Mixed Initiative Ambient Environments: A Self-Learning System to Support User Tasks in Interactive Environments

Abstract

In this paper we introduce a pro-active self-learning agent system that supports the user in an ambient intelligence environment. Interaction in an ambient intelligence environment is typically very complex and demands lots of attention of the end-user. In many cases it is extremely difficult for the user to evaluate what the consequences are from interactions accomplished in this space. In the system we present in this paper, an agent monitors the user's interactions with the environment to learn the user's expectations when certain activities are performed. Machine learning techniques allow the system to pro-actively react on behalf of the user. When the system is not entirely sure whether the user wants to execute a certain action, it will not execute it on his or her behalf, but makes it easier for the user to execute it himself by guiding the user through the options. This mixed initiative can help the user while not risking to lose his or her confidence by executing the wrong action on the user's behalf. The kind of system presented in this paper takes a lot of simple, often reoccurring work out of the user's hands.