An experimental study of action selection mechanisms to create an entertaining opponent

The final step in the Monte Carlo Tree Search algorithm is to select the action to play from the root level of the tree. Experimentation on modifying the selection mechanism has been somewhat limited to date, particularly with respect to consider aspects other than playing strength. This paper investigates the modification of selection mechanism as an attempt to produce a more entertaining opponent in the strategic card game Lords of War. These selection mechanisms are played against our most effective Information Set MCTS agent, and we investigate their performance in terms of measures of performance and complexity. An interesting side effect is that one of the action selection mechanisms results in a significant improvement in ISMCTS play strength. We also experiment with online tuning of configuration parameters in an attempt to create an agent with dynamically scaling play strength.