Trust in gendered voice assistants—the special case of gender ambiguity

Abstract Automation in communication makes it possible to rethink communicators. With technologies such as voice assistants, for example, it is possible to design voices that can no longer be acoustically assigned to a social gender and are sometimes perceived as “ambiguous”. A phenomenon previously conceivable only for humans and occurring only among a minority of humans promises to become significant for the development of speaking artificial agents: Such voices are now available for the US-American Siri and for “Q”, which aims to overcome stereotypes. However, while research suggests that gender attribution impacts the evaluation of voice-based agents (VBAs) and similar technologies, including trust towards them, such findings are lacking for ambiguous voices. Meanwhile, trust is crucial for the acceptance of technologies. On the one hand, voices with a clear social gender are perceived as more trustworthy and competent when they talk about stereotypically appropriate topics. In turn, acoustically ambiguous voices can be assigned to a gender based on the topic they are talking about. Based on stereotype theory, this would imply an overall higher trustworthiness as the perceived gender would match the topic. On the other hand, two major issues could endanger this aim: First, the disfluency effect indicates a perception of gender-ambiguous voices as less competent and trustworthy as their categorization is harder. Second, although gender can be assigned to acoustically ambiguous voices, it is unclear if this assigned gender has the same impact on perception and evaluation as distinctly gendered voices. In general, there is a dearth of research on ambiguous voices and their effects. Therefore, this study addresses two combined questions concerning the impact of gender ambiguity in communication: How do gender-ambiguous voices affect the users’ trust, especially in comparison with distinctly gendered ones? And are there differences in trust between voices that are acoustically ambiguous and those that are actually perceived as ambiguous? To determine how trust in gender-ambiguous voices actually plays out, an online experiment ( N = 343) was conducted with a 3 (male, ambiguous, female voice) x 3 (male, neutral, female topic) design. After listening to the randomly assigned voice and topic, participants rated their perception of different dimensions of trusting beliefs, intention, and trusting behavior as intention to adopt. The results show that the acoustically ambiguous as well as the perceived-as-ambiguous voice differ negatively from the gendered voices in several aspects, indicating disfluency. Even acoustically ambiguous voices with an ascribed gender do not automatically evoke similar effects as voices that are acoustically clearly assigned to a gender. This indicates that overall, gender-ambiguous voices in voice assistants—at the moment—might result in negative attitudes towards the device and the voice itself for the average user, thus hindering rather than enhancing acceptance of non-binary gender images and voices. However, all three grouping variables that were used to measure gender effects showed effects on different aspects of trust. This suggests that future studies need to clarify what form of gender ambiguity and what exact differences they are investigating and why this decision was made. Furthermore, the results show that even among the voices with a clear gender, trust did not conform to classic gender stereotypes: while the male voice was perceived as particularly secure, the female voice was considered more competent. Overall, it can be deduced from this that automated communicators are not necessarily subject to the same social patterns as human communicators. While early pioneers of human-machine communication such as Reeves and Nass still assumed that regularities and heuristics from social research could also be transferred to machines, the results of this study indicate that communicating technologies are now developing their own heuristics—in our case a kind of “social gender role” for voice assistants. However, it is also clear that this role has its limits: While gender stereotypes are being reinterpreted for voice assistants, they still contain a clear dichotomy of male and female. If a voice does not fall within this norm—acoustically or ascribed—it is less trusted than gender-specific variants.