Many modern NPC systems combine behaviour trees, utility AI, planners, and increasingly large language models. While these systems can generate convincing actions or dialogue, they often lack a coherent internal model explaining why an NPC behaves the way it does. This talk explores how computational cognitive models can bridge that gap. Drawing from cognitive science and game AI research, I present a practical architecture combining Belief–Desire–Intention (BDI), utility-based decision making, motivational systems inspired by Maslow’s hierarchy, emotional regulation through the Yerkes–Dodson model, and social reasoning to create more believable and explainable NPC behaviour. Rather than proposing another AI algorithm, this talk demonstrates how existing game AI techniques can be organised around computational models of cognition to produce NPCs that appear consistent, adaptive, and socially believable while remaining practical for real-time games. Attendees will leave with a perspective on cognitive computational models, implementation strategies, and lessons learned from building a living social simulation in Unity.