The Trouble with Chatbots: social skills in a social world
This paper argues that social intelligence is a critical component of any conversational agent. Conversational interfaces with a sense of social identity circumvent several of problems that commonly arise, and such agents can promote ease-of-use, encourage engagement, and naturally set limits. The paper explores these issues by retrospectively analysing two sets of wizard-of-Oz experiments in the light of the social intelligence concept. In these experiments, the wizards employed simple social strategies to deal with ‘difficult’ situations, and these are examined and expanded to general principles. Embracing the idea that a conversational agent is a social actor will, we believe, result in interfaces that participate seamlessly in our human world of social relations.