Talking with a chatbot: Simulated understanding of human-chatbot communication?

Abstract

<p>The main aim of this chapter was to find out what kind of communicative activity takes place between human participants and an AI-based system. This was studied with an experimental data in which human users discussed with entertainment chatbots. The interactions were based on the kind of turn-taking that favoured short sequences in which the topics were mostly everyday topics, but also the ones that mixed virtual reality with the real world. In these, the successful sequences turned out to be simple and ritualised or playful. However, a big part of the sequences turned out to be sporadic in which understanding was fragile, because the topical sequences were not pursued very long by the chatbots, the playfulness or jokes failed because they were off topic. Chatbots were unable to answer to questions on more specific domains, such as politics. One specific characteristic was mock or non-sense turns produced by chatbots. In these cases, the user resorted to metacomments that tried to explain the meaning or the user tried to follow the exceptional sequentiality by constructing the meaning for themselves.<br></p>