AI-driven non-playable characters (NPCs) have been developed to enhance the gameplay experience. However, recent studies reveal that such NPCs do not always achieve this goal. Despite their intelligence, AI-driven NPCs still rely on fixed behavioral models, similar to scripted NPCs, resulting in monotonous gameplay experience. This paper addresses this limitation by introducing artificial life as NPCs. Unlike conventional AI, artificial life can memorize, learn, and evolve through interactions with the game environment and players. This enhances games in two key aspects: variability, allowing games to change dynamically and unpredictably during and between playthroughs, and freedom, allowing players to interact with NPCs without being limited to fixed interactions, both identified as essential for player satisfaction in prior research. We implement artificial life using a neural network that determines actions based on the environment, updating its weights during gameplay to enable real-time learning within its lifespan. Experimental results in simulated gameplay scenarios show that it enhances gameplay variability and freedom, while computational efficiency analysis on a CPU platform confirms its real-time feasibility and minimal resource requirements. The code is available at .