top of page
  • Laurens Kasteleijn

Artworks made by Computers: (im)possibility and (un)desirability of copyright protection of AI made

April 2021 - Ignace Nédée & Laurens Kasteleijn for Art Law Services

Key words: Computers’ ability to produce art, Artificial Intelligence, legal scope of artworks, potential copyright protection of AI produced artworks, authorship of AI produced artworks, EU copyright legislation, (un)desirability of copyright protection

1. Introduction

Artificial Intelligence (AI) is real. It is not only widely exploited in the business world; it also increasingly enthuses the art world. Can computers make art? Doesn’t art require emotional involvement or at least human intervention? Apparently not. Today, artists are able to create art via AI. After generating an algorithmic formula and putting in certain data, AI mechanisms are able to ‘create’ works, to move spectators and to complement the cultural environment.

A. Elgammel beautifully states the following (1):

If you look just at the form, and ignore the things that art is about, then the algorithm is just generating visual forms and following aesthetic principles extracted from existing art. But if you consider the whole process, then what you have is something more like conceptual art than traditional painting. There is a human in the loop, asking questions, and the machine is giving answers. That whole thing is the art, not just the picture that comes out at the end. You could say that at this point it is a collaboration between two artists — one human, one a machine. And that leads me to think about the future in which AI will become a new medium for art”.

Whether art can (not) or should (not) be algorithmically created by machines is open for a philosophical or ethical discussion. Although this topic is indeed fascinating, this article merely focusses on the legal aspect of AI generated works. After briefly presenting the scope of ‘Artificial Intelligence’ and EU copyright, this article analyzes whether AI produced works can be copyright protected and whether this protection is in fact desirable.

2. Scope of copyright and AI

a. What is Artificial Intelligence?

Artificial Intelligence (AI) is a broad concept that is subject to multiple interpretations. In short, the term refers to computers that are not only programmed to execute certain commands, but that also are able to learn, observe, evaluate and adapt themselves. It refers to “the capability of a machine to imitate intelligent human behavior” (2). It is all about processing and analyzing data. Through using pre-programmed algorithms and parameters, AI is capable to make independent decisions based on the data they process. This is called ‘machine learning’ (3), which A. Guadamuz describes as (4): “a subset of artificial intelligence that studies autonomous systems that are capable of learning without being specifically programmed. The computer programme has a built-in algorithm that allows it to learn from data input, evolving and making future decisions in ways that can be either directed, or independent”.

Traditionally, a distinction is made between ‘weak’ and ‘strong’ Artificial Intelligence. ‘Weak AI’ refers to computers that are purely designed to perform specific tasks. Considering it can’t deviate from the programmed tasks, weak AI can’t make autonomous or independent decisions (5). Whereas the programmer conserves direct control during the process and can predict the result to a large extent, the actual ‘(creative) choices are still made by a natural person (6). This is an important aspect in terms of copyright.

Strong AI’, on the other hand, refers to computers that are able to make accurate decisions in an arbitrary, independent and autonomous way and that are able to monitor and reprogram themselves (7). In fact, the ‘machine learning’ capacity goes one step further: ‘deep learning’ (8). Strong AI can create new algorithms by itself, which allows them to make autonomous decisions. Unlike weak AI, the results of strong AI cannot be precisely predicted. Besides performing specific tasks, strong AI can analyze and adapt their behavior and is able to perform independent tasks without needing human intervention (9). Also this is an important aspect in terms of copyright. Contrary to weak AI, the programmer does not have direct control in the creative process.

Two famous examples are “Portrait of Edmond Belamy”, made by the Parisian collective Obvious, and A. Elgammal’s “Psychedelic”. These artworks have been created through using Artificial Intelligence and more specifically via ‘adversarial networks’. These are two-fold algorithms consisting of a ‘Generator’ and a ‘Discriminator’ (10).

Regarding the first work, Obvious has put the data of more than 10.000 portraits (painted between the 14th to the 20th century) into the AI system. While the Generator algorithmically creates a new image after having analyzed these data (11), the Discriminator attempts to distinguish the image being created by Generator from a human produced image (12). The objective of this method is “to fool the Discriminator into thinking that the new images are real-life portraits. Then we have a result (13). The result of this GAN method (‘Generative adversarial network’) is that it generates a different and non-preexisting work such as the Portrait of Edmond Belamy (14).

The algorithmic method used by Elgammal is called the CAN method (‘Creational Adversarial Network’). This is also a strong AI that operates with a two-fold algorithm – namely a Generator and a Discriminator – but with a specific focus to create something new. One the one hand, the AI is programmed to analyze and ‘learn’ the data of existing artworks, aesthetics, styles and artistic movements (15). On the other, it is programmed to algorithmically create and develop artworks that are ‘new’ and characteristically different vis-à-vis these existing works and artistic styles (16). More specifically, “it will be penalized if, when creating a work of its own, it too closely emulates an established style” (17).

In other words, there is no human assistance or direction in this process. The mechanisms form strong AI that independently analyze (thousands of) existing artworks, and generate new content on an arbitrary, autonomously and independent manner. Whereas the entire creation is done by the AI, the human intervention is limited to the selection of existing artworks (putting in as data) – at the beginning of the process – and the designation of one of several computer-generated works (‘creation by designation’) (18) at the end of the process. In comparison with weak AI, human intervention – let alone creative input – is a lot less significant and even negligible (19).

b. When can a work obtain copyright protection under EU law?

In theory, any person who makes an ‘original work’ within the EU automatically acquires a set of exclusive economic and moral rights from the moment of the creation. Whereas economic rights – such as the right of reproduction, distribution and communication to the public – are perfectly alienable and trade-able, moral rights – such as the right of disclosure, the paternity right and the right to allow or refuse modifications to his/her work – are strictly personal and inalienable. But what is an original work?

In fact, the scope of copyright protectable ‘works’ is extremely broad, because the minimum standard, namely the requirement for a work to be considered as ‘original’, is rather low. Under Article 2(1) of the Berne Convention, copyright protection must include "every production in the literary, scientific and artistic domain, whatever the mode or form of its expression" (20). No matter the form, color, volume or mass of a creation, it will be automatically copyright protected if the work forms an ‘original’ intellectual creation (21).

The EU Court of Justice has formulated three important elements to clarify this ‘originality’ requirement.

Firstly, a work is considered to be original if it is the own intellectual creation of an author that reflects his/her personality (22).

Secondly, a work is considered to be original if it appears that the author was able to express his creative abilities in the production of the work by making free and creative choices (23).

Thirdly, the EU Court of Justice has emphasized that “where the expression of those components is dictated by their technical function, the criterion of originality is not met, since the different methods of implementing an idea are so limited that the idea and the expression become indissociable” (24). In other words, when a work has a technical or functional character, it can only obtain copyright protection if it is observable that certain creative choices have been made. In case of a lack of creative choices, the idea coincides with its expression, which means that the originality requirement cannot be met (25). Consequently, no copyright protection can be obtained.

In short, the expression of personality, including making creative choices, is a key element in EU Copyright. Whereas these are strongly human related elements, the first question is whether works made by AI can be considered to be ‘original’. Given that they have a strong technical and functional character, a case-by-case assessment needs to be done to evaluate the presence of an expression of personality and the reflection of creative choices. If an AI produced work can be considered copyrightable, the second question is who can be legally considered to be the author. The programmer? The Computer? The person who makes use of AI? A third important question is whether copyright protection of AI created works is actually desirable in today’s society.

3. Can works made by AI be copyright protected?

As stated above, copyright protection is theoretically possible if works have a technical or functional character. However, according to the EU Court of Justice, the attribution of copyright is in principle excluded when the idea and the expression become indissociable, taking into consideration that an idea an sich is not copyrightable. This is the case when the final result is actually dictated by a technical function, implying that the originality requirement cannot be met (26). Therefore, it is essential to verify whether (additional) creative choices have been made in order to determine a potential copyright protection.

If it appears that a person has made creative choices and “utilises available formative freedom and thus gives it originality”, the creation can be copyright protected even if a “technical aid” is used (27).

Concerning works created by ‘weak AI’, copyright protection is conceivable. Taking into account that weak AI is merely designed to perform specific tasks, remains under the direct control of a natural person and isn’t technically able to make autonomous or independent or unpredictable decisions, copyright protection is possible if the human author has made free and creative choices and expresses his/her personality. In that sense the weak AI is merely used as a ‘technical aid’.

Concerning works made by ‘strong AI’, the chances of copyright protection appear to be less obvious. This form of AI isn’t as a simple ‘technical aid’ to create a work. Quite on the contrary, strong AI is able to produce works independently and autonomously. Via deep learning, strong AI mechanisms are capable to create new algorithms by their selves, without any human intervention or control. Considering the algorithm-generated products are purely arbitrary and unpredictable, precluding any (human) creative choices, the originality requirement cannot be met.

Taking the above illustrated works “Portrait of Edmond Belamy” and “Psychedelic” as an example, it appears strongly doubtful that these creations can be considered as original. Whereas these works might be seen as original in the common sense, in terms of EU Copyright they aren’t. There is a lack of an expression of the personality of the authors and a lack of creative choices; these works are entirely created by strong AI, without any human assistance or control. The only human choices that have been made, happened at the beginning of the process where existing artworks were selected and put in as data. It is difficult to consider these choices as ‘creative’. Consequently, copyright protection cannot be attributed to the human programmer (considering of the lack of originality) nor to the AI itself, considering it is a machine, an object and not a legal person.

4. Who is the actual author of an AI produced artwork?

Assuming that certain AI produced artworks are potentially copyrightable, the following question is who can be legally considered to be the author.

Considering that copyright protection can only be attributed within the EU when material is ‘original’, meaning that 1) it should be an ‘own intellectual creation’ of the author (28); 2) meaning that it should ‘express the personality’ of the author and (29); 3) that it should reflect the ‘creative choices’ of the author (30), and considering that copyright expires after 70 years after the death of the author, indirectly implies a human intervention criterium. Machines can’t legally ‘die’ or fully express the expression of the author or reflect creative choices. If animals – such as elephants making paintings or monkeys taking selfies (31) – can’t be considered as actual authors – how should it be possible for machines to be legally considered as authors and to benefit a copyright protection?

Mark Perry and Thomas Margoni stated that: “In those legal systems where a work needs to be creative to qualify for protection it might be argued that only a human is capable of authorship. Creativity, especially under the civil law tradition, is a concept that is strongly connected to the person, and may be seen as an emanation of the author’s personality” (32). In other words, following the civil law tradition within the EU, only a human can become the author of a copyrighted work and the involvement of a natural person is therefore an implicit requirement for the copyright protection.

Thus, considering that the EU Court of Justice has emphasized that copyright protection “presupposes that they are intellectual creations” (33), indirectly referring to creations of the human mind, it is strongly doubtful that strong AI produced works – where very little or no creative choices have been made by a natural person – can be copyright protected according to EU (case-)law.

But even if the EU originally requirement would be less strict, even if computers and AI could be considered as legal subjects instead of legal objects and even if AI could be qualified as creative entities and potential authors, the question remains whether copyright protection for (strong) AI produced works is actually desirable in our societies.

5. Is copyright protection desirable for works made by Artificial Intelligence?

In order to evaluate desirability of attributing exclusive economic and moral rights to those who create works via AI, it is important to keep in mind the ratio legis of EU copyright. The central reason to give copyrights – as an intangible form of property– to creators has multiple dimensions. In fact, not only ensures copyright an important legal protection for authors, it also incentivizes them to develop, to create and to innovate, given that they acquire a legal monopoly without any registration requirements and without any financial obligations. On an individual level, copyright both protects the creation of art and the personal interests of artists and encourages them to create. On a collective level, copyright stimulates progress and welfare in the society as a whole through incentivizing the intellectual creation, production and distribution of works both nationally and transnationally. It’s about promoting cultural, economic and social prosperity on a large scale.

Taking these elements into account, it is essential to question whether copyright protection for AI created works is actually desirable on an individual and collective level. On the one hand, it can be argued that a lack of copyright protection on (strong) AI produced works might discourage individuals to make use of, develop or exploit AI. In this view, no copyright means no legal protection for AI made works, no rights to stop people making copies of these works and most importantly, no rights to exploit them economically. On a collective level, one can argue that the development of Artificial Intelligence might be stagnated because both private individuals and companies are disincentivized to invest in AI. Why bothering spending big amounts of money to the development of Artificial Intelligence if you can’t exploit the AI produced works?

Nevertheless, this ‘chilling effect’ risk needs to be nuanced. The fact that works made by AI aren’t (always) copyrightable doesn’t necessarily imply a restriction on creation or innovation. Firstly, there are alternative legal IP regimes – such as patents on AI related inventions and copyrights on software – that protect inventors and creators and compensate them for their investments. Why should there be an additional protection on the results of the used AI? Besides, the whole idea of attributing copyrights to artworks is to protect and reward individual human creators and not computers. Computers don’t need to be protected or incentivized to create; they are programmed to do so.

Another and essential reason not to foresee copyright protection to (all of the) AI generated works consists of preventing ‘suffocation’. If AI produced works could be considered as original, if each and very AI made work is copyrightable or protectable by a sui generis copyright regime, there would an enormous risk of excessive and large-scale protection of algorithmically produced works, suffocating the market both economically and culturally. If big companies – such as Ikea, international clothing stores or other big chains – obtain copyright on everything they produce via AI, will there be still any room left for creation? Would it be desirable for a society as a whole, would it promote progress, cultural development and economical and social prosperity? Whereas AI might offer important opportunities in terms of art, attributing a legal monopoly on them might not be the answer to success…

Want to know more about AI art? Please contact us at:


(1) X, “Is artificial intelligence set to become art’s next medium?”, Christie’s, 12 December 2018, available at:

(2) Bernard Marr, “The Key Definitions Of Artificial Intelligence (AI) That Explain Its Importance”, Forbes, 14 February 2018, available at: .

(3) Andres Guadamuz, “Do Androids Dream of Electric Copyright? Comparative analysis of originality in artificial intelligence generated works”, Intellectual Property Quarterly, 2017, p. 169-186, available at:

(4) Ibid.

(5) Doug Rose, “What Is Artificial Intelligence?”, informit, 15 December 2020, available at:

(6) Madeleine De Cock Buning, “’Computer generated works’, een test voor de grondslagen van het auteursrecht?”, Computer Recht, 1993, available at:

(7) IBM Cloud Education, What is strong AI?, IBM, 31 August 2020, available at:

(8) Ibid.

(9) Madeleine De Cock Buning, “’Computer generated works’, een test voor de grondslagen van het auteursrecht?”, Computer Recht, 1993, available at:

(10) X, “Is artificial intelligence set to become art’s next medium?”, Christie’s, 12 December 2018, available at:

(11) James Vincent, “How three French students used borrowed code to put the first AI portrait in Christie’s”, The Verge, 23 October 2018, available at:

(12) Ibid.

(13)Alyssa Newcomb, “Painting sells for $432,500. The artist? An algorithm.”, NEWS, 29 October 2018, available at:

(14) Amanda Turnbull, “The price of AI art: Has the bubble burst?”, The Conversation, 6 January 2020, available at:

(16) Ibid.

(17) X, “75% of people think this AI artist is human”, Fast Company, 19 October 2018, available at:

(18)Burkhard Schafer, ‘A fourth law of robotics? Copyright and the law and ethics of machine co-production’, Artificial Intelligence and Law 2015/23, available at:


(20) Berne Convention for the Protection of Literary and Artistic Works, 9 September 1886, as revised at Paris on 24 July 1971 and amended in 1979; WIPO, Berne Convention for the Protection of Literary and Artistic Works, 9 September 1886, as revised at Paris on 24 July 1971 and amended in 1979; WIPO, “Summary of the Berne Convention for the Protection of Literary and Artistic Works (1886)”, available at:

(21) Court of Justice (EU), 16 July 2009, C-5/08, ECLI:EU:C:2009:465, Infopaq International A/S v Danske Dagblades Forening and Court of Justice (EU), 1 December 2011, C-145/10, ECLI:EU:C:2013:138, Eva-Marie Painer v. Standard VerlagsGmbH and Others.

(22) Court of Justice (EU), 1 December 2011, C-145/10, ECLI:EU:C:2013:138, Eva-Marie Painer v. Standard VerlagsGmbH and Others; Directive 2006/116/EC of the European Parliament and of the Council of 12 December 2006 on the term of protection of copyright and certain related rights, paragraph 16

(23)Court of Justice (EU), 4 October 2011, C-403/08 and C-429/08, ECLI:EU:C:2011:631, Football Association Premier League Ltd and Others v QC Leisure and Others and Karen Murphy v Media Protection Services Ltd.

(24) Court of Justice (EU), 22 December 2010, C‑393/09, ECLI:EU:C:2010:816, Bezpečnostní softwarová asociace – Svaz softwarové ochrany v Ministerstvo kultury.

(25) Ibid.

(26) Ibid.

(27) Opinion of Advocate General Trstenjak delivered on 12 April 2011; Court of Justice (EU), 1 December 2011, C-145/10, ECLI:EU:C:2013:138, Eva-Marie Painer v. Standard VerlagsGmbH and Others

(28) Court of Justice (EU), 16 July 2009, C-5/08, ECLI:EU:C:2009:465, Infopaq International A/S v Danske Dagblades Forening and Court of Justice (EU), 1 December 2011, C-145/10, ECLI:EU:C:2013:138.

(29) Court of Justice (EU), 1 December 2011, C-145/10, ECLI:EU:C:2013:138, Eva-Marie Painer v. Standard VerlagsGmbH and Others.


(31)Xavier Wolfgang, “Auteursrecht op (aap)selfies en andere ‘andere foto’s’? Deel II”, Alibi, 2018, available at:

(32)Mark Perry & Thomas Margoni, “From Music Tracks to Google Maps: Who Owns Computer-Generated Works?”, Computer Law and Security Review 2010, vol. 26 (6), 621-629, available at:

Court of Justice (EU), 16 July 2009, C-5/08, ECLI:EU:C:2009:465, Infopaq International A/S v Danske Dagblades Forening and Court of Justice (EU), 1 December 2011, C-145/10, ECLI:EU:C:2013:138.

306 views0 comments


bottom of page