Could Copyright Benefit From AI?

Copyright and AI are not besties, there’s no surprise here.

However, when topics become as “trendy” as Artificial Intelligence, it becomes increasingly hard to find different perspectives emerging from the established narrative.

The mainstream opinion wants Artificial Intelligence to be copyright’s worst enemy, but what if a deeper look at the basic functioning of this breakthrough technology can shed light on a viable path of peaceful and even fruitful coexistence?

What if Generative Artificial Intelligence may also be able to address the same challenges that it creates in this legal domain?

Let’s set the perimeter first.

Copyright protects the creative works of the human mind. The quantum of creativity required for the work to be suitable for protection is very low and does not take into account any artistic merit or commercial value.Although copyright emerges upon the creation of the work and does not require registration for its eligibility, it is always advisable, for evidentiary purposes, to undergo such formality — especially in the U.S., where registration is necessary to file copyright infringement lawsuits.Because the originality threshold for copyright protection is low, the Copyright Office’s registration is traditionally assumed to be nearly automatic.As a result of the low quantum of creativity required and of the absence of proper registration procedures for copyright to exist, the right’s legal scope is ultimately defined by courts in the event of conflicts.To do so, judges have to compare the allegedly infringing work with the pre-existing and supposedly infringed copyrighted work in order to establish substantial similarity and determine whether infringement has occurred.Judges perform this task on an ad-hoc basis by applying various principles and doctrines (substantial similarity, idea/expression dichotomy, fair use..); methods which have proven to be scientifically inaccurate and subjective in the outcome due to their abstract and elusive nature.

Generative AI can, based on an expressive input or prompt, generate human-like materials (text, image, music, videos..) in an ever-increasingly higher quality and shorter time frame.This disruptive technology is able to provide outputs based on a predictive method: by analyzing vast datasets to identify patterns and correlations, AI models learn to make predictions or decisions when presented with new, unseen data. AI’s operation heavily depends on training sets, which are massive collections of data used to teach the system to recognize patterns.However, Artificial Intelligence introduces considerable friction within the status quo, particularly concerning Copyright; this friction can be substantiated by roughly three basic issues:Firstly, the mentioned training set predominantly consists of copyrighted materials, and as such, these materials should only be used with the approval of the rights holder.Secondly, is the output produced by AI a lawful independent creation (thus eligible for copyright) or is it an infringing derivative of the copyrighted works which were used to train the system?Thirdly, who owns the copyright over the outputs produced by AI? Is it the prompter (human that interacts with the machine), is it the AI’s developer (OpenAI if we consider the example of ChatGPT) or is it the machine?Lastly, can we stretch the notion of authorship to the extent that machines can be regarded as authors?

While such questions deserve a separate and thorough discussion, the object of this article is to offer a different and intriguing perspective: what if Generative AI could positively shape the landscape of Copyright by providing a scientific and objective method to assess the originality of copyrighted works?

Let’s build on our premises.

At present, because the level of originality for copyright eligibility is low, most works will pass this threshold even if they incorporate many generic expressive compositions.Until recently, measuring the genericity of expressive compositions was not feasible at the registration phase and was done in a rather loose, intuitive, and imprecise manner in the occurrence of an infringement.However, copyrighted works are the results of various compositions of basic elements (so-called “building blocks” such as lines, shapes, colors, musical sounds..).Such elements are often very nuanced and hardly visible to the human eye, which only grasps the final work but not the underlying components of it. AI systems like DALL-E can generate a visual image from a textual prompt thanks to the aforementioned basic blocks: during training, models learn recurrent patterns among basic elements and then apply these patterns to generate new content in response to users’ prompts.This core feature of AI demonstrates its ability to detect, categorize, and study hidden interconnections among elements of copyrighted works.Thus, GenAI models can also be used to rank copyrighted works based on the novelty of their compositions and to assign different originality scores depending on the genericity of their elements.

The ability to harness GenAI to measure copyright originality has far-reaching implications for copyright law:

• Copyright registration offices might, similarly to patent law, adopt a novelty standard for copyright eligibility: Registration offices might evaluate the originality of expressive compositions of works at the time of their creation and reserve legal protection only to works showcasing “artistic distance” from pre-existing works in the field.

• Courts could rely on originality scores as a scientific method to assess substantial similarity in copyright infringement lawsuits. In this context, a lower originality score (indicating high resemblance) would suggest greater similarity with the allegedly infringed copyrighted work. Consequently, this would increase the burden of proof for the defendant to demonstrate that infringement has not occurred.

• Artists and right holders might benefit from originality scores as an objective index of market value when determining royalties in licensing practices.

Leveraging AI to measure originality could impact all the major phases in the lifecycle of copyrighted works, from registration and licensing to copyright infringement litigation.The resulting, more rigorous, copyright system would certainly adhere to copyright’s raison d’être, which is to enrich the domain of artistic works, enhance creativity, and thereby improve society’s well-being. Society has no interest in protecting generic expressions that do not entail any advancement in art or literature.

CC: Francesco Gaggioli