Rémi Barraquand - Designing Sociable Technologies

This thesis investigates the design of sociable technologies and is divided into three parts that are described below.

In the first part we introduce sociable technologies. Technologies are defined as anything created as an extension of techniques by intelligent means. From this definition we compare technologies according to the motivation underlying their design e.g. improvement of control (technological tools), improvement of cooperation (human communication, sociable technologies). In order to better understand the role of designers in technological innovation, technological evolution and changes are regarded from an evolutionary perspective : designers are initiators of new variants that undergo selection in society. Considering the fundamentally social and cooperative structure of human society and culture we argue that evolution of technologies is branching off toward a new type of technologies : sociable technologies. Sociable technologies are defined as an extension of techniques to improve social cohesion, social interaction and cooperation. From this perspective, the emergence of human language and the emergence of sociable technologies share a common motivation : the improvement of cooperation and social life. Two design principles are then presented and led to the introduction of a new direction of research : acquiring social common sense. The acquisition of social common sense is presented as fundamental for the emergence of sociable technologies. This doctoral work then focuses its investigation on a key aspect of social common sense : the ability to behave appropriately in social situations. The concept of polite technologies is introduced and approaches to design polite technologies are addressed in this thesis.

In the second part we introduce premises for the design of sociable technologies. First we present a preliminary approach that suggests acquiring polite behavior by learning an association between model of social situations and behavior. Reinforcement learning is proposed as mean to learn such association during social interactions between users and computer systems. Three increments to the standard Q-Learning algorithm are presented and evaluated into a set of experiments conducted in a smart-environment. The results obtained validate the approach but emphasize the limitation of technologies to reach a mutual understanding of social situations with humans. Without an ability to reach a mutual understanding of social situations the interaction of humans and technologies are doomed to remain autistic. The code-model of communication (Shannon) and used by technologies is presented as an obstacle toward reaching this mutual understanding. Based on recent research in the field of evolutionary anthropology, which studies evolution of human communication, human social learning and human culture, we argue that the ostensive/inferential model of communication, proposed by Sperber & Wilson to explain human communication, is more adapted to support human-machine interaction. This hypothesis is evaluated in a study conducted in a smart-environment and the outcomes of this study validate the need for a psychological infrastructure of shared intentionality for sociable technologies. The premises for the design of such infrastructure are then enunciated.

The final part of the thesis concerns the design of an infrastructure for the design of sociable technologies. This infrastructure is composed of three components : an inferential model of context, a digital intuition and a cooperative machine learning theory. First, a meta-model and an architecture are presented to support the inferential model of context. Ostensive interfaces are presented as a new form of user interfaces to support the ostensive part of the ostensive/inferential model of human communication. The architecture and ostensive interfaces are illustrated and evaluated in an experiment conducted in a smart-environment. Second, we provide the support for a digital intuition for technologies and introduce the concept of eigensituations. After introducing the theory and algorithms for cooperative machine learning theory we demonstrate the advantage of eigensituations to learn polite behavior from social interaction in an experiment conducted in a smart-environment. Results demonstrate the benefit of the whole infrastructure, namely the combination of an inferential model of context, a digital intuition and a cooperative machine learning theory.