Item

Abstract

This paper is based on a presentation given in December 2019 at the Lund’s University conference "Artificial Intelligence (AI), Data Protection and Intellectual Property in a European context". The purpose of this article is to analyse the suitability of the copyright system over the so-called "core components" of Machine Learning (ML) systems, the ML models. Computer programs protection has always caused certain difficulties for intellectual property law. Internationally is was agreed in the 1970s to protect computer programs as literary works of copyright. ML models have been called "learning algorithms", "AI computer programs", "super-software". Yet, unanimity lacks as to technically agree on what they are. This is relevant from a copyright perspective, because depending on whether the ML model qualifies as a computer program, as a mathematical method or as another type of work, the regime of protection granted by copyright will be different. Additionally, all proprietary and open-source software licensing relies on the copyright protection. In most open licenses, if the license is applied to something that is not protected by copyright (or related rights) the license is not triggered. Thus, it seems relevant to question whether EU copyright law provides adequate protection for the core components of machine learning systems, the ML models. This paper begins with an overview of justifications for copyright protection of computer programs. Understanding how this came about is important to consider protection insufficiencies and how this can later be applied to ML models. It follows a technical overview of the differences between AI software, ML models and algorithms, as to delineate and frame the type of work that a ML model would be. This part also focuses on the differences between traditional (non-AI) software and ML process design and development, and the role that ML models play therein. The third part explains what the legal framework for copyright protection of computer programs is and tests its applicability to ML models. The paper is finalized with a set of conclusions and thoughts for future reflection.