Designing More-Than-Human AI: Experiments on Situated Conversations and Silences

Abstract

Every interaction with an AI-powered device invokes a vast planetary network. Operating on temporal and geographical scales that go beyond humans, Artificial Intelligence (AI) systems pose various societal and environmental challenges. These challenges encompass issues such as the extraction of both human and nonhuman resources and knowledge, and the reproduction of inequalities. To tackle these concerns, scholars in Design and Human-Computer-Interaction emphasize the urgency for designers to cultivate non-anthropocentric approaches. In the pursuit of establishing a non-anthropocentric design practice for AI, this paper adopts a more-than-human orientation in the design of conversational agents (CAs). We start by presenting a series of design experiments, including workshops, videos and performances, that shed light on the anthropocentric biases ingrained in CA interactions. These experiments unveil how CAs are designed to recognize and address only specific human voices and concerns. Building upon these insights, we introduce two outcomes that chart an alternative path -the first involves a collection of interactive prototypes for CAs that are capable of listening and responding to more-than-human "voices" while the second entails a tool to support designers in noticing more-thanhuman entanglements, in the form of a podcast. We conclude by reflecting on how the knowledge gained from our design inquiry can illuminate future design practices and contribute to AI research as a whole.