Why US law should accommodate for computer-generated work
The advent of computer-generated work (“CGW”) is leading to a global re-evaluation of copyright law. The UK, for example, has enacted special provisions, which explicitly address AI-authorship. As per the Copyright Designs and Patents Act (“the CDPA”) the ‘author’ of such work “shall be the person by whom the arrangements necessary for the creation of the work are undertaken”.
The US has been far less progressive – with the law failing to modernise so as to include a regulatory provision on CGW specifically. Indeed, James Grimmelmann – a professor at Cornell Tech – has recently argued that the need for one is “more apparent than real”. He has wrongly insisted that CGW poses “no distinctive conceptual problems”, and that computers are simply digital tools. Consequently, the existing copyright framework requires no change. Such a view is flawed and outdated.
First, a SCR is needed to legally address circumstances where a computer evidences itself as capable of authorship. Regulation must reflect, contrary to Grimmelmann, that a program does not serve merely as a tool for the real author. While admittedly, his unapologetic contention that “whether you use Microsoft Word or a pen, Finale or drums…any work might have been computer-generated” may have had force a decade ago, it no longer holds true in light of the level of sophistication computers now exhibit.
For example, ‘The Rembrandt’ is an algorithm which analysed a data-set of 15,000 artworks and produced its own original painting. The resulting piece was so remarkable that media outlets have doubted observers’ ability to distinguish it from an original. Another example is ‘BRUTUS’, an AI capable of generating stories with varied plots and characters. It reportedly specialises in the literary concept of betrayal and is likened to a “synthetic version of Proust, Joyce or Kafka”. These examples demonstrate that technology no longer acts as a passive facilitator. It raises novel challenges which must be addressed by equally novel responses.
Second, a SCR is needed to clarify the indeterminate legal status of works created by ‘independently functioning’ AI. Currently, caselaw dictates that programmers are only able to obtain copyright over computer-generated output which is made under ‘direct human guidance’. As such, in the US, the by-products of ‘BRUTUS’ and ‘The Rembrandt’ would most likely be released to the public domain.
This counters the principle of transitivity: intuition rightly suggests that programmers should (at least sometimes) receive intellectual property rights in the works generated by their software. They are, after all, the ‘authors of the authors’ of the work. In addition, granting copyright in such cases would reward coders for their labour, thereby serving US copyright law’s stated purpose: promoting progress by incentivising creativity.
Third, a SCR must be enacted to address the challenges posed by sequential creation – where third-party users employ computer programs. Depending on the program, either the user, the programmer or the AI could merit copyright. As the analogue world seldom requires us to contemplate whether rights should be granted to a penmaker over a novelist or indeed the pen itself, the existing framework unsurprisingly offers no guidance for such cases.
Moreover, while it has been suggested that a recipe raises similar difficulties, this is unconvincing since the active involvement of the AI and user may result in an output entirely distinct to that initially envisioned by the programmer. A further challenge raised by ‘sequential’ work is the need to legally distinguish between computer-users which are genuine authors, such as those using Finale to create a string quartet, and those who merely push a button, such as an animation by pressing the Space bar. A SCR would answer these questions.
Finally, a SCR is needed to square algorithmic creativity with the existing definition of originality; a prerequisite to US copyright grants. Here, the case of Naruto v Slater is authority for authorship being confined to ‘original intellectual creations’. Moreover, as per Feist Publications, authorship is also contingent on “a spark of creativity” and an exertion of ‘consciousness’.
While computers cannot meet these criteria, they can indeed be original. The Cybernetic Poet, for one, analyses human poems and generates new ones. The output’s originality is validated via an incorporated software module which prevents plagiarism. Computer originality is likewise seen in computers’ non-deterministic creation: the same program can be run repeatedly with identical inputs to generate different works. A SCR would therefore develop the ‘originality’ standard to one more suited to the digital age.
Overall, it is clear that the US needs a special copyright regime for computer generated work. It must follow the UK in legally acknowledging that computers can now ‘create’ because of, but separate to their programmers. Doing so would promote progress by rewarding coders for their AI’s outputs whilst also tackling the challenges of sequential work and modernising the ‘originality’ criterion.
Bridy, A. (2012). Coding Creativity: Copyright and the Artificially Intelligent Author. Stanford Technology Law Review.
Cohen, J. E. (2007). Creativity and Culture in Copyright Theory. UC David Law Review vol.40, 1151-1205.
Murray, A. (2019). Information Technology Law: The Law and Society (4th Edition). Oxford University Press: Ch. 10.