To Be Or Not To Be: The Quest For Liability Of Artificial Intelligence

1. Introduction

Legal systems have historically been built upon a human-centric conception of liability grounded in fault, causation, and intent. By contrast, artificial intelligence models operating at varying levels of autonomy—from decision-support systems categorized as soft AI, to hard AI approaches capable of optimizing their own objectives without human intervention, and to generative models that can independently produce content—have complicated the question of whether harmful outcomes should be attributed solely to human conduct or to the independent functioning of these algorithmic structures. These differing categories introduce new debates, both technical and legal, regarding the allocation and determination of liability.