Yet, these approaches rely on large datasets from which they process isolated snippets of spoken language use, and this is what they enable – the processing of single spoken inputs, or the generation of answers to individual requests. This is seemingly in contrast to the considerable progress in recent years with the advent of powerful machine learning techniques that have leaped language processing to a new level. Even the voice-driven applications people use in everyday life – in-car navigation, smart speakers at home, or personal assistants on smartphones – merely enable only task-specific “dialogs” that consist of spoken instructions or questions by the user and direct responses by the system that often do not meet user expectations ( Luger and Sellen, 2016). So far, no system is able to lead a half-decent coherent and engaging conversation with a human user. Still, today, conversational agents have not fulfilled the common expectation that, by trying to advance and combine abilities for verbal and nonverbal communication, the interaction with computers and technology will be facilitated and rendered more intuitive. The history was marked by ups and downs, and there have been numerous coordinated research endeavors in particular in the last two decades directed to the realization of so-called “conversational agents” ( Cassell et al., 2000) or “conversational artificial intelligence (AI)” ( Ram et al., 2017). We contrast those with current conceptualizations of human-agent interaction and formulate suggestions for the development of future systems.īuilding computer systems that are able to converse autonomously and coherently with a human is a long-standing goal of Artificial Intelligence and Human-Computer Interaction (going back, at least, to the seminal Eliza system presented by Weizenbaum, 1966).
We base our argument on a vast body of work on human-human communication and its psychological processes that we reason to be relevant and necessary to take into account when modeling human-agent communication. In this paper, we argue that we must start to re-consider the hallmarks of cooperative communication and the core capabilities that we have developed for it, and which conversational agents need to be equipped with: incremental joint co-construction and mentalizing. Yet, despite frequently made claims of “super-human performance” in, e.g., speech recognition or image processing, so far, no system is able to lead a half-decent coherent conversation with a human. The study of human-human communication and the development of computational models for human-agent communication have diverged significantly throughout the last decade. 2Department of Social Psychology, Media and Communication, University of Duisburg-Essen, Duisburg, Germany.1Social Cognitive Systems Group, Faculty of Technology, Bielefeld University, Bielefeld, Germany.
0 kommentar(er)
