Evaluating the Improvements of StarCraft Gameplay in the ABL Agent EISBot by Implementing Dynamic Specificities through an External Component

Abstract

I studied how artificial intelligence (AI) adapts to failure. I worked with AI framework ABL, which uses a behavior tree. The tree begins by creating behaviors to accomplish an AI goal. Behaviors are chosen based on priorities. However, these priorities are not effective in all situations, and cannot change in ABL, which prevents the AI to learn from failure. For my study, I gave ABL the ability to change its priorities. This essentially allows an ABL AI to adapt from its mistakes. This ability was then tested for effectiveness using the ABL AI EISBot (an AI that plays videogames) and the game of StarCraft. The new ability caused EISBot to improve significantly showing that the ability to change priorities improves an AI’s performance.