Stranger than Sci-Fi Part 1: Can (and should) Artificial Intelligence machines own intellectual property?
Artificial Intelligence, or AI, refers to the capability of a machine to imitate intelligent human behaviour. Though it sounds like something out of a sci-fi film, the reality is that AI is quickly becoming a norm in our everyday lives, from the simple AI systems used to sort email inboxes, to complex AI systems known as “Generative Adversarial Networks” (GANs), which can be trained to autonomously produce new products without human input. This has resulted in a new type of intellectual property: AI-generated IP.
The creation of IP solely by machines without human contribution poses novel challenges to existing IP laws including questions, like who owns IP created by machines without human contributions and how can businesses generating products via AI protect their IP rights?
In the first article of this two-part series, we explore how existing Australian copyright and patent legislation deals with ownership of AI-generated IP. In the second article, we consider whether AI machines should be recognised as the ‘owners’ of the IP create.
Examples of AI-generated IP
GANs can produce a large range of IP, with some examples being:
- Obvious AI’s Portrait of Edmond Belamy, the first AI-generated artwork to be auctioned off, fetching a whopping $432,500 at Christie’s in October 2018. The Obvious AI team obtained GAN algorithms from an open source online forum and trained the algorithm with classical portraits and super-resolution algorithms. The end product is a painting which looks like an 18th century gentleman.
- DABUS AI, an AI system created by Dr Stephen Thaler, created food containers based on fractal geometry. We discuss below the issues faced by the DABUS AI team in obtaining patents over the end product of their AI system.
Can Artificial Intelligence machines be recognised as owners of intellectual property in Australia?
The Australian patents and copyright legislative frameworks currently do not allow for the recognition of wholly AI-created work. This is perhaps unsurprising, given that these frameworks were first conceived of long before AI was anything more than a faraway concept in a sci-fi novel (and given Australia’s IP legislation inability to keep up with the ever-increasing pace of technological change more generally).
The current state of the law views machines as merely tools used by human creators to produce works, the same way that Microsoft Word is used a writing tool, but not the author of a final masterpiece.
Conversely, where there is no human author involved, no ‘original work’ is recognised for the purposes of copyright protection. One example is satellite images of weather patterns that are collected entirely automatically, where information is automatically processed by specialised computer programs and the only human contribution comes in downloading the final image with the press of a button. In this situation, is it right to recognise the button-presser as the ‘author’ of the satellite images for copyright purposes?
So what? The implications of non-recognition of AI Ownership
The current lack of legal framework for recognising non-human authorship and inventorship means that work created by AI machines may struggle to obtain IP protection.
A stumbling block to obtaining copyright protection in the past has been the inability to identify specific human authors behind work. This was the issue for phone book directories created by automation processes in the case of Telstra Corp v Phone Directories, as well as datasheets created by the operation of computer programs in Acohs Pty Ltd v Ucorp Pty Ltd. Though there has yet to be a case regarding purely AI-generated work in Australia, these cases indicate that it would likely be difficult for works generated by AI machines to obtain copyright protection.
Though Australian courts have yet to decide specifically on the subject of patenting AI-generated products, patent protection may be similarly difficult to obtain. The High Court has affirmed as recently as 2018 that a subject matter is only patentable if it is the result of ‘human action’. Similar resistance has been seen in overseas jurisdictions. In 2019, the European Patent Convention (EPC) rejected a patent application by DABUS AI on the basis that the AI machine as the inventor. The EPC’s rationale in rejecting the application was that inventors must always be humans and that machines are merely tools.
As noted above, IP legislation has always been notoriously slow at keeping pace of new technological and is often not drafted flexibly enough to apply to novel scenarios. The issues AI-generated products face in obtaining IP protection clearly illustrate the problems that arise from the law’s inflexibility in this area.
In our next article, we consider directions for further IP law reforms, including the pros and cons of recognising AI machines as the owners of intellectual property. Is this a necessary step in the evolution of our IP laws to reflect properly the future of technology, or is this another way in which the machines are trying to take over?
The authors wish to thank summer clerk Ying Yi Lim for her contribution to the research and drafting of this article.
 Section 15 of the Patents Act 1990 (Cth) allows patents to be granted only to eligible “persons”. Similarly, section 32 of the Copyright Act 1968 (Cth) requires a work to be the result of intellectual effort by a human author in order to enjoy copyright protection.