Personality dimensions of intelligent personal assistants

Abstract

An earlier study of intelligent personal assistants (IPAs) identified a large number of user utterances aimed at understanding an IPA's personality by testing its responses to questions about itself and about users. Based on primary and secondary sources, we developed a dataset of IPA-directed personality utterances and analyzed IPA responses to such utterances from four major IPAs-Amazon Alexa, Google Assistant, Apple Siri and Microsoft Cortana-using Fiske's stereotype content model. The model supports classification of user perceptions of other people or objects on warmth and competence dimensions. The findings suggest that while all four IPAs were frequently perceived as competent, Alexa and Siri were more frequently perceived as low-warmth compared to the higher-warmth Google Assistant and Cortana. Future work aims to verify the initial classification of IPA personality responses using warmth-competence dimensions, and test user demographic and situational preferences for personality manifestations in IPA responses.