Safeguarding children from harassment in social VR: integrating child, parent and expert perspectives into the design of automated embodied moderators

Children grow up in a digital world with constant online connection, changing the way they socialise, interact and communicate. Social Virtual Reality (VR) has emerged as one of the technologies that attracts children and teenagers, despite age restrictions typically 13+. In contrast to 2D digital media, social VR leverages VR headsets’ unique affordances that allow embodiment, presence and immersion. Users can now interact via an embodied avatar synchronously in 3D-360-degree high-fidelity social virtual environments, providing rich visual, auditory, haptic and phantom touch sensations, which allow for mimicking face-to-face interactions. However, with these new forms of socialising come challenges, including new forms of harassment, not only verbal, but also physical (e.g., avatars invading personal space) and environmental (e.g., displaying inappropriate content). Children and adults now share these unsupervised social VR spaces, while still being anonymous, which has led to numerous harassment events, including group harassment. First, a gap remains in understanding child and parental involvement in this space. Second, child-friendly solutions are lacking, and the effectiveness of the existing safety features and their usage by children are not evaluated. Third, there is a noticeable absence of involvement from key stakeholders such as experts, parents, and children themselves in research studies. To address these gaps, in Part I of this thesis, through three empirical studies (a mixed-methods questionnaire with 79 parents and 70 non-parent social-VR aware adult users, interviews with 16 experts and a user study with 40 children 8-16 years old) the research aims to 1) explore and better understand issues and concerns associated with child use of social VR, 2) inform the design of child safety features in social VR. The findings reveal concerns regarding the freedom children have in social VR, the increased harassment situations and the current lack of safety and supervision tools. Then, the research underscores the critical insight that existing safety solutions place responsibility on children and the necessity of creating child-friendly, real-time safety tools such as quick-access, discreet and engaging features to effectively safeguard children against one-to-one and group harassment in social VR. With AI moderation becoming promising and leveraging embodiment, in Part II, I propose a novel approach to safeguard children in these spaces using AEMs, AI-driven embodied safety agents to respond to harassment in social VR in real-time, designed with parents, children and experts. I developed “Big Buddy”, a Wizard-of-Oz AEM prototype and through five studies with children aged 8-16 years old, parents and experts (user studies with 43 children and 17 parents, 34 children, 34 parent-child dyads respectively, workshops with 13 children 8-16 years old, 5 parents, 3 grandparents and interviews with 16 experts), the research aims to 1) explore both experiential and conceptual aspects of AEMs from stakeholder perspectives and 2) identify the requirements for AEMs to effectively engage both children and their parents, while ensuring children remain empowered and supported in the process. Considering the lifecycle of a moderation incident (before, during and after a child is harassed in social VR), and informed by the findings, I present how I envision an AEM to work in practice, highlighting considerations around the role of AEMs, child users and parental settings and how parents and children can be involved. Finally, I propose design guidelines to develop child-friendly social VR safety features. This thesis addresses critical gaps and establishes a solid foundation for child safety in social VR.