Building a Decision Making Model for Communication Game by using Ontology

Abstract

This research proposes a method to incorporate ontology into game AI. We constructed human-like AI agents that play " INSIDER" by using ontology. "INSIDER" is a communication game. Each player selects one role as one Master, one Insider, or the Common. Master choses a topic from all and the topic is hidden from Common. This game consists of two parts with Cooperation part and Confrontation part. Cooperation part is to guess the hidden topic and Confrontation part is to guess who Insider is. We show two simulations: one is that AI agents don’t ask many questions about the bottom class of the ontology, and one is that they ask random questions from all classes. AI agents exclude unwanted topic candidates based on the parent-child relationship of the ontology by using the information they get. We observed the ontology’s change as AI agents get an information.