Smart Game Development Systems: Artificial Intelligence for Emerging Gaming Experiences

The purpose of this study was to evaluate the application of modern AI techniques to game development from a comparative perspective and to examine how different AI approaches influence gameplay behavior and adaptability in game environments. In this study three general approaches were implemented. Using the UE5 engine, agents were created with elaborate control systems and numerous attack states in the form of behavior trees. Within Unity, training data was used to implement reinforcement learning agents using PPO in the ML-Agents environment, allowing the agents to develop behavior through a reward-based learning process that optimizes behavior according to the consequences of agent actions within the environment. The NEAT algorithm was implemented in a Python-based simulation environment, where agent intelligence evolved through multiple generations by iteratively improving neural network structures and behaviors. Behavior trees are used in this context to provide structure for the creation of intricate series of actions; RL allows the use of experience for optimized tactical approaches based upon player interaction; NEAT offers a way for agents' skills to iteratively improve. The comparative analysis demonstrates how these AI paradigms differ in terms of decision-making, adaptability, and behavioral diversity within game development environments. Overall, this study provides a comparative perspective on the implementation and practical characteristics of modern AI approaches for game development.