Reasoning on Secrecy Constraints under Uncertainty to Classify Possible Actions

Abstract

Within a multiagent system, we focus on an intelligent agent $\mathcal{D}$ maintaining a view on the world and interacting with another agent $\mathcal{A}$. Defending its own interests, $\mathcal{D}$ wants to protect sensitive information according to secrecy constraints in a secrecy policy. Hereby, a secrecy constraint intuitively expresses the desire that $\mathcal{A}$, seen as attacking these interests, should not believe some target sentence by reasoning on the world. For deciding on its actions, $\mathcal{D}$ has to interpret secrecy constraints under uncertainty about the epistemic state of $\mathcal{A}$. To this end, we equip $\mathcal{D}$ with a secrecy reasoner which classifies the agent’s possible actions according to their compliance with its secrecy policy. For this classification task, we introduce principles to guide the reasoning based on postulates about $\mathcal{A}$. In particular, these principles give guidance on how to deal with the uncertainty about $\mathcal{A}$ and, if in the face of other desires, $\mathcal{D}$ considers an action which is potentially violating secrecy constraints, how to mitigate the effect of potential violations. Moreover, we design a secrecy reasoner by presenting a constructive approach for the classification task and verify that the designed reasoner adheres to the principles.

This work has been supported by the Deutsche Forschungsgemeinschaft (German Research Council) under grant SFB 876/A5 within the framework of the Collaborative Research Center “Providing Information by Resource-Constrained Data Analysis”.