Simulated Mining Game

Background

In this project, we consider the problem of computationally identifying behavior tendencies from a player’s game decision-making data. “Behavior tendencies” refers to a player’s consideration of the relevance of attributes that helps them make a specific decision. The theory we used is called Theory of Mind, which is used for humans to identify behavior tendencies by observing someone’s gameplay, which is often attributed to successful collaboration in teams and other environments.

The Game

Approach

We separated the decision making process into two decisions, the first decision for staying or moving and the second decision for which direction to go next. We modeled the first decision using logistic functions while we modeled the second decision using a softmax function.

We embedded the cognitive factors into these two decision-making models. Then we tested our models by implementing AI bots with different setting of cognitive variables. We generated gameplay logs using these bots with different cognitive factor settings.

We use Markov-Chain-Monte-Carlo simulation for inverse inference. Based on the log files generated by bots, we use inverse inference to infer the cognitive variables in the decision models. The result shows that we can successfully infer these variables and differentiate bots based on their different behaviors.

Teams