Communicative and social consequences of interactions with voice assistants

• In Event: Poster Session 02
  In Poster Session: PS 02 Topic - Social Cognition

Abstract

The growing prevalence of artificial intelligence and digital media in children’s lives provides them with the opportunity to interact with novel non-human agents such as robots and voice assistants. Previous studies show that children eagerly adopt and interact with these technologies, but we have only limited evidence of children’s distinction between artificial intelligence and humans. In this study, the communication patterns and prosocial outcomes of interactions with voice assistants were investigated.

Children between 5 and 6 years (N = 72) of age were administered two tasks, a treasure hunt task and a prosocial task. In the treasure hunt task, children interacted with two human experimenters in one of four conditions. The conditions were a combination of the interaction partner (a human or a voice assistant) and the information type (knowledge or experience) during the treasure hunt, resulting in four between-subjects conditions (human – knowledge, human – experience, voice assistant – knowledge, voice assistant – experience). While human experimenter 1 (E1) interacted with the child in the study room during the familiarisation for the treasure hunt, human experimenter 2 (E2) was only present via her voice in all conditions. After the introduction, children completed a treasure hunt together with E2. During the treasure hunt, E2 either provided personal experience-based or knowledge-based information about the treasure hunt objects (information type). In the second part of the study, children were administered tasks on sticker sharing (Fehr et al., 2008; Moore, 2009) and helping (Beier et al., 2019; Warneken & Tomasello, 2006) to investigate the prosocial consequences.
In the treasure hunt task, we measured how often children talked to the interaction partner during the search phases in which information sharing was optional but not obligatory: children talked mainly about the search process (“I can't find the box”) and about the opening of the boxes (“Should I open the box?“) or they did not communicate with E2 (communication). In the sharing task, we coded for each trial whether children chose the prosocial option (choosing the (1, 1) distribution in the no cost sharing and the costly sharing, or the (1, 2) option in the no cost envy, or the (2, 3) option in the costly envy trials) or not (sharing).

Results revealed that children provided voice assistants with less information than humans and that only the type of information given by a human interaction partner was related to children’s information selection, see Figure 1. Sharing was influenced by an interaction between type of information and interaction partner, showing that the type of information shared influenced children’s decisions more when interacting with a human, but less when interacting with a voice assistant, see Figure 2. Children in all conditions enjoyed the treasure hunt with the interaction partner.
Overall, these results suggest that children do not impose the same expectations on voice assistants as they do on humans. Consequently, cooperation between humans and cooperation between humans and computers differ.

Authors

• Anja Gampe, University of Duisburg-Essen
  Presenting Author

• Sara Aeschlimann
  Non-Presenting Author

• Marco Bleiker, University of Zurich
  Non-Presenting Author

• Michael Wechner
  Non-Presenting Author