Robots and AI as Legal Subjects? Disentangling the Ontological and Functional Perspective.

Robotics and AI-based applications (RAI) are often said to be so technologically advanced that they should be held responsible for their actions, instead of the human who designs or operates them. The paper aims to prove that this thesis ("the exceptionalist claim")-as it stands-is both theoretically incorrect and practically inadequate. Indeed, the paper argues that such claim is based on a series of misunderstanding over the very notion and functions of "legal responsibility", which it then seeks to clarify by developing and interdisciplinary conceptual taxonomy. In doing so, it aims to set the premises for a more constructive debate over the feasibility of granting legal standing to robotic application. After a short Introduction setting the stage of the debate, the paper addresses the ontological claim, distinguishing the philosophical from the legal debate on the notion of i) subjectivity and ii) agency, with their respective implications. The analysis allows us to conclude that the attribution of legal subjectivity and agency are purely fictional and technical solutions to facilitate legal interactions, and is not dependent upon the intrinsic nature of the RAI. A similar structure is maintained with respect to the notion of responsibility, addressed first in a philosophical and then legal perspective, to demonstrate how the latter is often utilized to both pursue ex ante deterrence and ex post compensation. The focus on the second objective allows us to bridge the analysis towards functional (law and economics based) considerations, to discuss how even the attribution of legal personhood may be conceived as an attempt to simplify certain legal interactions and relations. Within such a framework, the discussion whether to attribute legal subjectivity to the machine needs to be kept entirely within the legal domain, and grounded on technical (legal) considerations, to be argued on a functional, bottom-up analysis of specific classes of RAI. That does not entail the attribution of animacy or the ascription of a moral status to the entity itself.