IS THE CURRENT UK COPYRIGHT FRAMEWORK SUFFICIENT FOR PROTECTING EMERGING TECHNOLOGY IN AI-GENERATED ART?
| Author | Singh, Asha Lee Jai |
INTRODUCTION
This paper seeks to address the nexus between art generated by artificial intelligence (AI) and UK copyright law--a relationship that has become increasingly complex over the last ten years. Despite advances in AI, laws in the UK have not adapted to accommodate emerging forms of technology that will impact society in profound and, probably, permanent ways. These changes are particularly true in the creative realm--one of the last areas where AI is most expected to have an impact.
This paper has two aims. First, to identify and clarify the legal issues surrounding AI-generated visual art and, second, to explain why the current copyright system is ill-equipped to accommodate AI art today. The first half of the paper will describe advances in AI art with the use of generative adversarial networks (GANs) and the legal uncertainties facing creators. The second half of the paper will explore the copyright system, raising specific concerns about the originality doctrine. The paper will conclude with proposed modifications to the copyright structure, ultimately calling for a shift in the concept of 'originality'.
ADVANCES IN ARTIFICIAL INTELLIGENCE
The pace at which technology has advanced over the last 30 years is striking. In 1995, the internet started to enter homes. At that time, less than one per cent (0.68 per cent) of the global population used the internet, (1) but by 2019, this had grown to 57 per cent. (2) Today, an estimated 75 per cent of the world's population own a smartphone, thereby giving internet access anywhere and at any time. (3) These statistics point to the undeniable fact that technology is now part and parcel of life. AI is perhaps one of the most exciting technologies, but it carries with it the potential to destabilise and test to the limit traditional notions of copyright law.
There is no single definition of 'artificial intelligence'. In fact, US computer scientists Russell and Norvig propose eight definitions spanning four themes. (4) These include the capacity to think like a human, to act like a human, to reason and to act rationally. (5) The UK Government similarly acknowledges the lack of a formal definition of AI, but does observe that there exist common themes relating to AI. These include "using statistics to find patterns in large amounts of data" and performing "repetitive tasks with data without the need for constant human guidance". (6)
The lack of formal definition reflects the difficult task of defining AI concepts. AI is continually evolving in novel ways. How does one, after all, define 'intelligence', 'autonomy' or 'consciousness'? There is general consensus that Alan Turing brought the concept of AI to the fore. Known for cracking the Enigma Code during the Second World War, Turing also conceptualised a test to determine whether a machine had 'intelligence'. The test, called the 'Imitation Game', deems a machine 'intelligent' if a human interacting with the machine believes it to be a human. In other words, the machine must match human intelligence in terms of reactions and reasoning.
To date, no machine has come close to passing the Turing test. Nevertheless, machines are quite clearly advancing in ostensibly 'intelligent' ways. The EU Parliament goes so far as to hint at the possibility of AI surpassing human intelligence, commenting in a draft report that:
...there is a possibility that within the space of a few decades, AI could surpass human intellectual capacity in a manner which, if not prepared for, could pose a challenge to humanity's ability to control its own creation and, consequently, perhaps also to its capacity to be in charge of its own destiny and to ensure the survival of the species. (7) In the arts, machines need not pass the Turing test to prove convincingly creative. Sophisticated algorithms can train machines to write poetry, (8) create cheesy film scripts, (9) generate tales of Harry Potter, (10) and even mimic the techniques of master painters. (11) The science-fiction ideal of anthropomorphised robots remains in the distant future. While machines today can produce works without the aid of human hands, the devices nonetheless require a human hand to come into being. Hence, references to art generated by AI or AI-generated art refer to art created semi-autonomously using algorithms. Today, there is no art known to this author that has been created entirely without human influence.
When the UK Government published a consultation paper on AI in 2020, it highlighted the 'Next Rembrandt' project. (12) This project, led by the University of Sussex, is one of the more widely known and more sophisticated undertakings utilising AI technology to create visual art. (13) The project's goal is to generate visual art that resembles Rembrandt's style.
When examining this project and the resulting visual art, it is tempting to credit AI as the creator of the artwork. One can fall prey to believing that a machine is creating 'Rembrandts' without human aid. A closer examination of the machine debunks this. However, to understand the Next Rembrandt project and how questions surrounding copyright and AI can become convoluted, one first needs to understand how the machine works. This requires a basic grasp of generative adversarial networks (GANs), a model of AI used to produce the works.
GENERATIVE ADVERSARIAL NETWORKS
At a basic level, GANs underpin AI machines. AI machines have two primary models: generative and discriminator codes. The generative code creates new works. It 'generates'. This work is then judged or 'discriminated' by the 'discriminator code'.
For example, a coder may create an AI machine that can produce pictures of dogs. In this case, the generator creates dog pictures, which get passed on to the discriminator, which evaluates whether the work is sufficiently 'dog-like' for the parameters set for dog pictures. The codes, therefore, play off each other in both dialectic and adversarial fashion. They are 'adversarial' because the generator constantly tries to trick the discriminator.
Another central component of any GAN is the data set, curated by the programmer to train the model. The act of curating this data is known as 'mining'. Anything may be used as data. Researchers for the Next Rembrandt project, for example, collated data from Rembrandt's works, scanning over 300 of the artist's works. (14)
After mining, the programmer applies the generative code to produce a particular type of work. In the case of the Next Rembrandt project, the code generates paintings in the style of Rembrandt. This work passes to the discriminator model, which assesses- -based upon the data set--whether the work 'passes' by resembling the data set. The product gets passed through if it does, resulting in the 'next' Rembrandt.
One of the more beguiling characteristics of GANs is unsupervised learning through the neural network. Neural networks are named after neural systems in the human brain. Like neurons firing in the brain during a learning process, neural networks absorb information, identify patterns and create new connections from a given data set. (15)
This characteristic is unlike more basic forms of AI, such as supervised learning. In supervised learning, humans correct and guide the machines in a somewhat heavy- handed fashion. Unsupervised machines, in contrast, learn 'independently', generating patterns off previous patterns and parameters not part of the original data.
This generative learning, akin to the human brain, has interesting implications for artists and art professionals. In contrast to supervised learning, where the programmer directs the machine's learning, unsupervised learning machines can detect patterns that humans overlook. Theoretically, the machine could see patterns in Rembrandt's work that may be imperceptible to the prestigious Rembrandt scholar--an issue that will be evaluated below.
THE COPYRIGHT FRAMEWORK
Understanding the legal issues facing Al-generated artwork requires an understanding of copyright law. Copyright gives the holder the exclusive right to make or to authorise the making of copies, and therefore to prevent the making of unauthorised copies of a work. The creator of the work will also have various moral rights under the copyright legislation.
The UK has two crucial criteria for copyright under the Copyright, Designs and Patents Act 1988 (CDPA). First, a 'work' must fall within a recognised category of works eligible for copyright protection. (16) Visual art created with AI will fall within the Artistic Works category. (17) These include:
(a) a graphic work, photograph, sculpture or collage, irrespective of artistic quality,
(b) a work of architecture being a building or a model for a building, or
(c) a work of artistic craftsmanship. (18)
In addition, databases are also eligible for copyright protection. (19) The CDPA defines a database as "a collection of independent works, data or other materials", which are "arranged in a systematic or methodical way" and which "are individually accessible by electronic or other means." (20)
The second requirement for copyright protection is originality. The work must be "original" or "the author's intellectual creation". (21) This requirement is pivotal in the discussion of AI-generated art. It is also perhaps the most complicated.
In the UK, the notion of originality has changed over time. Before joining the European Union, the UK's concept of originality was labour, skill and judgment. (22) The terminology used has not always been consistent, with the courts referring sometimes to "labour, skill and judgment", and at other times to "labour, skill or judgment". In either case, the three ingredients remained the common denominators. After joining the EU, the UK aligned its case law with that of the Court of Justice of the European Union (CJEU), which defines originality as "the author's own intellectual creation". (23) This definition of originality...
To continue reading
Request your trialCOPYRIGHT GALE, Cengage Learning. All rights reserved.
