What Makes an AI Generated Work “Original”?
The emergence of generative artificial intelligence (AI) products in the past couple of years has significantly increased the capacity for individuals, businesses, and organisations to utilise AI to produce a wide range of content at great speed and low cost for personal, professional, and business purposes. ChatGPT, DALL-E, Midjourney, Claude, and Grok are a small number of prominent generative AI tools that are increasingly being used by business and the general public in an ever-growing range of use cases, producing literature, music, images, and art.
The rapid rise of generative AI has drawn attention to a number of copyright issues, including the question of copyright protection for content generated by those tools. In the United Kingdom, a provision of the Copyright, Designs and Patents Act 1988 (CDPA) (section 9(3)), part of a broader set of rules regarding the authorship of different categories of works) provides that “in the case of a literary, dramatic, musical or artistic work which is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken”.
On its face, s. 9(3) appears to proceed on the basis that copyright ought to subsist in literary, dramatic, musical, and artistic works which are created by a computer, which would seem to include works created by generative AI. In fact, Parliamentary records (Hansard) indicate that the proponents of the CDPA, which was enacted in 1988, “[believed] this to be the first copyright legislation anywhere in the world which attempts to deal specifically with the advent of artificial intelligence.”
The legislature, therefore, envisaged that literary, dramatic, musical, or artistic works may be generated by computers – indeed by “artificial intelligence” – and it specifically provides that the authorship of such works should be attributed – as a matter of law – to the person making the arrangements for the creation of the work. There may be a question in specific cases which person exactly should be identified as the person who made those ‘arrangements,’ but the principle seems clear that the question of authorship itself does not come into question merely because a work is computer-generated.
However, the correct legal effect of s. 9(3) CDPA has increasingly been drawn into question, particularly in view of a significant shift that took place in the past few years in the understanding of the concept of originality under the law.
Changes in the concept of originality in UK copyright law
Under the CDPA, copyright subsists in the following types of works:
- original literary, dramatic, musical or artistic works;
- sound recordings, films or broadcasts, and
- the typographical arrangement of published editions.
Authorship and originality are two related concepts. Under the CDPA, literary, dramatic, musical, and artistic works (sometimes known as “authorial works”) require originality to attract copyright protection. Under Section 9 of the CDPA, with the exception of computer-generated works, the author of such works is their creator. With regard to the types of works that do not require originality to enjoy copyright protection (sound recordings, films or broadcasts, and typographical arrangements), other provisions were made in s. 9 attributing authorship to persons by reference to their responsibilities in relation to the generation of the work (for example, the authors of a film are deemed to be the producer and principal director and the author of a sound recording is its producer). However, the rule under s. 9(3) is the only rule of deemed authorship that applies to types of works which are protected in copyright subject to a requirement of originality. This raises the question of how deemed authorship can be reconciled with originality.
The context in which this question falls has changed significantly in the past few years, not necessarily as a result of technological developments, but rather due to legal ones. At the time the CDPA was enacted, and until recently, the test for originality under English law was the ‘skill and labour’ test which considers whether a sufficient degree of skill, knowledge, mental labour, taste, or judgement were expended by the author in creating the work. The traditional rule required only a minimal degree of skill and labour (more than an insubstantial amount). Under this test of originality, the English courts had historically been able to recognise originality in works that were generated in circumstances where there was no human author (for example, in Nova Productions Ltd v Mazooma Games Ltd [2006] EWHC 24 (Ch) the Court accepted the subsistence of copyright in artistic images produced by a computer game in the course of play).
However, since the decision in 2010 of the European Court of Justice’s in Infopaq International A/S v Danske Dagblades Forening (Case C-5/08) EU:C:2009:465, the question has been debated in a number of subsequent cases by the English courts, and it has now been established by the Court of Appeal in its decision in THJ Systems Ltd v Sheridan [2023] EWCA Civ 1354 that the traditional common law test for originality in copyright is no longer applicable. Through its interpretation of EU copyright legislation, the Infopaq decision adopted a uniform test of originality which defines an original work as one which is “the author’s own intellectual creation.” Although the UK (following its withdrawal from the EU) is no longer subject to the jurisdiction of the EU courts, the Court of Appeal in THJ Systems treated the EU test of originality as a test that had already been incorporated into UK law before Brexit through past cases that applied the Infopaq test.
As summarised by the Court of Appeal in THJ Systems, the test of originality of the “author’s own intellectual creation” considers at its core whether “the author was able to express their creative abilities in the production of the work by making free and creative choices so as to stamp the work created with their personal touch.” This test is widely considered to set a higher standard than the traditional common law ‘skill and labour’ test. Although the Court of Appeal emphasises that the EU test is an objective one and not one of “artistic merit” and the court will not seek to assess the artistic quality of the work, it is still more than the minimalistic test previously applied by the English courts. The work must be an expression of the author’s creativity, choice, or judgment.
Can a computer generated work meet the test of originality of the “author’s own intellectual creation?”
Computers (including so called “artificial intelligence” software) do not possess an intellect and the works generated by AI do not express the “intellectual creative choices” of the software. In fact, the process by which AI generated content is produced is as far removed as possible from the idea of an intellectual creation. However, this does not necessarily mean that works created by AI cannot meet the originality test of the “author’s intellectual creation.”
The aforementioned THJ Systems case, although it did not deal with an AI-generated work, can provide some guidance. The dispute concerned “risk and price charts” that were generated by a business software application. The case focused on the authorship (which dictated the ownership) of the works generated through the use of the software, although the question of subsistence of copyright was also considered. The claimant, the creator of the software, argued that the outputs it generates where his copyright works. The defendant denied there was copyright in the charts and argued that if there was, then under s. 9(3) CDPR, authorship ought to be attributed to the user of the software, the defendant, not to the software developer.
At trial it was established that the graphical works in which the claimant claimed rights protected by copyright and which were reproduced by the software every time it was used were created by the claimant in developing the software. Although the user’s use of the software resulted in different figures being populated in the charts, those figures were not the copyrighted work in question, but rather it was the structure of the charts and other design elements displayed on the screens that attracted copyright and those were not the result of any choice or action taken by the user.
On the question of the subsistence of copyright, the Court of Appeal held that the trial court misapplied the law by considering the matter against the now-discarded “skill and labour” test, rather than the Infopaq test of the “author’s own intellectual creation”. However, the Court of Appeal did not interfere with the trial court’s conclusion that there was sufficient originality in the structure and arrangements of the charts to justify copyright protection. Instead, the functionality of the software was held to be irrelevant to the question of originality. The enquiry concerned the visual appearance of the graphical works. Even though the charts were generated by the software each time it was used, the Court found that they were carefully arranged by the software designer who applied judgment and choice to achieve a certain visual result. It was not suggested that the particular way in which the charts were organised was dictated by technical requirements such that would exclude a finding of originality and copyright protection. The Court held that the degree of originality was clearly low (with the result being that only an identical copy could amount to an infringement) but the resulting artistic work was still the expression of the author’s (that is, the human software designer’s) intellectual creation and choices and that the works were sufficiently stamped with his personal touch, so that they can attract copyright protection.
Accordingly, a graphical work can be at the same time a computer-generated work for the purpose of s. 9(3) CDPA whilst originality can be assessed by reference to the work carried out by the human designer who was responsible for making the relevant choices or judgments which resulted in the work being created (by the computer) in a certain way.
The analysis with regard to AI-generated works could proceed along very similar lines. Although the charts in the THJ Systems case were directly arranged by a human developer (and not automatically by an AI), the Court of Appeal emphasised that the functionality of the software does not matter and, at least with regard to artistic works, a key element of evidence as to originality is the work itself.
If a graphical work is generated through an AI tool and looks, on its face, like an original artistic work, a court would look at the creative choices and judgments made by the user (the author) and the extent to which the work reflects those choices and whether it is stamped by the author’s personal touch. It should not matter in principle that the graphical output itself is generated automatically by the software. It should not be a determinative factor whether the user applied any brushwork or drew any lines (virtual or otherwise). It would seem contrary to the intention of Parliament in enacting s. 9(3) CDPA to exclude copyright in an artistic work purely because it was generated by a computer. As long as it can be demonstrated that a human author has made sufficient free creative choices or applied sufficient judgment so that the resulting work expresses his own “intellectual creativity,” that should be sufficient for finding originality in the work.
But the decision in the THJ Systems case leaves important questions unresolved. The question can be taken to the extreme, such as in the case of a work that is generated by AI with minimal user input. Here, a court might find insufficient authorship to amount to ‘creative choices or judgments’ or making the work a reflection of the author’s “own intellectual creation” – for instance an image produced by an AI tool based on the instructions: “flowers in a vase; natural style; green, red, yellow.” It would be difficult to argue that such banal instructions amount to creative choices that make the resulting work a reflection of the user’s creative free choices and judgment. But given the provision of s. 9(3) CDPA – which suggests that copyright should not be denied from an authorial work merely because it is computer-generated, perhaps in such a case authorship of the resulting work should be attributed (as in the THJ Systems case) to the designer of the software tool, rather than the user, and the creative choices and judgment can be said to have been those made by the developers who trained and developed the AI tool. After all, without their work the computer-generated image could not have been created. As long as the image is not a copy of images that were used to train the AI, perhaps it is the creative efforts of the software developers that should justify protecting the resulting image in copyright?
On the other hand, can it seriously be argued that the unlimited number of different images that an AI image-generating tool can produce are all the reflection of the software developers’ creative choices and judgment? Can it be said that all such works (as long as they are not sufficiently moulded and individualised by the user of the tool) are stamped with the “personal touch” of the software developers? It seems doubtful.
It is perhaps more likely that courts in the UK would take the approach that s. 9(3) CDPA does not mean that all computer-generated works must necessarily attract copyright. It is still necessary to show in each case that the work is original in the sense that it reflects its author’s free creativity and intellectual choice and judgments – whether that author is the user of the software or its creator. That would be a highly factual investigation and may depend on the nature of the AI tool, how it was developed, what it does, and the way it was used by humans to generate a particular output.
Potential legislative reform
The UK Government is currently conducting a public consultation on copyright issues related to AI. One issue under consideration is the special rule of authorship for computer-generated works. The consultation paper invites feedback from the public and relevant stakeholders, but also expresses an initial view that it may be preferable to abolish section 9(3) CDPA on the basis that the provision creates more uncertainty than it resolves. In forming that initial view, the Government took note of the fact that other major legal systems such as the European Union and the United States did not adopt a similar rule of deemed authorship regarding computer-generated works.
The Government’s view (which is not binding on the courts) is that even without the special statutory rule of authorship under s. 9(3) CDPA, literary, musical, artistic, and dramatic works which are computer-generated but which exhibit human creativity (that is, works created by humans with the assistance of AI tools) would be protected in copyright on the basis of the general rule of originality. It is also noted that computer-generated films, videos, sound recordings, and broadcasts will enjoy copyright protection in any event since the requirement of originality does not apply to those types of works.
It is yet to be seen if s. 9(3) CDPA is to be abolished or if any further clarifications might be made in legislation regarding the copyright protection of AI-generated works. As long as the law remains unchanged, courts are likely to consider each case on its facts with a focus on the degree of individual creativity that contributed to the creation of each work.
