Machine Learning en Interaction: Outil, Matériau, Culture
Baptiste Caramiaux · HAL (Le Centre pour la Communication Scientifique Directe) · 2023
This habilitation deals with the study of machine learning in situated interaction, where I propose and articulate three interdependent analytical points of view: tool, material and culture. These three perspectives define the structure of the manuscript.In the first part, I consider machine learning as a tool. I propose to begin with theoretical work on the use of tools, from cognitive science to human-computer interaction. This work emphasises that the use of a tool involves situated actions. I argue that this also applies to machine learning, which can only be considered a tool when it is situated. I illustrate this with two sets of my previous work: machine learning as a research tool in human perception studies and machine learning as a creative tool in musical performance. Finally, I show the limits of this perspective and call for a broader vision of machine learning in interaction.In the second part, I propose a broader vision in which the materiality of machine learning is analysed. I show the theoretical elements that characterise digital materiality, a notion that comes from Design research and the HCI community. However, appropriating the material expression of machine learning is no easy task. I will illustrate this with two series of previous works. The first takes up the research work of the creative community and explicitly highlights the elements that make up the materiality of this technology. The second presents work in pedagogy where the materiality of machine learning is highlighted to better convey its underlying concepts to novices. I discuss the fact that machine learning as a material requires tools and pedagogy.In the third part, interactions with machine learning are considered from a cultural and political point of view. Thanks to ambiguous terminology, the technology has been communicated using a normative vision. I explain and highlight how this discourse can be deconstructed to imagine more exciting and richer narratives related to interactions with machine learning. I illustrate this with two previous works. First, I show how visual artists make explicit this dominant culture in machine learning and artificial intelligence and discuss its limitations. Second, I present a performance art piece to which I contributed that questions the role of artificial intelligence in contemporary algorithmic societies and how this role elicits violent behaviour.In conclusion, I propose to highlight how this work and reasoning define rich and interdisciplinary research directions that will shape my future research agenda.