entity outside of oneself, is. Yet, in most cases when thinking about trust, and especially when measuring it, there is an implicit notion of the first person perspective, i.e. a subject-object relation. Following this line of reasoning which focuses on the subject’s perceptions of an object, it is possible to trust various nonhuman entities the subject is potentially vulnerable to, such as animals, organizations, institutions, governments, and technologies. The degree of reciprocity in these relations of trust varies – while one can certainly build a reciprocal trust relationship with a nonhuman animal, this may not always be the case with an institution or a government (as will be explored in the following section), and is most certainly not the case with the technologies available to humans at present. However, it is not implausible that further technological development will enable reciprocal trust relationships with technology, particularly with AGI, the advent of which adds another layer of complexity to trust in technology.
From interpersonal to impersonal trust
Upon defining interpersonal trust, let us consider how this definition may apply to other relational objects. One such example is trust in different governmental institutions, also known as political trust. Just like interpersonal trust, this type of trust is defined in the literature as consisting of confidence in the trustee’s (in this case, a particular government’s or governmental institution’s) competence and benevolence. However, while reciprocity is commonly a feature of interpersonal trust, it is to a lesser extent present in political trust, possibly due to the indirect nature of the relationship that can often conceal trust signals that would otherwise be more apparent through direct communication typical for interpersonal contexts.
For example, signals of trust from the institutions to the citizens, as shown through policies and administrative procedures, are more transparent than signals of trust from the citizens to their governments, which are usually opaque unless citizens are asked about their political trust directly. Thus, while more trust exhibited by the governmental institutions is likely to lead to more trust experienced by citizens, it is arguably less likely that high political trust that citizens give to their governments would translate into high trust that governments grant their citizens.
Another potential reason for this is power: previous research has shown that when a relationship contains a power disparity, the trust of an entity with more power does not increase with increased trust of the entity with less power. For example, while supervisors’ expressions of trust will increase employees trust in the supervisor, that is not the case for employees’ expressions of trust – on the contrary, what will increase supervisors’ trust is the quality of employees’ performance, rather than, as would be expected according to the reciprocity principle, trust employees demonstrate towards supervisors. Put simply, it seems that in a trust relationship characterized by an unequal power dynamic, the side that holds more power also has greater ability to set the tone for reciprocity of trust within the relationship – i.e. the level of trust of the more powerful party is more likely to be reciprocated by the less powerful party, than the other way around.
In technology we trust?
Similarly to the extrapolation from interpersonal to political trust, there have been numerous attempts to apply trust to understanding of the relationship of humans with technology. Conversely, due to the impossibility of creating reciprocal relationships with technology, some deem the concept of trust inapplicable in this context. However, as seen in the previous sections defining interpersonal and political trust, while reciprocity commonly occurs in trust relationships, it is not a necessary condition for trust as exemplified in instances where trust exists despite a lack of reciprocity. In parallel, critics of the trust in technology concept argue that because technology cannot “betray” us, it is not appropriate to say that we form trust relationships with technology. Instead, when technology fails, it disappoints us (due to its (in)competence), rather than betraying us (as would be the case had the issue been its (un)benevolence). Therefore, critics argue, the relationships we have with technologies should more accurately be termed as relationships of reliance than trust (Deley and Duboy, 2020). In other words, while it is obvious that the competence component, i.e. the confidence in technologies’ ability and reliability, is relevant in how we as humans relate to technologies, it is less obvious whether the confidence in the trustee’s benevolence is applicable in this case. According to some, since technologies are inanimate objects lacking intentionality and agency, they cannot be malicious or benevolent, and therefore miss a key component that defines the trustee.
Be that as it may, within the first-person perspective, the benevolence component is subjectively experienced – dispositions such as personality traits, previous experiences, and risk perceptions, all inform the perceived benevolence of the trustee. Although this may constitute projection, it is highly likely that the trustor perceives technologies as having differing degrees of benevolence, even if technologies do not (yet) possess conscience and agency. This is especially the case nowadays, because the information about the development process and designers of technology is often publicly available and well known. Therefore, it is common for trustors to extrapolate their impressions about the benevolence of individuals and corporations that design and develop emerging technologies into benevolence perceptions of technologies themselves. Under the view that technologies are not value neutral, and the values of designers become encoded in technologies, such spillover of perceived benevolence from creators to artifacts is quite reasonable.
What about AI?
The applicability of trust to the relationship of humans with technologies becomes greater the more technologies resemble humans. In particular, AI applications are commonly designed to mimic human social interactions (e.g. through the use of language, social norms and scripts, and even physical features of embodied AI and humanoid robots). Such design choices are likely deliberately geared towards increasing trust, since research shows that anthropomorphic design of AI contributes to trust because similarity between parties increases trust. While humans for the most part remain conscious of the fact that computers are not people, we often subconsciously anthropomorphize them (e.g. the computer is “tired”, or it got “confused”). Such anthropomorphizing is particularly prominent with prolonged exposure to AI.
While all just outlined arguments in support for trust being a relevant theoretical lens through which we can observe human-AI relationships refer to already widely used technologies, another argument could be made in reference to upcoming technologies, namely AGI (Artificial General Intelligence). Specifically, since AGI by definition rivals humans in competence, its benevolence becomes particularly important. Moreover, the path that the evolution of this emerging technology is going to take in terms of autonomous consciousness and agency is unknown. Therefore, the perception of risk is highlighted, and many are therefore distrustful, as evident from wide-spread societal debates.
Lastly, a particular concern that is becoming apparent in the public discourse around AGI could justly be termed new-age luddism. While the term luddism is often used pejoratively to describe the irrational fear of technological progress, a historical perspective reveals a different picture. Namely, the Luddite movement began in the early 19th century in protest of cost-saving machinery that put mill and factory workers out of job opportunities and reduced wages. With the advent of AGI, defined as being in most tasks equal or superior to humans, the Luddite sentiment of concern over replaceability of workers with emerging technologies is rising, especially among precarious workers.
Conclusion
This opinion piece offers an overview of existing arguments for and against the appropriateness of trust as a lens through which the relationship between humans and technology can be viewed. Thus far, the scientific debate focused on the lack of reciprocity and benevolence of technology as the arguments against, versus the first-person perspective, perceptions of risk, and anthropomorphic design as arguments in favor of using trust as a conceptual lens for understanding the relationships between humans and technology. This opinion piece offers three additional arguments supporting the relevance of trust in technology, which simultaneously explain the observable mistrust in emerging technologies among some factions of the society. The first argument relates to how (mis)trust in the individuals and/or corporations associated with development of certain technologies “colors” the perceptions of, and determines the (mis)trust in technologies themselves. The remaining two additional arguments relate to the emergence of AGI: namely, the apparent AI evolution in the direction of increasing autonomy, which increases its similarity to human trustees; and the concerns around societal consequences of AI development, such as labor replaceability. In sum, trust appears to a great extent applicable to human relationships with technology, and is becoming increasingly applicable with emerging technological developments. The more anthropomorphized and integrated technology becomes in social life, the more appropriate it is to understand our relationship with it using social psychology constructs, such as trust. However, given the multidimensionality and subjectivity of trust as outlined in this piece, it is questionable whether the efforts to develop “trustworthy AI” can ever be achieved. Similarly, the efforts to regulate the trustworthiness of AI may fall short of their promises to achieve this lofty goal, by equating trust with acceptability of risks (Laux et al., 2023), which is, as discussed here, only one component of trust. This is potentially misleading because it conflates the risk acceptability as determined by domain experts with trust as a dynamic and highly contextual phenomenon elusive of objective definition that would provide grounds for a legally regulated seal of approval.
Antonia Stanojević
Antonia Stanojević is an interdisciplinary social and behavioral science researcher, whose research analyses power dynamics at different levels (interpersonal, organizational, intergroup and transnational) and emphasizes the interconnectedness of these levels, advancing as such a holistic approach to burning societal issues with a focus on positive social impact. Antonia is currently employed as a postdoctoral researcher at the Tilburg Institute for Law, Technology and Society, where she conducts qualitative and conceptual research into the governance of and governance by emerging technologies.
