Do we really want AI answering on our behalf? A study of smart replies usage

Abstract

This research focuses on the acceptability of smart replies in emails and text messages. Our goal is to find criteria influencing the use of smart replies in AI-mediated communications and define the acceptability of such practices from the sender's perception. We conducted qualitative and quantitative research using surveys and interviews with a population of French native speakers. During our experiment, we shared various communication scenarios (including both professional and personal contexts) with fifty college students. We offered them the choice between smart replies or their own handwritten reply. We then followed up with interviews with a subset of students to better understand their replies to the survey and their relationship to AI-mediated communication. Analysis of the collected data points toward a broader acceptance of smart replies when the author only intends to acknowledge that the message was received and understood. On the contrary, reply suggestions are often dismissed by the sender as too casual for professional communications and too formal for family or friends.