Poster #111 - "Alexa, What do you Think?” - Children’s Theories of Artificial Minds and Learning with Digital Assistants

• In Event: Poster Session 06

Abstract

Voice-based digital assistants (DA), such as Amazon Alexa and Google Assistant, are common in western homes. However, its impact on child development is still under-researched. Of interest is how children at critical periods for development of Theory of Mind and other social cognition skills (4-8 years) conceive of DAs. Children’s conceptions of Artificial Intelligence (AI) systems are the foundation of an inter-disciplinary area of study regarding Theory of Artificial Minds (ToAM). We are investigating whether mechanisms underlying how children conceive of other human minds might be similar to how children understand “artificial minds” or AI technologies.

This preregistration poster outlines how we will conduct an in-person mixed-methods study with English-speaking 4–8-year-old children (15-30 per age, n = 150) and their parents (n = 150). The three guiding research questions are RQ1: How do young children conceive of voice-based artificial intelligent technologies, i.e., their ToAM for DAs? RQ2: What is the impact of DA use on children's ToAM? RQ3: How does children’s ToAM affect how they learn from DAs? Three aspects of children’s ToAM will be explored and measured. First, a parent survey will be conducted assessing 5 areas: demographics, DA uses (items, n = 13), types of DA queries (17), child-DA interactions (5), child-DA relationships (10) and parental mediation strategies (6). We developed these items based on a pilot study (n=50) and a scoping review of existing DA research and industry categorizations (Amazon, Google etc.). For the DA parental mediation scale, Livingstone et al. (2017) parental mediation scale of internet use was adapted for DA use. Second, children will complete three individually administered behavioral tasks. Two are ToAM tasks assessing whether children ascribe any beliefs, desires, or intentions to DAs. These ToAM tasks will be adapted from classic false belief and intention-explanation ToM tasks. The last DA task will explore how children learn from conversations with DAs. Third, a brief semi-structured interview will follow the tasks so that participants can elaborate on answers from previous tasks.

To analyze the parental survey, descriptive statistics will detail frequency of use, interactions, and parental mediation. For ToAM tasks, children’s responses will be coded to understand the language children use to describe DA actions. The coding scheme will differentiate whether children use more human-like mentalistic explanations or more computer-like object-oriented language to explain DA behavior. For the learning task, a coding scheme will code for how children gauge the information they receive from DA in terms of trust and verification. Results will indicate how children perceive AI entities around them, how they understand the inner workings of DAs, and whether they imbibe DAs with human-like qualities (intention, personality, belief, etc.). It is hypothesized that these ToAM competencies will affect how children reason about information learned from the DA. Factors that may account for performance differences are age, gender, prior DA use, and parental mediation. Findings will contribute to a growing body of knowledge about the influence of AI-based technology on children and how aspects of child development may play out in connected homes.

Authors

• Nandini Asavari Bharadwaj, McGill University
  Presenting Author

• Adam Kenneth Dubé, McGill University

• Victoria Talwar, McGill University