Cognitive interaction with virtual assistants: From philosophical foundations to illustrative examples in aeronautics

Why do we perceive virtual assistants as something radically new? Our hypothesis is that today virtual assistants are raising an expectation for natural interaction with the human. Human interaction is by nature cognitive and collaborative. Human sciences help to flesh the ingredients of such cognitive interaction. Uttering a sentence is at the same time: producing sound; making a well-formed sentence; giving meaning; enriching a common background of beliefs and intentions; making something together. In this paper, we remind the basics of human cognitive communication as developed by human sciences, particularly philosophy of mind. We propose a definition of this way of interacting with computer as 'cognitive interaction', and we summarize the main characteristics of this interaction mode into a layered model. Finally we develop case studies to illustrate concretely the concepts. We analyze in light of our theoretical model three approaches of conversational systems in Al, to illustrate the different available options to implement the pragmatic dimension of cognitive interaction. We analyze first the seminal approach of Grosz and Sidner [20], and then we describe how the now classical approach of discourse structure developed by Asher and Lascarides [5] could capture the pragmatic dimension of interaction with an intelligent virtual assistant. Finally, we wonder whether a state-of-the-art chat bot framework actually implements the needed level of cognitive interaction. The contribution of this paper is: to remind and summarize essential ideas from other disciplines which are relevant to understand what should be the interaction with virtual assistants should be; to explain why the cooperation with virtual assistants is something special; to delineate the challenges we have to solve if we are to develop truly collaborative virtual assistants. (C) 2019 Elsevier B.V. All rights reserved.