We all know that artificial intelligence (AI) has limitless potential, that goes beyond the power of human imagination. But what if AI starts creating pieces of art?

This does not seem that new, since we’ve already got used to technologies making creative stuff. Today, 3D printers produce artifacts that in the past we used to do with our own hands. Recent applications of AI to visual art adds another piece (even broader) to this mosaic. Indeed, analytics and reuse of large amounts of data relating to old creative works, digitally copied and reproduced by way of complex systems of AI, can transform machines into creators of brand new artworks. How is this possible?

The Edmond de Belamy’s case – or AI’s love for portraits

Let’s begin from a very recent fact. In late October 2018, the portrait of Edmond de Belamy, an unknown gentleman with a black suit, has been sold for 432,500 US dollars. This may not seem an attention-grabbing piece of news, since this work looks basically like a standard portrait from the 18th or 19th century. Moreover, the clearing price is not that high, considering that the highest price ever reached at an auction is 450 million US dollars, paid by a Saudi prince for Leonardo’s Salvator Mundi.

Actually, the sale of the portrait of Edmond de Belamy has been sensational not for its own features, but for its author’s. Indeed, such work has been created by a system of AI called GANs (Generative Adversarial Networks) – journalistically known as an algorithm. The French collective of artists Obvious developed such technology, that was “fed” with data of about 15,000 portraits from the 15th to the 19th Century and, as a result, made the auction-breaking work.

Algorithmic authors

Leaving aside ethical and philosophical questions on the role of algorithmic artists, several legal challenges clearly arise here. Can a creative work made by an artificial intelligence system be protected under the current copyright law?

As is widely known, creative works must be original to get copyright protection. At first glance, such requirement prevents an algorithmic work from access to copyright protection, since originality is eminently deemed a human feature. And yet, such requirement is fundamental for both the main legal systems of the Western legal tradition – i.e. common law and civil law.

  • Member States of the EU unanimously link originality to individuals. For example, in Italy there is no clear legal definition of an author; however, several provisions of the Italian Copyright Law suggest that being author is an exclusively human trait. Under Section 6 of Italian Copyright Law, “the original entitlement of the acquisition of copyright consists of the creation of the work, as a particular expression of intellectual work“. Then, under Section 8 of the Italian Copyright Law, “it is considered the author of the creative work, unless proven otherwise, who is indicated therein as such in the forms of use, that is, it is considered as such in the performance, execution, representation or radio-diffusion of the work itself”.
  • At EU level, the originality standard has been subject to limited harmonization, since EU directives generally provide that softwares, databanks and photographs enjoy copyright protection only where they are their “author’s own intellectual creation”.
  • Also, recitals 9 and 11 of the InfoSoc Directive undoubtedly sets out that the standard of originality is related purely to a “human intellectual creation” and states that copyright protection enables the flourishing of human culture, since “helps to ensure the maintenance and development of creativity in the interests of authors, performers, producers, consumers, culture, industry and the public at large“.
  • Similar conclusions are relevant for the US scenario. Indeed, according to the landmark Feist decision of the US Supreme Court, copyright law only protects “the fruits of intellectual labor” which “are founded in the creative powers of the mind”.

Algorithmic artworks = unprotected works?

The abovementioned legal approaches exclude machines from the notion of authorship. However, this does not prevent someone from enjoying copyright protection. Indeed, in several countries from all over the world the threshold to get copyright protection is somewhat low. For example, in the Netherlands an individual is given copyright protection as long as human choices are involved (for example, as in the Edmond de Belamy’s case, embedding data and information on old paintings in the software). As a consequence, the person designing the software would enjoy copyright protection for both the software itself and the artwork.

Low thresholds to get copyright protection for algorithmic artworks raises the question of the bloat of IP protection, in a world without scarcity. Indeed, the software designer may enjoy different IP rights on basically the same assets – copyright protection for both the computer program and the artwork, and perhaps the database sui generis right or trade secrets protection for structured information resulting from the analysis of digital versions of old paintings.

In conclusion, the debate on the relationship between copyright law and AI creative works is still at the beginning. In the future, broader legal and ethical questions will clearly arise in relation to these issues. In the meantime, we’ve started believing that Lucio Dalla’s words on creativity can be extended to machines: “creativity is one amongst many bells that rings as out of place in a well-oiled machine“.