Artificial intelligence (AI) has been the subject of a great deal of academic interest in recent times; it has captured the attention of copyright lawyers fascinated by the thought of machines creating works of art, music and literature. There is no doubt that, as has often happened in the past during previous waves of technological advances, AI platforms—and especially, machine learning—have brought with them new opportunities as well as challenges. Machine learning is an AI application enabling programs to learn and progress automatically from experience. Its main feature is accessing data and often using it for the purpose of creating outputs, including music, literature, movies and art. Amounts of data are observed and analysed by the machine, which enables the latter to learn and then make creative decisions leading to final outputs that, as precise works of art, are often not foreseeable by the people who developed and started the initial program. Such a process is characterised by the absence of substantial human intervention or assistance after the program is operated, and by the use of algorithms—namely a sequence of instructions aimed at solving a problem or performing a computation. It can be deemed “algorithmic creativity”, or the way by which AI/machines create new works.Against this background, the article delves into thorny legal and policy issues. Is an AI-created work protected by copyright? Should it be? Who would be viewed as the author? Who should own such a work? These questions are the focus of the first part of this article: “AI as producer”. In the second part, “AI as consumer”, our analysis shifts to the questions of whether and to what extent the use of data fed into the system, for example to train the algorithms, may amount to copyright infringement, or may in certain circumstances be exempted under fair use, fair dealing or similar doctrines. In the third part we consider possible legal regimes for dealing with machine produced works including alternatives to copyright, such as a public domain scenario and a sui generis system. Finally, we provide our conclusions
