This invention relates to the design of intelligent agents, specifically a simulation-driven agent for card games with adaptable intelligence levels. Prior to making a move in the actual game, the agent's simulation module uses Monte Carlo simulations to generate multiple simulated gameplay scenarios. These simulations may include at least two participants, where the hand of one participant remains unknown to the agent. A prediction module employs a Bayesian network to estimate the unknown hand within the simulated games. A logic module applies rule-based reasoning to determine a sequence of actions throughout the simulations. Each action is evaluated with a corresponding payoff, and the action with the highest overall payoff is selected as the move to be executed in the real game.
Figure (1) ISBA in action; (2) Opponent Hand Prediction model; (3) Bayesian network model for opponent hand prediction
Traditional card game agents often struggle with imperfect information environments like Rummy due to the complexity of predicting opponents' hands and evaluating the best moves in real-time. Human players find it challenging to anticipate opponent strategies and optimize their own moves consistently. Existing AI agents are often limited by relying on fixed datasets of human play, restricting their intelligence to human skill levels and lacking adaptability. There is a need for an intelligent agent that can dynamically predict opponents' hands, simulate multiple game scenarios, and make optimal decisions to outperform human players and traditional agents.
- Monte Carlo Simulation-based Decision Making: The agent uses multiple simulations to evaluate various possible moves, improving decision accuracy in uncertain game states.
- Bayesian Network for Opponent Hand Prediction: The agent predicts the opponent’s hand probabilistically, enhancing strategy formulation beyond random guessing.
- Adjustable Intelligence Level: The number of simulation rollouts controls the intelligence level, allowing a customizable challenge for players of different skill levels.
- Combination of Rule-Based Logic and Neural Network Models: It enables realistic and strategic actions within simulations, including card picking and discarding decisions.
- Superior Performance over Traditional Agents: It was demonstrated to outperform agents lacking predictive or simulation capabilities by significant point margins.
- Applicability to Multiple Player Variants: While demonstrated on two-player Rummy, the approach generalizes to games with three or more players.
- Real-Time Decision Making within Time Constraints: The agent efficiently generates simulations and evaluates moves within gameplay time limits.
The Intelligent Simulation-Based Agent (ISBA) for card games integrates three core modules: a simulation module that generates multiple Monte Carlo simulated games to evaluate possible moves, a prediction module that uses a Bayesian network to estimate the opponent’s hand, and a logic module that applies rule-based pick and discard strategies within each simulation. By iteratively simulating game scenarios, calculating cumulative payoffs for various enforcement actions, and selecting the move with the highest payoff, the ISBA adapts its intelligence level based on the number of simulations run. This approach enables the agent to make informed decisions in real-time during actual card games like Rummy, outperforming traditional agents and offering adjustable difficulty to suit players of varying skill levels.
This technology is available for licensing.
9
This technology enhances gaming by offering an adaptive AI opponent that helps players improve their skills and enjoy the game. It acts as an educational tool for beginners and advances AI research in decision-making under uncertainty, with potential applications beyond gaming. By enabling smarter digital card games, it boosts user engagement and supports industry growth. Its adjustable difficulty makes gaming more inclusive, appealing to both casual and competitive players.
- Online and Mobile card games
- Gaming AI development
- Educational gaming platforms
- Simulation and decision support systems
- Artificial Intelligence research
Geography of IP
Type of IP
202221061265
554181