Investigation of the conditions for continuous information conveyance by two autonomous conversational agents

In recent years, information conveyance through conversation using agents such as robots and avatars has gained attention. Among them, conversation by two agents has been shown to encourage effective information conveyance. Previous studies have also demonstrated that incorporating subjective information, such as emotions, into conversations enhances this effect. Therefore, a medium for information conveyance involving two autonomous agents and including subjective information is expected to be effective.

In this study, such a medium was implemented, and the conditions necessary for it to convey information continuously were investigated. Objective information was defined as the content of existing news, and subjective information was defined as the preference toward the news. A frame structure was used for organizing objective information, and a network structure was used for subjective information. A method was developed to autonomously obtain both types of information. This knowledge was then distributed to two agents, who exchanged it and attempted to understand each other through conversation.

Experiments were conducted to determine whether the subjective information obtained autonomously by the agents was as natural and consistent as that of humans. Further experiments examined the conditions for enabling continuous information conveyance using the medium.

The results indicated that conveying important information first and using robots rather than text were effective strategies for maintaining continuous information conveyance.