Inteligência artificial (IA): considerações sobre personalidade, imputação e responsabilidade

Abstract

In today’s social and economic interactions, a new category of being is noticed with increasing frequency: the artificial agent. It is responsible for buying and selling goods online, for determining eligibility for benefits such as obtaining credit or medical assistance, for trading on the stock market, among numerous different everyday applications. However, the vast majority of these artificial agents do not yet have an essential feature: the autonomy to act independently from the direct inputs of human persons. They are, at most, semiautonomous, but research in the field of artificial intelligence indicates that this will not be the case in the near future. To the extent that artificial agents become able to act in a truly autonomous way, that is, regardless of direct human orientations, and based on information the agents themselves acquire and analyze, it will be possible for them to perform actions not anticipated by human beings. As a result, questions arise as to how the legal system should treat these autonomous artificial intelligence agents, especially in regards to the attribution of liability in the occurrence of damages, and the possibility of changing their legal status to that of persons, as opposed to mere objects. Thus, this research focuses on the following questions: the possibility of personifying autonomous artificial agents and the applicable liability regime for cases in which such personification does not occur. The analysis of the personification of artificial agents is carried out, therefore, within the context of the existence of a liability gap, and thus aiming to solve it in cases where the occurrence of damages caused by the agent cannot be directly attributable to a human person. In order to do so, initially, the main characteristics capable of defining an artificial intelligence as an autonomous agent will be presented, as well as the central problems caused by its autonomy and the possibility of independent action regarding its programmers and users, especially in the field of law. Then, the main reasons that inform the personification of non-human entities, particularly legal entities such as corporations, will be investigated in order to verify whether these reasons are also present in autonomous artificial agents. The main obstacles currently facing the personification of autonomous agents of artificial intelligence will also be presented. Finally, the different liability regimes that could be applied to them will be evaluated under two different contexts, namely, damages caused by artificial intelligence that can be traced back to the performance of its programmers, to the design of the machine, to the data used to train it, or to the behavior of its users; and damages not directly attributable to human individuals, raising the ultimate question of who should bear the costs.