AI-Based Digital Assistants

Abstract

Artificial intelligence (AI) is becoming omnipresent; it permeates our work and private lives in many areas. A key area of application is AI-based digital assistants, which are now becoming available in large numbers and a wide variety of usage scenarios. Research into AI-based digital assistants has a long history, dating back to Joseph Weizenbaum’s well-known ELIZA in 1966. In parallel, global technology companies such as Microsoft, IBM, Google, and Amazon have been working intensively for decades on advancing AI-based digital assistants and have recently made them suitable for the mass market. Empowered by recent advances in AI, these assistants are becoming part of our daily lives. We are observing the ever-growing usage of various digital assistants, for instance, voice-based assistants such as Amazon Alexa, or text-based assistants (chatbots), such as those embedded in Facebook Messenger. It is foreseen that AI-based digital assistants will become a key element in the future of work. Today’s enterprise communication platforms such as Slack or Microsoft Teams already provide many different bot types to augment work, and Gartner (2019) predicts that by 2021, one-quarter of all digital workers will use a virtual employee assistant daily. AI-based digital assistants provide significant opportunities, but also might become a threat. On the one hand, they are expected to take over routine tasks from humans and to free up time and resources for more demanding tasks. For instance, IBM argues that chatbots can help to reduce customer service costs by 30% (Reddy 2017). On the other hand, a recently announced advanced AI-based digital assistant by Google named Duplex (Google AI Blog 2018) has led to a debate about potential misuses for deception and fraud, owing to its human likeness. More generally, while the pervasiveness of AI-based digital assistants increases, most people ignore their underlying architecture and algorithms (Frey and Osborne 2017), resulting in serious concerns and user aversion regarding their uses (Dietvorst et al. 2015, 2018). From a conceptual perspective, AI-based digital assistants – like every IS – can be understood from two different yet complementary perspectives (Fig. 1): first and broadly speaking, AI-based digital assistants represent a sociotechnical system that relies on the interplays of three key elements (Goodhue and Thompson 1995; Heinrich et al. 2011): the individual user, who seeks to achieve certain Prof. Dr. A. Maedche Dr. S. Morana Karlsruhe Institute of Technology, Karlsruhe, Germany