Development of a Video Game Where NPCs are Submitted by Users and Moderated Through AI

Abstract

This thesis presents a novel system for integrating user-generated content, specifically Non-Playable Characters (NPCs), into a video game environment through an AI-moderated submission process. The video game was developed using the Godot Engine, with GDScript as the chosen language for scripts. Models for the game were created with Blender. The game features a mix of interaction and action stages, with the NPCs appearing in the interaction stages to contribute to the atmosphere of the game. It was concluded that developing a video game can be more time consuming than initially stimated, even if the result is a short game. A companion website, built using NextJS (following an initial development phase with Create React App), provides a user interface for NPC submission, dialogue contribution, and character browsing. Characters are stored and fetched from an API that uses Amazon Web Services (including API Gateway, Lambda, DynamoDB, and Amplify), all constructed to be serverless. Content is moderated by sending it to Anthropic’s Claude AI before submitting, the system incorporates a confidence score from the AI, enabling prioritization of submissions for human review. The research concludes that a hybrid approach, combining AI's capability to understand context with human judgment for more complex cases, offers the most effective and efficient content moderation strategy for user-generated content in this context.