Baptiste Caramiaux, researcher CNRS at ISIR, defended her Habilitation to Supervise Research (HDR) on Monday, March 15, 2023 at 9:30 am, on the Pierre and Marie Curie campus of Sorbonne University.
Title of the work: Machine Learning in Interaction: Tool, Material, Culture.
The jury was composed as follows:
- Susanne Bødker, Professeure, Aarhus University (Reviewer)
- Elisa Giaccardi, Professeure, Delft University of Technology (Reviewer)
- Myriam Lewkowicz, Professeure, Université de Troyes (Reviewer)
- Alex Taylor, Reader, City University of London (Examiner)
- Rebecca Fiebrink, Professeure, University of the Arts London (Examiner)
- Jean-Daniel Fekete, Directeur de Recherche, Inria (Examiner)
Summary of the habilitation
In this habilitation I emphasise the importance of a perspective on machine learning (and artificial intelligence) technologies situated in interactions with people and contexts of development and use. Through my research over the past ten years, in collaboration with my students, post-docs, and colleagues, I articulate three perspectives: machine learning as a tool, material, and culture. First, I show that machine learning can be seen as a tool acting on objects situated in their context. This view has several limitations, which are highlighted in creative applications. Secondly, I show that machine learning has materiality, useful for system design. This materiality makes the technology dynamic and expressive, highlighting its cultural dimension developed in the third part. As a culture, I show that this technology is often communicated in a normative way, reducing the possibilities of discourse and interaction. And I propose examples of alternative discourses on technology from artistic works. In conclusion, I discuss the interrelationships between these three perspectives and how they can be linked in a human-computer interaction research programme. Summary of the habilitation, by Baptiste Caramiaux.
Machine learning is a technology which, based on data, is able to make predictions for complex tasks such as recognition or generation. This technology is now ubiquitous in various fields of application and an important issue is to create interactions between these systems and users in communities of practice where these technologies are used. Baptiste Caramiaux’s research focuses on the design of these interactions, the study of their use, and the study of the socio-technical and socio-cultural impact they underlie. In particular, the user communities concerned are artists, health practitioners and the general public.
Design of interactive learning systems
Interactive learning systems involve users in the learning loop, either by editing the learning data or by editing the model. The design of such systems has a strong impact in the creative environment where users have a need for customisation for their expressive and aesthetic purposes. More generally, these systems provide a means for expert users to audit learning systems. The design of these systems is made possible thanks to an open source application design library, called Marcelle, which Baptiste Caramiaux has helped to develop and disseminate.
Study of the uses of systems incorporating machine learning
As part of his research, these systems are deployed in artistic and medical communities, but also to the general public. The study of the uses in each of these communities allows a better understanding of the impact of these technologies on the practices and in their socio-technical context. Firstly, the study of uses will make it possible to improve the interactions designed. Secondly, the study of uses allows us to highlight the cultural and social representations of users with regard to these systems and technology.
This research evolves in the fields of Human-Computer Interaction, Design research, and Science and Technology studies. It is an interdisciplinary research that emphasizes the incorporation of various actors (human and non-human) in the design and socio-technical integration of intelligent systems.
Contact: Baptiste Caramiaux, Researcher CNRS
