Poster #88 - Do children trust a smart speaker when learning factual information and making moral decisions?

• In Event: Poster Session 06

Abstract

Smart speakers (Amazon Alexa, Google Home) are designed as trustworthy sources of information about such things as the weather and traffic. Recent research suggests children (4-8 years) prefer learning facts from smart speakers over humans (Girouard-Hallam & Danovitch, 2022). Yet, smart speakers’ answers to children’s queries can have moral implications. Recently, when asked for a challenge by a 10-year-old, an Alexa suggested the child press a penny onto a plug partially inserted into an outlet (BBC, 2021). Fortunately, the child knew better. As smart speakers become increasing ubiquitous – 41% of families with children 8 years and younger have a smart speaker in their homes (Common Sense Media, 2020) – important questions emerge regarding children’s trust in smart speakers especially when there are moral implications. The current study aims to address this question by investigating children’s use of confidence as a credibility cue when learning factual information and making moral decisions. Participants (N=128 planned; 5-8 years) were randomly assigned to either the factual or moral condition. Using the selective social learning paradigm, participants viewed videos of two pairs of informants – one confident and one hesitant – provide different answers to questions about animals (Figure 1). Participants listened to four trials with human informants (confident and hesitant) and four trials with smart speaker informants (confident and hesitant) (order of informant type was counterbalanced). In the factual condition, participants were presented with factual questions about two animals (e.g., “Here is a cow and a horse. Which of these does not have a toma?”). Novel “facts” were used to control for children’s actual knowledge. In the moral condition, participants were presented with questions about animals that invoked moral principles, such as fairness and harm (e.g., “An elephant and a giraffe at the zoo are both very sick, but there is only enough medicine for one of them. Which of these should get to take the medicine?”). Across both conditions, the confident informant selected one animal using linguistic and paralinguistic cues of confidence, whereas the hesitant informant selected the other animal using linguistic and paralinguistic cues of uncertainty. To assess learning preferences, participants were asked which answer they endorsed for each trial (e.g., “What do you think – the horse or the cow?”). Additionally, participants rated both informants in the pair on a 4-point scale (0=not at all to 3=a lot) in terms of the informant’s level of confidence, likeability, and smartness. The procedure was then repeated with the other informant pair (humans or smart speakers). Participants’ responses (0 = Hesitant informant; 1 = Confident informant) to the four learning trials were modeled simultaneously with random-intercept logistic regression with age, condition (factual, moral), and informant type (smart speaker, human) as predictors and participant ID as a random effect to account for the repeated responses across trials. Mixed Effects Models were used to test to the effects of condition (factual, moral), informant type (smart speaker, human), and participant age on ratings of the informants’ confidence, likeability, and smartness.

Authors

• Rachel L. Severson, University of Montana
  Presenting Author

• Sarah Sweezy, University of Montana

• Tom Macko, University of Montana

• Shailee Woodard, Keene State College