Assessing The affect of Predictability on Interaction With AI in VR

Abstract

In this era of technological innovation, the fusion of artificial intelligence (AI) with user experience (UX) presents a transformative shift in human-computer interaction (HCI). This thesis investigates the impact of predictability in AI-driven virtual reality (VR) environments on user experience and task performance. The pre-study aimed to operationalize the concept of "predictability," revealing significant differences between predictable, unpredictable, and manual conditions. The main study focused on task completion duration, user satisfaction, perceived control, cognitive load, and error rates across these conditions. Results indicated that participants completed tasks significantly faster in the predictable condition compared to the manual and unpredictable conditions, suggesting that predictability enhances efficiency and streamlines user interactions. However, no significant differences were found in user satisfaction, perceived control, NASA Task Load Index (NASA-TLX) scores, or error rates across conditions. Qualitative feedback revealed that while the predictable condition was described as "smooth" and "satisfying," the unpredictable condition elicited frustration and confusion, highlighting the importance of considering both quantitative and qualitative data in user experience evaluation. The study’s limitations include the novelty effect of VR and the focus on a specific task, which may not fully capture real-world AI interactions. Future research should explore predictability in diverse AI applications, conduct longitudinal studies, and consider user diversity to enhance the generalizability of findings. This study underscores the critical role of predictability in AI systems, providing valuable insights for designing more intuitive and efficient AI-driven environments.