Heuristic optimization of AI agents for Dhumbal

Abstract This study evaluates Artificial Intelligence (AI) agents for Dhumbal, a culturally significant multiplayer card game with imperfect information, by systematically comparing rule-based, search-based, and learning-based strategies. We formalize Dhumbal’s mechanics and implement diverse agents, including heuristic approaches (Aggressive, Conservative, Balanced, Opportunistic), search-based methods such as Monte Carlo Tree Search (MCTS) and Information Set Monte Carlo Tree Search (ISMCTS), reinforcement learning approaches, including Deep Q-Network (DQN) and Proximal Policy Optimization (PPO), and a random baseline. Evaluation uses multiple tournaments, each comprising independently shuffled rounds, while accounting for dependence among agents’ outcomes within each round. Performance is measured via win rate, economic outcome, Jhyap success, cards discarded per round, risk assessment, and decision efficiency. Primary comparisons rely on effect sizes (Cohen’s d ), 95% confidence intervals, and a round-level bootstrap sensitivity analysis. Across 1024 simulated rounds, the rule-based Aggressive agent achieves the highest win rate (88.9%, 95% CI: [86.9, 90.8]), outperforming ISMCTS (9.3%) and PPO (0.2%) under the evaluated configurations. The study contributes a reproducible AI framework, insights into heuristic efficacy under partial information, a discussion of the ethical risks of deploying dominant agents in a game traditionally played for money, and open-source code, thereby advancing AI research and supporting digital preservation of cultural games.