The field of game AI and procedural content generation is moving towards more sophisticated and dynamic methods for generating game content and improving agent decision-making. Recent research has focused on combining machine learning techniques, such as deep reinforcement learning and learning from demonstration, with traditional game tree search methods to create more realistic and challenging game experiences. Another area of interest is the development of hybrid procedural content generation methods, which leverage the strengths of both machine learning-based generation and search-based repair to create high-quality, functional game levels. These advances have the potential to greatly enhance the replayability and engagement of games, and could also have applications in other areas such as simulation and planning. Noteworthy papers include:

• Elevating Styled Mahjong Agents with Learning from Demonstration, which presents a novel algorithm for learning from demonstration that preserves the unique play styles of Mahjong agents.
• NTRL: Encounter Generation via Reinforcement Learning for Dynamic Difficulty Adjustment in Dungeons and Dragons, which proposes a reinforcement learning approach to automating dynamic difficulty adjustment in Dungeons & Dragons.