Conducted research (with Dr Selen Turkay and Joel Harman as my supervisors) using a virtual reality medium with an AI (ChatGPT) configured program to investigate the impact of intelligent NPCs on user experience. Built a VR environment (using Joel Harman's base project), with NPCs that used ChatGPT, over a 10-week period and conducted a study to collect data. Utilised Unity, C#, ChatGPT and research methodology. The aim of the game is to budget for weekly expenses with Isabelle and Freya. 

Controls: WASD movement, R to interact with objects, hold E to speak when in range with Isabelle or Freya.

​

Abstract:

Since the recent advancements of artificial intelligence, particularly with Large Language Models (LLMs), the notion of virtual reality (VR) and intelligent agents assisting with user learning and enhancing their interactive experiences has been of great interest. However, there is still substantial dislike of intelligent agents from the public. This research intends to engage users in a positive way with intelligent agents and investigate what attributes, in the agents, evoke a positive emotional response, increase user motivation and sense of presence. This study aims to investigate the impact that embodied intelligent agents who express uncertainty have on user experience in VR and whether users can identify this uncertainty. The research question that guided this investigation was: 

​

How do embodied agents that express uncertainty impact users’ experience in VR?

​

The study was exploratory. Ten participants completed surveys before and after the VR experience to measure their emotional response, intrinsic motivation and sense of presence. Then, they interacted with the virtual environment and the two embodied agents (one that appeared uncertain, and the other was sure about the information they provided). After, participants also engaged in an interview about their experience with the researcher. From the initial interviews, participants often preferred one agent over the other (and not necessarily the agent that expressed uncertainty). This preference came from the agent appearing assertive, friendly, human-like or approachable. It is hypothesised that users will have a more positive emotional response to agents, the more human-like they appear. However, users will be more motivated to engage with an activity when agents appear sure of themselves and give clear instructions. This study shows potential for intelligent agents to appear approachable and enhance users’ sense of presence in VR.