Chatbots' gender stereotypes: individual perceptions of the gender of chatbots in South Africa

Ndala, Faith Nokuthula
Journal Title
Journal ISSN
Volume Title
Chatbots are getting closer to being integrated into our social environment and embodying our personalities, and how individuals react to information conveyed by chatbots is influenced by how they view chatbots. Research shows a small body of scholars studying individuals’ views on chatbots and a minimal focus on gender. The views of gender affect how people interact with each other and potentially with chatbots. Although progressive steps towards gender equality have been taken in many countries over the last few decades, no country in the world has yet achieved gender equality. Considering the history of South Africa, gender imbalances date back to the apartheid years, when white men were primarily in leadership roles. Also, in the new age of democracy, women still suffer from the dominance of the patriarchy in the cultural, social, and political spheres. We may conclude that using chatbots with the unknown effects on how gender-based chatbots may reinforce gender inequalities in South Africa may increase inequality in the light of history. The study’s purpose was to explore the individual perceptions of the gender of chatbots in South Africa to raise awareness among chatbot designers of the implications that the gender of chatbots may have in society. To achieve the study’s aim, the researcher adopted the Social Role Theory, Social Identity Theory, and Similarity-Attraction Paradigm to understand how gender roles are reinforced by people behaving according to them. In turn, these theories were used to address how people’s perceptions of chatbots are influenced by its gender–by focussing on (1) how people respond emotionally in a controlled environment to chatbots of different genders performing the same role, (2) the influence of a chatbot’s gender on the user’s responses during the interaction, (3) the influence of a chatbot’s gender on user’s experiences, and (4) how people’s prior-existing expectations of the role of the chatbot are affected by its gender. Data were collected qualitatively through chatbot interaction and interviews. The researcher provided an experience for the participants by creating chatbots to assess the participants’ experiences while communicating with chatbots of different genders and the same role and measuring how participants reacted emotionally to these chatbots in a regulated environment. The research design was based on a 2 x chatbot gender (male vs. female), 2 x participant gender (male vs. female), and 2 x stereotyped subject domain (auto mechanic and midwife) matrix. Following this, semi-structured interviews were conducted to obtain in-depth insights into the phenomenon while also aligning the results of the interviews with those gathered after the participants interacted with the chatbots. The participants interacted with the chatbots and were interviewed on the same day. The researcher was not present during the interactions with the bots, but interactions were recorded for analysis. The researcher only joined the participants during the interviews, which were also recorded for analysis. In South Africa, the results indicate that gender stereotypes in the offline world also exist in the online world and are more endorsed by males than females. People tend to associate feminine characteristics with female bots and masculine with male bots. These stereotypes are conditioned mainly by the societal norms developed to stipulate how people should act in societies and what work roles they should undertake. These attributes then make female bots more suited to providing health services and male bots to providing mechanic services. The female mechanic chatbot was more forgiven when it could not help with car diagnosis and was commonly perceived as incompetent. On the contrary, the perceptions about the male chatbot in the health sector were more about feelings around vulnerability. While results revealed the same-gender preference as found in similar studies, this research also indicates cross-gender preferences such that some female participants preferred male bots, and male participants preferred female bots. The theory of social roles has emerged as more substantial than the other theories in defining and understanding gender stereotypes concerning chatbots. The findings also support Giddens’s view that individual agency can reinforce and can change the social structure of gender stereotypes in this context. The use of Social Role, Social Identity Theories, and Similarity-Attraction Paradigm present more of an application of these theories to clarify chatbots’ gender stereotypes where they have not been implemented in the past. The Structuration Theory brings together these ideas under the umbrella called structures to understand how gender roles are reinforced by people acting in accordance with them. This study can pave the way for future research into how current social norms and stereotypes manifest in Artificial Intelligence interaction. In addition to methodological and theoretical contributions, the thesis can make practical insights into the decision-making process regarding chatbots’ gender and Artificial Intelligence in general. This, in turn, may help drive gender equality through the use of attribute-based chatbots to facilitate improvements in the understanding of gender roles in South Africa
A research report submitted in partial fulfilment of the requirements for a degree of Masters in Commerce at the University of Witwatersrand, Johannesburg, Faculty of Commerce, Law and Management, 2020