Machine learning e responsabilidade penal : reflexões a partir do risco permitido

Abstract

This research aims to investigate the contribution of permitted risk as a criterion for removing the criminal liability of the programmer/producer for damage associated with machine learning systems. It seeks to verify, within the framework of imputation proposed by Claus Roxin, whether an unexpected result produced by the machine/software can be attributed to the conduct of the agent responsible for producing or programming the intelligent product. To this end, in a first chapter, we introduce the concepts of artificial intelligence and machine learning, relating them to the challenges imposed on criminal law. In a second chapter, we analyze the contribution of the permitted risk in the process of criminal imputation for disvaluable results arising from machine learning and the possibility of adopting the parameters already developed for criminal product liability. In the third and final chapter, we confront the proposed theory with the resolution of cases inspired by real situations, verifying whether the criminal liability of the programmer/producer will be configured (or not). The approach used was the hypothetical-deductive method, with emphasis on a transdisciplinary dialog