To Understand Sentience in AI, First Understand it in Animals. Commentary to Jonathan Birch and Kristin Andrews.

Birch and Andrews' article examines the problem of sentience in AI systems using a method based on the identification of specific markers. This approach seeks to overcome two major obstacles: the “gaming problem” and the “N=1 problem”, while highlighting the limitations of studies based on the linguistic productions of these systems, which would only be superficial reproductions of human linguistic behaviour. However, the phenomenon of ‘hallucination’, combined with the inability of these systems to truly grasp the meaning of the words and structures they generate, seems to us to rule out any possibility of attributing any form of sentience to them. Such a question therefore seems more likely to be an anthropomorphic projection of characteristics specific to living beings.