Agent auto-generation system: Interact with your favorite things

Abstract

This paper proposes a framework for an Agent Auto-Generation System (AAGS) which gets information from sensor devices and improvises a conversational agent from arbitrary things. AAGS has agent-types to prepare a perception for an improvised agent according to the shape of the agentization targets. It also has virtual-input, which generates knowledge representation related to information extracted from the networked sensor devices. The virtual-input employs a viewpoint based on the agent-type to generate a knowledge expression, and gives it to the improvised agent. Experiments to evaluate AAGS revealed that it is necessary to consider the perception of the improvised agents based on the agent-types. In addition, the agent's viewpoint for the perception has an effect on how people recognize the improvised agent.