A recent rapid improvement in generative AI technologies is seeing them used to create a wide range of outputs. After pushing past the novelty of having an AI tool write you song lyrics about dating a racecar driver in the style of Taylor Swift, or create a ‘photograph’ of Spiderman as if his outfit was designed by Pablo Picasso, many have begun putting generative AI tools to real applications, often with very practical outcomes.
The useful outputs of these processes will often be the kinds of works traditionally protected by intellectual property – literary works, software source code, images, sound recordings, and even inventions. However, there are some very real questions about who would own the intellectual property rights arising from the use of generative AI tools, or perhaps whether they are capable of being owned at all.
In our recent article, we discussed the training processes used by ‘generative AI’ tools, and the associated intellectual property implications. In that article, we did consider some issues associated with the output of AI tools – specifically, whether there may be instances where the output created could infringe the training data on which it is based on.
In this piece, we consider those outputs further, asking whether we can apply existing intellectual property frameworks to the ‘creative’ output of a generative AI tool.
Can computers and machines create intellectual property?
Although generative AI is a relatively new field, the legal question of whether machines and computers can create intellectual property rights is not an entirely new one. These questions have been before the Courts in years past, in respect of issues such as whether copyright subsists in a telephone directory compiled via automated means.
As set out below, existing laws tend to require the involvement of some human author or inventor, and so the output of generative AI tools is arguably unlikely to attract IP protection, although there may be scope to debate whether a person prompting the tool has had sufficient input to be deemed an author.
Patents
In recent years, there have been attempts to patent inventions created by machines and computers. The first such case in Australia concerned an invention by DABUS, an AI system created by Dr Stephen Thaler.
DABUS was named as the inventor in a patent application for an invention said to be ‘autonomously generated by an artificial intelligence’. After being refused registration by the Commissioner of Patent, Dr Thaler filed a judicial review. Although DABUS enjoyed some initial success in the Federal Court, the Full Court unanimously set that ruling aside, finding that only natural persons can be named as an ‘inventor’ for the purposes of the Patents Act 1990 (Cth) (Patents Act). The legal battle in Australia ended with the High Court refusing special leave to appeal the decision of the Full Court of the Federal Court.
Section 15(1) of the Patents Act provides a list of ‘persons’ entitled to be granted a patent for an invention. Although the Full Court acknowledged that section 15(1) of the Patents Act does not definitively state that an inventor has to be human, there is a line of authority that established that the grant of a patent is premised upon an invention from the mind of a natural person.1 That said, it is worth noting that the Full Court did not rule out the possibility of the creator of that system being registered as the patentee:2
If the final concept of the invention as described in the specification and claimed in the claims would not have come about without a particular person’s involvement, then that person has an entitlement to the invention.
Copyright
Unlike patents which protect inventions and innovations, copyright is a right that subsists in the original works of an author, such as literary works and artistic works. The output of an AI tool like ChatGPT is clearly in the nature of a literary work, and similarly the output of an image generation tool like Stable Diffusion is in the nature of an artistic work. However, merely being a relevant work is not sufficient to attract copyright protection.
At the heart of copyright protection in Australia is the concept of authorship. Section 32 of the Copyright Act 1968 (Cth) (Copyright Act) makes it clear that copyright subsists in the original works of an author who was a ‘qualified person’, defined to include an Australian citizen or a person residing in Australia. There is also common law jurisprudence that established that the creation of original works must involve ‘independent intellectual effort’,3 such that an automated process cannot attract copyright.4
As such, the absence of a human author for these works is likely an insurmountable hurdle to any copyright arising, although as discussed below, there may be scope to argue that the prompter using the tool provides that human input.
Who, if anyone, owns the copyright in AI-generated content?
For IP rights to exist in machine generated content, some change to the law would likely be necessary to clearly recognise those ownership rights. But in that scenario, who should be entitled to own those rights? The end user of the tool, or the creator of the tool itself?
End user as the owner of copyright?
The Copyright Act recognises photographers as the author of a photograph for the purposes of copyright. On a modern digital camera, the process of capturing a photograph can involve all manner of computation largely beyond the control of the photographer, but there is little dispute that the person framing the shot and pressing the shutter button is exercising sufficient artistic control to be considered the ‘author’ of the resultant image.
One could argue that the user of a generative AI tool is in a similar position to that photographer – while they are not necessarily intimately familiar with the technical workings of the tool, they play a role in sculpting its output, potentially iterating on prompts to refine results. ‘Prompt crafting’ or ‘prompt engineering’ is already being touted as an up-and-coming discipline, with some claiming that there is a skill and technique in getting the best results from generative tools.
The key issue is whether a largely, if not entirely, automated and autonomous process can satisfy the element of ‘independent intellectual effort’, including whether the inputting of text descriptions or prompts can be considered an intellectual effort.
In the US at least, the US Copyright Office’s view seems to be that the inputting of text description and prompts is insufficient to satisfy the ‘traditional elements of authorship’. The US Copyright Office recently revoked artist Kristina Kashtonova’s copyright on comic images generated using Midjourney AI, concluding:
Though [Kashtanova] claims to have ‘guided’ the structure and content of each image, the process described in the [letter] makes clear that it was Midjourney—not Kashtanova—that originated the ‘traditional elements of authorship’ in the images.
The Copyright Office later released guidance clarifying the issue of human authorship:
AI technology receives solely a prompt from a human and produces complex written, visual, or musical works in response, the ‘traditional elements of authorship’ are determined and executed by the technology—not the human user. Based on the Office’s understanding of the generative AI technologies currently available, users do not exercise ultimate creative control over how such systems interpret prompts and generate material.
However, the Copyright Office also added that copyright may be accorded in some circumstances, for example, where a human ‘selects or arranges AI-generated material in a sufficiently creative way that the resulting work as a whole constitutes an original work of authorship’.
Unlike the US, copyright in Australia is not a registrable right. In the event of a dispute, the party claiming the subsistence of copyright bears the burden to demonstrate the requisite elements of copyright. We are not aware of this having been tested by an Australian Court, but we would be sceptical that the mere prompting of a generative AI tool would be enough human involvement in the creative process for copyright to be granted to the person using that tool.
There remains an open policy decision for lawmakers to consider about whether the law should change to provide the human users of these tools some degree of ownership of their output, given the current conundrum.
Programmer as the owner of copyright?
Another theory suggests the authorship of the works belongs to the programmer that developed the AI technology. But for the creation of that tool, none of the resultant works would exist, and all of the relevant creative decisions are being ‘made’ by the tool functioning in the manner dictated by its programming.
There is an argument that this is already the case under the copyright law in the UK. Section 9(3) of the Copyright, Designs and Patents Act 1988 (UK) (CDPA) states that the person ‘by whom the arrangements necessary for the creation of the work are undertaken’ is the author of a work that is computer-generated (that is, work generated by computer in circumstances such that there is no human author of the work).
Whilst it is arguable who the person making ‘the arrangements necessary for the creation of the work to be undertaken’ may be (ie the end user, programmer, or even engineer involved in the machine learning process), one Judge has suggested that as a general rule, ‘the owner of a thing is owner of the fruits of that thing’.5 This would suggest that the copyright owner of the AI technology is also generally the copyright owner of the output generated by that AI, but that does not necessarily reflect the legal position in Australia.
Copyright in the model?
Inarguably more valuable than the output of any generative AI system is the model itself.
Surprisingly, there might also be doubt as to whether that model is capable of being owned via copyright.
These generative AI systems tend to consist of a relatively concise piece of software, backed by a very large model, generated via a machine learning process. To use a perhaps overly simplistic analogy, the software effectively defines the way in which the ‘brain’ works, but the model is the knowledge and experience of that brain. This makes the model supremely valuable, particularly given the time and expense involved in the training process.
Notwithstanding that the model is plainly commercially valuable, it may not necessarily be able to be protected by copyright.
- As explained above, copyright is typically reserved for works created by human authors, which may be lacking in an automated machine learning process.
- Under the Copyright Act in Australia, copyright protects original works, and it is not clear that machine learning leveraging prior materials qualifies in this respect.
- As an even more fundamental issue, the Copyright Act provides protection for software, but it is up for debate whether these models themselves are software, or merely some form of data for use in conjunction with another piece of software.
A number of the major generative AI tools, like ChatGPT, are currently offered only as online services, rather than distributed as locally installable tools. No doubt some of this is driven by the technical demands of running those systems, but this also helps the creators of those models keep them confidential, and effectively protect them as trade secrets in the absence of certainty around copyright.
Conclusion
The application of traditional intellectual property rights to the output of generative AI is complex and unresolved. For now, users of generative AI tools should work from the starting assumption that they are unlikely to own the output created and take a cautious approach as a result.
How can HWL Ebsworth help?
HWL Ebsworth’s Technology and Intellectual Property team has extensive experience in advising businesses on intellectual property. If you are concerned about intellectual property infringements, please do not hesitate to contact us for further information on how we can assist you.
This article was written by Daniel Kiley, Partner and Paul Sigar, Solicitor.
1Commissioner of Patents v Thaler (2022) 401 ALR 551 [105].
2Ibid [103].
3IceTV Pty Ltd v Nine Network Australia Pty Ltd (2009) 254 ALR 386 [33].
4See Telstra Corporation Limited v Phone Directories Company Pty Ltd [2010] FCAFC 149.
5Thaler v Comptroller-General of Patents, Designs and Trade Marks [2020] EWHC 2412 (Pat) [49].
