With Laura Goudet. Contribution at the conference Replaying Japan University of Alberta, 11 / 13 August 2021.
Rather than focusing on emotions as a symptom of an AI becoming human or as a human response to an AI character, this paper aims to pick up the interrogations on AI and humanness via the notion of failure, in its two philosophical meanings : as an omission which refers to an agent not doing something which was expected (Clarke, 2014); and as the opposite of success which, in regard with the human condition, threatens the self-wholeness (Desmond, 1988).
Using a composite corpus, made of action games, RPGs, fighting games, etc., we focus on the notion of failure and equilibrium between AI and human interactions.
We propose three readings of AI failures. Firstly, in a technocentric approach, we study the malfunctioning of AI programs used both in the video game structure and as agents which test the game to understand how the machine and the human player engage with each other, through glitches and abnormal uses of these games. Secondly, in a socio-cultural approach, we study failures of AI characters as either involuntary proof of humanness or voluntary strategies to be understood as human. Thirdly, in a phenomenological approach aiming to move aside to the anthropocentric views imposed on AI, we question AI failures to function as a human as part of a quest for their own identity.
A posthuman encounter between a human and a non-human agent has meaning and value. It is creative because there is an equilibrium between failure and success on both sides, since humans and AIs have their own agendas which can combine harmoniously through their forces and shortcomings.