On Deception Detection in Multi-Agent Systems and Deception Intent

摘要

Deception detection plays an important role in the military decision-making process, but detecting deception is a challenging task. The deception planning process involves a number of human factors. It is intent-driven where intentions are usually hidden or not easily observable. As a result, in order to detect deception, any adversary model must have the capability to capture the adversary's intent. This paper discusses deception detection in multi-agent systems and in adversary modeling. We examined psychological and cognitive science research on deception and implemented various theories of deception within our approach. First, in multi-agent expert systems, one detection method uses correlations between agents to predict reasonable opinions/responses of other agents (Santos & Johnson, 2004). We further explore this idea and present studies that show the impact of different factors on detection success rate. Second, from adversary modeling, our detection method focuses on inferring adversary intent. By combining deception "branches" with intent inference models, we can estimate an adversary's deceptive activities and at the same time enhance intent inference. Two major kinds of deceptions are developed in this approach in different fashions. Simulative deception attempts to find inconsistency in observables, while dissimulative deception emphasizes the inference of enemy intentions.