Unaware of Reality: Inconsistent Grounding in Conversational AI

Anonymous

Abstract: Conversational AI is one of the most promising applications of NLP research. It will be a factor in the success of technologies designed to improve our lives through human-machine interaction. However, current conversational AI methods based on neural networks are often unreliable. This short paper discusses two different ways of interpreting the (in)consistency of conversational agents' responses, which we call horizontal and vertical consistency. We frame their limits with respect to grounding and present a broader outlook on the general problem of conversational agent design.