Negotiating with Deceptive Agents in Mixed-Motive Interaction Environments using Theory of Mind

Abstract

Deception is the act of spreading a false belief through communication or actions, which is commonly used among people in situations where motives differ. Being able to perform such act is because of an ability called Theory of Mind. This is the ability to reason about the mental content of others, which often happens naturally in people. Higher orders Theory of Mind enable reasoning about how others use this ability. Such ability is used by people to get an idea of what someone's intentions are when that person performs a certain action. Using this principle a prediction of future actions can be made, which is a key aspect of deception. In competitive negotiation settings, it is expected that higher orders of Theory of Mind enable the agent to look further into the negotiation process that in turn can lead to a higher payoff through deception. This thesis explores the conditions needed for deceptive behaviour to emerge and which order Theory of Mind enables deceptive behaviour in computational agents using a mixed-motive interaction environment. An environment called Colored Trails is built to achieve this. The environment enables the agents to negotiate with each other using various proposals. Here, the agents have partial information about each others preferences which allows for the investigation of deceptive behaviour for different orders of Theory of Mind.