AI Director vs. No Director: Evaluating Dynamic Difficulty in a Dungeon Crawler

This study explores the design, implementation, and evaluation of an adaptive AI Director system in a dungeon crawler prototype, Somniat. The AI Director dynamically adjusts enemy spawning and item distribution based on player behaviour, combat style, and in-game progression to enhance engagement and fairness. By classifying players into archetypes and responding to factors like combat preferences, stat distributions, and current health (lucidity), the system tailors gameplay in real time. A controlled A/B testing framework compared sessions with and without the AI Director, gathering telemetry and qualitative feedback. Results indicate that while no statistically significant differences were observed due to limited sample size, the adaptive version produced more consistent and satisfying experiences. Most testers preferred the adaptive system for its balanced stat allocation, responsive enemy encounters, and dynamic difficulty scaling. The findings suggest that AI-driven difficulty adjustment enhances player engagement. with broader implications for game accessibility and adaptive learning environments. Future work will involve expanded testing, affective data integration, and exploration of transparency and player agency in adaptive systems.