Visualizing Decision Trees in Games

Introduction:
This article by Haworth, Sheida, Bostani and Sedig involved an experiment in which the researchers designed a basic digital maze-game for subjects to play, filled with tile-based decisions which could be represented by a decision tree. The goal was to determine the effects of having access to the visualized decision tree on player decision-making.

Summary:
In the digital maze program designed for this experiment, the mazes were filled with tiles of varying effects - some forced a player to move one square in the indicated direction, while others might change the color of the player's avatar, enabling it to collect certain objects. Touching maze walls forced players to restart, and players' goal was to reach the door marking the end of each level. Many levels were designed with clever traps and delays for players to fall into. While there was no time constraint to force players to rush, players' careful consideration of each move was balanced by rewarding efficient clears of maze levels in the fewest possible number of moves, while punishing overly long routes. In this context, the ability to visualize one's options for each decision and its consequences would save moves and thus be beneficial to play, in theory.

Four variations on the game were part of the test:
1. IT+NK: Interactive Tree, No keys - Navigation by mouse (clicking on options in the decision tree) only
2. NIT+K: Non-interactive Tree, keys - Navigation by keyboard input only. Tree available for viewing.
3. IT+K: Navigation by mouse or keyboard, by choice. Tree available for viewing or navigation.
4. K: No tree available, navigation through maze by keyboard input only.

A sample of the decision tree available to players in game variants 1, 2, and 3.

The results of the test were interesting. None of the groups used the tree for decision-making early on, when paths were obvious and decisions straightforward, but those with access to the decision tree relied on it more heavily the more difficult and "trap-filled" the levels became. Trees were used to avoid traps at first, but eventually (with practice), subjects reported using trees to plan out navigation into specific areas or toward specific objectives. While participants tended to dislike navigation through clicking on the tree, they preferred the game variants with access to the tree to the one without it. Most of the participants felt that the game would be too difficult without access to the decision tree, requiring extensive trial and error which would rapidly turn it from a fun challenge into a repetitive chore.

Decision trees are integral to many games, both on the side of players, and on the side of designers in how their game functions and how AI makes decisions. This study ends with recommendations for further study into how to make the helping-power of decision trees support game design without intrusively ruining the experience.

Source:
Robert Haworth, Sousan Sheida Tagh Bostani, and Kamran Sedig, “Visualizing Decision Trees in Games to Support Children's Analytic Reasoning: Any Negative Effects on Gameplay?,” International Journal of Computer Games Technology, vol. 2010, Article ID 578784, 11 pages, 2010. doi:10.1155/2010/578784 Available online at http://www.hindawi.com/journals/ijcgt/2010/578784/cta/