An Interactive Space That Learns to Influence Human Behavior

Abstract
A key question in the design of intelligent environments is how a space can influence the actions of its users, and how such behavior can be learned. We present the results of experiments conducted as part of the Ada project, an interactive entertainment exhibit deployed at the Swiss national exhibition Expo.02. We used a learning model called distributed adaptive control (DAC) that is based on the animal learning paradigms of classical and operant conditioning. DAC has been developed using mobile robots in foraging tasks. Here, it was applied to the learning of effective cues for guiding visitors in a given direction. Our results show that, by using this learning mechanism, Ada was able to influence the behavior of visitors by learning to deploy particular types of cues. Many visitors could be induced to move toward a region of the space that they normally avoided visiting-an effect that can be seen as a spatial classification of visitors into interactive and noninteractive categories. In our analysis, we also introduce a measure of human activity that combines different types of data to capture key aspects of human behavior in interactive spaces.