Beyond determinism: Legal responsibility in the age of probabilistic AI systems

Abstract

This paper examines the fundamental distinction between classical deterministic programming and contemporary artificial intelligence systems, with particular emphasis on the legal consequences arising from the non-deterministic behavior of AI models. By analyzing the basic postulates of programming, the Softmax function as a key mechanism for output generation, and stochastic factors in training and generative processes, the paper demonstrates that artificial intelligence deviates from expected norms of predictability-an aspect that has far-reaching implications for the law of obligations, consumer protection, and tort liability regimes. In addition to a theoretical overview, the paper provides a legal analysis of the European Union's AI Act in the context of non-determinism, highlighting regulatory efforts to address this characteristic of AI models through a risk-classification system as well as technical and organizational requirements. The paper aims to contribute to an interdisciplinary understanding of AI-related challenges by positioning legal scholarship as a key factor in developing a new model of liability in the digital age.