AIPPI Congress Report 4: Copyright in AI generated works

Does and should copyright vest in works created solely by Artificial Intelligence (AI)? This is the  very issue discussed in this fourth report [previous reports here, here and here] from the 2019 AIPPI World Congress. 

Toby Bond (Bird & Bird) reports:

AIPPI Congress Report: Copyright in AI generated works

by Toby Bond

The next IPKat?

Should copyright (or related rights) subsist in works created using an AI system? If not, should these works obtain some other form of protection? These are thorny questions. They touch on the legal, economic and moral justification for intellectual property protection. They also require us to consider the relative value we attribute to human endeavor in creating works as compared to the human endeavor in creating AI systems which can themselves create works. With AI generated music available on streaming services, algorithms being signed to record labels and AI generated artworks sold at auction for six figure sums the answers to these questions already have real economic significance.

Not an organisation to shy away from grappling with difficult questions, AIPPI’s 2019 Study Question “Copyright in artificially generated works” took stock of the current position and explored possibilities for future harmonisation. 30 AIPPI groups responded to the Study Question, published in January, expressing a wide range of views on the eligibility of AI generated works for existing forms of IP protection and the desirability for new forms of protection. The responses were helpfully summarised in advance of this year’s AIPPI Congress in London by the Reporter General’s Summary Report followed by a draft resolution. Over a seven hour session on Sunday, a three hour plenary on Monday and a vote on a final resolution on Wednesday, the promise of lively debate on this challenging and complex subject did not fail to disappoint.

Expertly guided over the three sessions by Chair Jan Bernd Nordemann, Vice Chairs Edouard Treppoz and Dale Nelson and Secretary Christian Thomae along with Rapporteur Guillaume Henry, discussions were focused on three key issues:

(1) the application of existing copyright protection (in the sense of the copyright which applies to works which fall within Article 2 of the Berne Convention) to AI generated works;

(2) the application of other existing copyright types and sui generis IP rights (e.g. copyright applying to works such as sound recordings which don’t fall within Article 2 of the Berne Convention) to AI generated works; and

(3) whether there should be a new sui generis type right for AI generated works.

Rather than a blow by blow account, what follows below are the author’s reflections on the discussions as a whole and the final resolution which emerged. 

Not necessarily AI generated

Existing (Berne Convention) Copyright and Related Rights

As a starting point many agreed that in order to obtain copyright protection the creation of AI generated works must involve some form of human input. Given that many see the purpose of copyright as rewarding human creativity, the broad consensus in favour of this proposition was perhaps not surprising. The devil is, of course, in the details and much time was spent on Sunday discussing the level of human input required. This focused on two aspects.

The first was the causal involvement which a human must have in the creation of the work. After discussion a general consensus emerged that there must be some form of human “intervention” in the creation of the work in order for that work to become eligible for copyright protection. Agreement was also reached on what would not be considered a human intervention in the creation of a work. Merely creating an AI system which goes on to create a work would not (on its own) be considered a human intervention in the creation of a work generated by that system. Merely selecting a single work from a number of works created by an AI was also not a human input in the creation of that work (although it was recognised that copyright may subsist in a compilation as a new work created by the selection from multiple existing works). 

Human intervention was therefore proposed as a necessary condition for copyright to arise in an AI generated work. However human intervention alone was not felt to be a sufficient condition for a work to obtain copyright protection. It was agreed that there must also be some form of human intellectual input in the creation of the work. Much discussion followed regarding whether the mental requirement should be described in the resolution as “originality”, “creativity” or something else. The divergence of views mainly reflected the differences in wording used around the world to describe the qualifying criteria for obtaining copyright for works falling within Article 2 of the Berne Convention, rather than reflecting differing views on a specific test for AI generated works. The requirement that “Originality (as interpreted by national laws) of the generated work resulting from the human intervention should be a condition for the protection by Copyright” was accepted as a compromise to allow countries to continue to apply their existing rules on the qualifying criteria for copyright protection. 

Indeed the overall view reached was that copyright should apply to AI generated works in exactly the same way as it applies to any other work. This applied to both the subsistence of copyright and to the issues relating to the scope of protection, the application of moral rights, the term of protection and initial ownership.

The same conclusion was reached in relation to existing related and sui generis IP rights and was perhaps expressed more clearly in the final resolution as follows:

AI generated works should not be precluded from obtaining protection by existing Related Rights on the basis they are AI generated and should obtain that protection as long as they meet the required criteria for obtaining protection.

New protection for AI generated works

Having tackled existing rights, the discussion turned to the more controversial topic of introducing a new right to protect AI generated works. This had been suggested by around 40% of the respondents to the study questions to address situations where an AI generated work would fail to obtain protection under existing copyright or related rights. The UK group for example had proposed the introduction of a new sui generis right for AI generated works which fail to obtain existing copyright protection, but would have done so if they had been created by a human author (see the UK group’s responses 26 - 30 to the Study Questions). 

Taking a similar approach to that proposed by the UK group, the initial draft of the resolution included a proposal for a new sui generis/related right in order to incentivise and protect investments made concerning the AI generation of specific works. Obtaining this right would require a significant investment and the term and scope of protection would be lower than the protection provided by copyright.

This proposal for a new right was met with strong resistance during the Sunday session. Some expressed concern that the low cost of producing AI generated works could lead to a situation where these works vastly outnumbered human created works. It was suggested that offering protection to these AI works would give rise to a multiplicity of rights which could be used to stifle human expression. Others felt that it was still too early in the development of AI technology to reach a conclusion and that much more work would be required to understand the economic, legal and moral implications of a new right before a consensus could be reached on whether one was required.

At the end of Sunday’s session those against a new right prevailed and the draft resolution was amended to state that AI generated works should not be protected by a new related right. However after some time to reflect and some lively behind the scenes diplomacy, this definitive anti-new-rights stance was softened during the plenary session on Monday and the following wording was adopted: 

As AI is still developing, it is too early to take a position on the question of whether AI generated works not covered by such existing protection should be eligible for exclusive rights protection as a Related Right or as exclusive rights under copyright (not in the meaning of the [Berne Convention]).

With this contentious issue addressed, delegates at Wednesday’s session agreed to finalise the wording clarifications proposed by the UK group and subsequently approved the resolution as a whole.

Final reflections

AIPPI’s resolution (yet to be published officially at the time of writing) provides a valuable contribution to the debate on the application of copyright and related rights to AI generated works. It concludes that the existing copyright framework for assessing whether the creation of a work involves sufficient causal and intellectual input from a human are already capable of dealing with AI generated works. It also identifies a gap in protection for some forms of AI generated works and suggests that further work is required in order to understand the need for a new right to protect them. While not providing a definitive answer, the resolution provides a strong platform for future work.