Who can you trust: Dealing with Deception

Michael Schillo, Petra Funk, Michael Rovatsos

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Trust is one of the most important concepts guiding decision making and contracting in human societies. In artificial societies, this concept has until recently been neglected. The inherent benevolence assumption implemented in many multi-agent system can have hazardous consequences when dealing with deceit in open systems. Therefore we present a formalization of trust, an algorithm for evaluating it and an application scenario. This approach can be applied to avoid the use of trusted third parties. it includes social behaviour for implementing punishment without empowered authorities. We argue that by adopting our ideas in open systems, agents can autonomously deal with deception and identify trustworthy parties. Our approach to this problem is twofold: Agents can observe the behaviour of others and thus collect information for establishing a trust model. Still, agents cannot rely exclusively on the information gathered. In order to adapt quickly to a new or rapidly changing environment and to ensure fast responses to new opportunities, we enable agents to make use of observations from other agents (witnesses).

However, testimonial evidence from such witnesses may be brittle, as witnesses may have differing motives and may try to deceive other agents about their true observations. Our solution is based on stochastic process. For its evaluation,we chose a game theoretic abstraction of contracting between agents. The setting consists of a number of phases, which we consider also natural for computing. The interaction of agents improves if they use communication about observations and apply the trust model to evaluating new contract partners. We demonstrate practical relevance of our ideas with a direct mapping between our scenario and electronic commerce.
Original languageEnglish
Title of host publication Proceedings of the Workshop "Deception, Fraud and Trust" of the Autonomous Agents Conference
Number of pages12
Publication statusPublished - 1999

Fingerprint Dive into the research topics of 'Who can you trust: Dealing with Deception'. Together they form a unique fingerprint.

Cite this