The use of artificial intelligence (AI) has exploded rapidly in the last decade, particularly with respect to product development and data analytics.

In a recent test of Australia’s patent system, IP Australia decided that a patent cannot be granted where an AI system is identified as the inventor on the basis that it is not possible to identify the patentee for the purposes of section 15(1) of the Patents Act 1990 (Cth) (Patents Act).

While the decision is consistent with patent office decisions in other jurisdictions, it may be problematic since it has the potential to disrupt investment and stifle innovation due to the uncertainty about which, if any, intellectual property rights subsist in technology created by AI systems.

An application for judicial review of the decision was recently filed in the Federal Court.[1] However, the issue of inventorship will need to form part of a broader law reform conversation focussing on patents and copyright in particular. For now, innovators utilising AI systems will need to examine whether their current practices are sufficient to ensure that ‘their’ inventions are protected.

Dr Thaler’s application

In September 2019, AI pioneer Dr Stephen Thaler applied for an Australian patent titled ‘Food container and devices and methods of attracting enhanced attention’. The application named DABUS (Device for the Autonomous Bootstrapping of Unified Sentience), an AI system developed by Dr Thaler, as the inventor. Dr Thaler states that ‘DABUS is capable of devising inventions without the involvement of a natural person who traditionally qualifies as an inventor.’[2]

Section 15(1) of the Patents Act provides that a patent may only be granted to a person who:

  • is the inventor;
  • would, on the grant of the patent, be entitled to have the patent assigned to them;
  • derives title to the invention from the inventor or a person entitled to have the patent assigned to them; or
  • is the legal representative of a deceased person referred to in points one and two above.

Dr Thaler’s application was based on section 15(1)(c), as Dr Thaler stated that he derived title from DABUS on the basis that he ‘owns and created DABUS and is entitled to the output produced by DABUS.’[3] When asked for further information from IP Australia, Dr Thaler submitted that:

The common law confers ownership of anything produced by DABUS to its owner, the applicant. There is a general rule that the owner of a thing is the owner of the fruits of that thing, much like the owner of a fruit tree is entitled to the fruit produced by that tree. The principle of accession or first possession can apply.[4]

IP Australia’s decision

The Deputy Commissioner of Patents (Delegate) first looked at the question of whether an AI system can be named as an inventor for the purpose of the formalities of the application.

As acknowledged by the Delegate, neither the Patents Act nor the Patents Regulations 1991 (Cth) contain a definition of ‘inventor’. Given this, the question was not answered specifically in the decision, and so the Delegate looked to whether Dr Thaler could rely on section 15(1) of the Patents Act in order for Dr Thaler to be named as the patentee.

Section 15(1) requires the patentee to be a person, and the Delegate acknowledged that a person could be natural person, a body politic or corporate. Section 15(1)(a) provides that the patentee will be a person who is the inventor, so if the inventor is not a person then section 15(1)(a) will not apply.

Next, the Delegate looked at section 15(1)(b) which provides that the patentee must be a person who would be entitled to have the patent assigned to them. The Delegate concluded that section 15(1)(b) would not apply in the circumstances since Australian law does not recognise the capacity of AI to own or assign property.[5]

The Delegate reached the same conclusion in relation to section 15(1)(c), as that section requires the patentee to derive title ‘from the inventor’. Dr Thaler’s reliance on the concept of accession failed because, in relation to the doctrine of accession:

It seems that ownership automatically vests in the owner of the primary property by virtue of their ownership of that property (or possession of the new property) and is not regarded as transferred from another person. While the principle of accession (or possession) might well deal with the ownership of an invention created by an artificial intelligence machine, it does not do so by conceptually moving title "from" the artificial intelligence machine to the owner of the machine. It follows that the principle of accession (or possession) does not provide a means to identify a person who can be granted a patent that is consistent with section 15(1)(c).[6]

Given all of the above, the application lapsed for failing to meet the formality requirements of Regulation 3.2C(2)(aa), as it was not possible to identify a person who could be granted the patent.[7]

On 9 March 2021, Dr Thaler filed an application for judicial review of the Delegate’s decision in the Federal Court of Australia.

International approaches

The Delegate’s decision is consistent with those of the UK Intellectual Property Office,[8] European Patents Office[9] and US Patent and Trademark Office,[10] in relation to similar applications filed by Dr Thaler which name DABUS as the inventor.

In 2020, the UK Intellectual Property Office refused two applications by Dr Thaler in relation to DABUS on the basis that a natural person must be identified as the inventor. These decisions were upheld by the England & Wales High Court, which found that a patent can only be granted to a person, because only a person can hold property, and that DABUS is incapable of transferring property to Dr Thaler.[11]

The European Patents Office and US Patents Office refused applications by Dr Thaler on similar grounds, namely, that under their relevant statutory schemes an inventor must be a natural person.

Having regard to section 15 of the Patents Act, it would have been surprising had the Delegate reached a different conclusion in Australia.

WIPO’s Issues Paper

Dr Thaler’s application is not simply of academic interest. The patentability of AI-generated inventions is an emerging issue due to the increasing prevalence of AI in product development, for example in the pharmaceutical sector where it is being used to rapidly screen drug candidates and even design novel drugs and vaccines.

A range of patent issues relating to AI are considered by the World Intellectual Property Organisation (WIPO) in its Revised Issues Paper on Intellectual Property Policy and Artificial Intelligence, released in May 2020 (Revised Issues Paper). The Revised Issues Paper forms part of WIPO’s conversation on Intellectual Property and AI, and identifies the following key issues with respect to patentability of AI-generated inventions:

  • Do AI generated-inventions require patent protection?
  • Should the law require a human to be named as the inventor or permit an AI application to be named as the inventor?
  • Should AI applications only be permitted to be named as joint inventors alongside a human inventor?
  • Are any additional requirements necessary regarding the patentability of AI-generated inventions?
  • If AI-generated inventions are excluded from patent protection, what alternative protection mechanisms are available for such inventions?

In posing these questions, the WIPO makes an essential distinction between AI-generated inventions, which are created without human intervention, and AI-assisted outputs, which are generated with material human intervention and/or direction.

The Australian Government’s response to WIPO’s draft Issues Paper demonstrates that the Australian Government is aware of the importance of issues outlined in the draft Issues Paper, especially in light of the decisions with respect to DABUS. The Australian Government also raised a number of pertinent questions that it recommended WIPO include in the Revised Issues Paper, including:

  • If the law is amended to permit AI to be named as an inventor or owner, who is the beneficiary of royalty payments or commercialisation/licensing agreements with respect to the invention? Additionally, how would disputes over inventorship and ownership be resolved?
  • Does the prior art base include information used to train the AI?
  • Is any invention produced by an AI automatically obvious if the use of an AI would have been regarded as obvious to try?
  • Should there be a different test for inventive step when AI is involved?
  • For the purposes of disclosure, does the AI algorithm need to be disclosed?

WIPO is currently considering the views of member states in order to develop a set of ‘preliminary considerations’ to questions raised in respect of policy development and will publish these preliminary considerations in due course. WIPO also plans to hold the Fourth Session of the WIPO Conversation on IP and AI at a time to be scheduled during 2021.

These are only a small subset of the questions that governments are faced with as the adoption of AI in research and development continues to grow. Given the complexity of the issues, the need for international harmonisation and the substantial stakeholder consultation processes that will be required, it seems that any law reform in this area will take some time. Similar issues in respect of copyright, discussed in our September 2020 article referred to above, will necessarily be required to be included within the AI policy and law reform framework.

Looking ahead

The growing intersection between AI and intellectual property is raising a number of crucial issues for organisations seeking to protect and capitalise on their inventions. The recent line of international decisions in relation to AI inventorship reflects the fundamental complexities associated with determining the patentability and ownership of AI-generated inventions.

However, lawmakers around the world will be looking to adopt a harmonised approach to law reform and the extent to which their respective patent systems should accommodate emerging technologies such as AI. Until that time, innovators who are utilising, or planning to utilise, AI within their innovation toolkit need to carefully consider how best they can protect their intellectual property. This may include modifying their technology so that human intervention or direction is a material component of the invention, or ensuring that their technology (or how it works) remains secret and confidential.