## Abstract This thesis, conducted in collaboration with Sørlandet sykehus helseforetak and their specialists from the Selvmordsforebyggende nettverk, explores the use of artificial intelligence (AI) as a virtual patient in a virtual reality (VR) simulation. This application extends the hospital’s current use of VR headsets for simulation training with 360-degree video, aiming to assess whether AI can convincingly replicate human behavior for effective training. Simulation training allows healthcare personnel to practice handling challenging scenarios in a risk-free environment, preparing them for real-world situations. The hospital currently uses staff role-playing and 360-degree videos as primary simulation methods. To develop a realistic simulation, data was gathered from specialists through a mixed-methods approach. Initial insights were obtained from semi-structured group interviews and the first prototype. Further feedback on the final version came from individual interviews and surveys. The simulation was built using Inworld, Unreal Engine, and MetaHumans. The final simulation version offers efficiency in cost and time, along with replayability due to the AI’s varied responses. However, some issues remain, such as unnatural facial expressions and occasionally unrealistic, lengthy answers from the AI. Future research will focus on continued refinement of the AI and exploring additional applications for this AI-supported simulation. ## Methodology Overview ### Usability and User Experience Goals Our methodology begins with a focus on usability and user experience (UX), ensuring that interactive products are both functional and engaging. Usability goals include: - **Effectiveness**: Can users achieve their objectives with the product? - **Efficiency**: How productive are users once they are familiar with the product? - **Safety**: What measures are in place to help users recover from errors? - **Utility**: Does the product offer all necessary functions? - **Learnability**: How easily can users learn to use the product? - **Memorability**: How well can users remember how to use the product? User experience goals, focusing on emotional and subjective aspects, complement these usability goals, emphasizing the overall user satisfaction and engagement. ## Design Solutions and Prototyping **Design Solutions**: Following the specification of user requirements, we move to creating design solutions through iterative prototyping. This process involves:

**Conceptual Models**: Initial high-level visual representations guide the development of detailed designs.
**Low-Fidelity Prototyping**: Early sketches and storyboards outline basic interface and environment designs, leading to the creation of a functional patient model.
**Digital Prototyping**: A basic working prototype is developed, allowing for user interaction with preliminary features. Feedback from specialists leads to improved designs.
**High-Fidelity Prototyping**: Advanced prototypes incorporate detailed animations, realistic environments, and refined interactions based on continuous feedback.

## Evaluating the Design **Evaluation Process**: The final stage involves comprehensive usability testing to ensure the design meets user needs. This includes:

**Developing the Test Plan**: Outlines testing procedures, including VR prototypes, interview questions, and survey instruments.
**Setting Up the Testing Environment**: Conducts tests in designated meeting rooms with time constraints managed effectively.
**Selecting Participants**: Engages specialists and users from relevant departments to provide diverse feedback.
**Preparing Test Materials**: Includes VR equipment, prototypes, and consent forms for structured testing.
**Conducting Test Sessions**: Implements semi-structured interviews, observations, and surveys to gather detailed feedback.
**Analyzing Data**: Uses thematic analysis to organize and interpret qualitative data, ensuring insights are aligned with research objectives.

## Discussions ### From Theory to Practice The simulation aligns with several key learning theories: - **Constructivism**: Requires active user engagement with the AI patient, rather than passive observation. - **Experiential Learning**: Users benefit from interacting with the AI patient across multiple sessions, allowing for action and reflection. - **Situational Learning Theory**: Provides a realistic environment to practice communication skills, reflecting real-life interactions with patients. - **Cognitive Load Theory**: Allows adjustment of interaction difficulty to manage cognitive resources effectively and prevent overwhelm. These theories support a learning environment that fosters active engagement, reflection, and skill development in a controlled setting. ### Designing AI-Supported Virtual Reality Environments for Suicide Prevention Training The design of the VR simulation evolved from Prototype Lisa to the final version, Molly. Key improvements included: - **AI Enhancements**: Added situational awareness, long-term memory, and relationship features to create a more realistic and functional simulation. - **Patient Background**: Updated to reflect a believable context, including considerations for gender differences and realistic patient scenarios. - **Communication and Body Language**: Adjusted to better reflect typical patterns of individuals with depression. Reduced facial expressions and focused on realistic body movements. - **Visual Fidelity**: Created an office environment based on user feedback, with potential for expansion to other settings. - **Relationship Building**: Integrated new features, though their effectiveness in fostering trust and openness was not fully tested. - **Emotional Dynamics**: Maintained a consistent portrayal of sadness, reflecting real patient emotional stability. - **Topic Reactions**: AI’s lack of topic preferences aligns with real patient behavior, but future versions could include some positive feedback. The simulation combines VR and role-playing methods to offer a cost-effective and immersive training tool. ### AI-Based Characters and Real-Time Interaction in Simulation Training The AI's role primarily involves dialogue, with mixed feedback on its realism: - **Engagement**: Participants felt engaged, though there were concerns about the AI’s response length and voice quality. - **Visual Enhancement**: The visual components significantly improved the training experience. The AI’s effectiveness in representing real-time interactions remains unclear, but the overall feedback was positive. The integration of AI with visual elements creates a promising training tool, with potential for further development.