21 July 2021

Could Australia Become the First Country to Recognise Non-Human Inventors?

Machine inventorOn 2 July 2021, a hearing took place at the Federal Court of Australia in Melbourne, before Justice Jonathan Beach, in the matter of Stephen Thaler v Commissioner of Patents.  This case concerns the question of whether a patent may be granted for an invention that was devised by a machine, rather than by a human inventor.  Back in February, I reported on the refusal by the Australian Patent Office to accept as valid an Australian patent application naming an ‘artificial intelligence’ going by the name DABUS (‘Device for the Autonomous Bootstrapping of Unified Sentience’) as inventor.  And in March I reported that an application had been filed in the Federal Court for review of the Patent Office decision.

Regular readers will know my position on this issue – I do not consider it appropriate at this time (or, potentially, ever) to grant patents for inventions devised entirely by automated means, such that there is no human inventor.  I have written an article targeted to a more general audience, which has been published by InnovationAus, providing an overview of the Australian case, and broadly discussing my concerns.  Here I will be going into more detail of the arguments presented at the recent hearing, and why I think it would be very unfortunate if Justice Beach were to decide that this is a suitable case for judicial development of the law to embrace machine inventors, as he is being encouraged to do by Thaler.

I was able to attend the hearing virtually, since it was being held via web conference.  Thaler’s team, led by experienced and highly-regarded barrister David Shavin QC, appeared in person in the Melbourne courtroom with Justice Beach, while the Commissioner of Patents was represented by Hamish Bevan, appearing via video from Sydney (subject to restrictions, due to an ongoing COVID outbreak).  Although I disagree with the proposition, I thought that Mr Shavin presented a persuasive argument that the relevant provisions of the Australian Patents Act 1990 can, and should, be interpreted to encompass non-human inventors, and that Mr Bevan perhaps did not do enough to counter this argument.  I formed the impression that Justice Beach just might be minded to ‘develop’ the Australian law to permit patent applications having no human inventor, in part because he was not presented with any particularly good reasons not to do so.

The Case for Machine Inventors

To understand the arguments for and against allowing a machine (or, indeed, any non-human entity) to be an inventor, it is necessary to consider subsection 15(1) of the Patents Act 1990, which provides that:

Subject to this Act, a patent for an invention may only be granted to a person who:
        (a) is the inventor; or
        (b) would, on grant of a patent for the invention, be entitled to have the patent assigned to the person; or
        (c) derives title to the invention from the inventor or a person mentioned in paragraph (b); or
        (d) is the legal representative of a deceased person mentioned in paragraph (a), (b) or (c).

Importantly, there is no definition of the word ‘inventor’ anywhere in the Act.  It was common ground at the hearing that, historically, an inventor has always been a natural person, and that there is no evidence, or other reason to believe, that anything else was contemplated by the parliament when section 15 was enacted.  Thaler thus needs to convince the court that it is now appropriate to interpret the word ‘inventor’ more broadly, such that it encompasses machines.

There were – as I understood it when observing the hearing – essentially three elements to the argument presented by Mr Shavin:

  1. establish that the word ‘inventor’ should not be limited to humans, despite the fact that this might have been true historically;
  2. show that there is nothing in the scheme or structure of the legislation that reasonably prevents a broader interpretation of the term ‘inventor’; and
  3. convince the court that there are no barriers – e.g. policy concerns – to it developing the law by ruling that ‘inventor’ should be construed to include machine inventors.

The Meaning of ‘Inventor’

Mr Shavin spent some time on the meaning of the word ‘inventor’, and made reference to dictionaries as evidence of the ways in which the English language evolves.  Most (perhaps all) dictionaries naturally assume that an ‘inventor’ is a person.  But this is not conclusive, because there are many examples of words that once referred only to people, but can now refer to machines.  One such example is ‘dishwasher’ (which can now be either a person or a machine).  But the example which appeared to most capture the attention of Justice Beach is ‘computer’.  This word once exclusively referred to people who performed calculations, and often to women such as were employed by NASA in the early days of the US space program – a historical fact in which Mr Shavin and Justice Beach seemed to share a common interest, judging by their exchange on the subject.  Now, of course, the word ‘computer’ refers almost exclusively to machines.  As Mr Shavin put it, words such as ‘computer’, ‘dishwasher’, and ‘inventor’ are ‘agent nouns’ that assume nothing about the nature of the entity to which they refer – which may therefore evolve over time.

The Scheme Under the Patents Act

Paragraph 15(1)(a) provides that a patent may be granted to ‘a person who … is the inventor’.  It was common ground that a machine is not a ‘person’ (legal or otherwise).  Arguably, then, the inventor must be a natural person (i.e. a human being).  This is essentially the view taken by the delegate in the Patent Office decision under review, and it was also substantially the conclusion reached by the High Court in England when considering similar provisions under the UK Patents Act.

In his submissions on behalf of Stephen Thaler, Mr Shavin agreed that for the purposes of paragraph 15(1)(a), i.e. where a patent is to be granted to the inventor, then the inventor must be a natural person.  However, he sought to break the link between the inventor, and the mechanisms for a patent to be granted to some other person or entity under either of paragraphs 15(1)(b) or (c).  In particular, he argued that the requirement for a person to be be entitled to have the patent assigned to them, or to derive title to the invention should not inherently require that the inventor be a person capable of owning and assigning (or otherwise transferring) property.  Rather, what is at the heart of these provisions is being in ‘possession of the invention’.  This involved much elaboration on the legal meaning of ‘possession’ (which itself produced some banter between Mr Shavin and Justice Beach over who owns the earliest edition of a particular historical legal text).

There is – in Mr Shavin’s submission – no requirement for title to ‘move’ from a human inventor to an applicant who may be granted a patent.  All that is required is ‘possession’ that has been ‘derived’ from an inventor.  In this case, it is said to be sufficient that Thaler is in legal possession of an invention, derived by virtue of his ownership of the inventor DABUS.  The fact that DABUS could not itself be granted a patent under paragraph 15(1)(a) is irrelevant.

Policy Considerations and the Role of the Court

Assuming that there is nothing in the legislation expressly preventing the court from finding that the term ‘inventor’ is not limited to natural persons, the question arises of whether the Act should now be construed to encompass machine inventors.  Mr Shavin submitted that Justice Beach should construe section 15 of the Act by reference to the purposes and objectives of the legislation.  He pointed out that the High Court of Australia has not shied away from developing the law with an eye on policy objectives, including in patent cases such as D'Arcy v Myriad Genetics Inc [2015] HCA 35 (in which the plurality rejected extending patent protection to isolated genes on the basis that, inter alia, it would not be consistent with the purposes of the Act) and Calidad Pty Ltd v Seiko Epson Corporation [2020] HCA 41 (in which the majority overturned a century of established case law to find that the first sale of a patented product exhausts the patentee’s rights, rather than creating an implied licence, observing that the ‘exhaustion doctrine has the virtues of logic, simplicity and coherence with legal principle’).  It was not lost on Justice Beach that Mr Shavin was involved in both of these high profile cases (or that he was on the losing side in Myriad).

The object of the Patents Act is now expressly set out in section 2A, and includes promoting ‘economic wellbeing through technological innovation and the transfer and dissemination of technology’.  Mr Shavin argued that this objective would be advanced by allowing patents on inventions made by machines, citing in particular the increasing role of software – including AI technologies – in the pharmaceutical industry, e.g. through the development of ‘in silico’ tools for drug discovery and biological modelling.  This appeared to be an area of interest to Justice Beach, who holds a bachelor degree in physical chemistry, and an MA in philosophy of science.

The Case Against Machine Inventors

On the face of it Mr Bevan, for the Commissioner, had an easier task than Mr Shavin.  He really only needed to make the case for maintaining the status quo – at least until such time as the parliament might decide otherwise.

There is an established expectation that the right to obtain a patent derives ultimately from one or more human inventors who were responsible for the original conception of the invention.  The Patents Act does not recite this explicitly, but over the years there have been many cases that have been concerned with what makes somebody an ‘inventor’, and the various legal means by which somebody other than an inventor can obtain the right to be granted a patent.  All of these cases are founded on the assumptions that an invention is a product of the human mind, and that title originates with the inventor and may subsequently be transferred by assignment or devolution by law.  No Australian legislator or court has ever seriously contemplated the possibility than an inventor might not be human. 

How hard could it be to make the argument that there is no basis for the court to upset the settled law and expectations at this time, by expanding the meaning ‘inventor’ to include machines?

Unsurprisingly, Mr Bevan’s submissions closely followed the established position and ran (as I understood them in observing the hearing) along the following lines. 

  1. The word ‘inventor’ is not defined in the Act, and thus has its ordinary meaning, which does not (currently, at least) include a machine.  The argument put forward for Thaler is based on no more than the proposition that ‘inventor’ could, at some future time, come to include a machine. 
  2. Section 15 presupposes the existence of ‘title’, as a prerequisite to establishing the entity that holds the title.  This is a recognised aspect of the law, whereby ownership can be legally transferred.  Entitlement to a patent under paragraph 15(1)(b) or (c) is ultimately derived from the inventor who is originally entitled under paragraph 15(1)(a), and who must therefore be a natural person.
  3. There is a distinction between the invention itself, and title to the invention, such that someone (such as the owner or operator of an AI machine) could have possession of the invention without also having possession of the title.
  4. Thaler’s submissions are premised on the idea that paragraphs 15(1)(b) and (c) do not require the inventor to possess and transfer title, and conflate possession of the invention with the inchoate rights contemplated by the Act.
  5. The object of the Act set out in section 2A is not new, but was intended to reflect the existing position.  The Act itself sets out the mechanisms by which the object is to be achieved, whereby an incentive or reward is provided to individuals in return for their industry and innovation.

This argument is, in my view, correct.  But it is somewhat dull, compared with the rather more exciting proposition of pushing the boundaries of the law to recognise an entire new category of automated inventors!  In my observation, Justice Beach seemed far less enthused by, and more critical of, Mr Bevan’s submissions than he was of Mr Shavin’s. Justice Beach appeared sympathetic to Thaler on the construction of section 15 of the Act, at one point criticising as a ‘logical fallacy’ Mr Bevan’s argument that entitlement under paragraph 15(1)(b) or (c) requires derivation from an inventor who would be entitled under paragraph 15(1)(a).

I also thought it unfortunate that the Commissioner did not more strongly counter Thaler’s contention that allowing patents on inventions made by machines would be a positive policy development that would serve to further the objectives of the Act.  There is no empirical evidence to support this proposition.  A single anecdotal example of the use of computers in the pharmaceutical industry does not provide a balanced or comprehensive view of the potential consequences of allowing machine inventors.

Serious Issues Raised by Allowing Machine Inventors

Currently, the notion that machines are actually capable of invention exists at the fringe.  Stephen Thaler is one of a tiny minority of developers dabbling in artificial intelligence claiming to have developed machines that can autonomously invent.  Notably, not one of the major industry or academic research and development teams that have collectively invested countless billions of dollars in AI and machine learning technologies has made any such claim.  On the contrary, in a submission responding to a USPTO request for comments on patenting artificial intelligence inventions, IBM stated that ‘[i]ntelligent machines will remain tools that assist humans, rather than invent independently, for a considerable time.’  And responding to a World Intellectual Property Organization (WIPO) consultation process on AI and IP policy, the IP Owners Association – whose members include Microsoft, Google, Apple, Oracle, and IBM – observed (diplomatically) that the assertion that ‘inventions can be autonomously generated by AI … is still subject to debate from a technical and legal standpoint.’

A recognition in Australian law that the term ‘inventor’ can encompass a machine would thus not only be well ahead of the dictionaries, it would also be ahead of any significant usage of the word in this way in society at large, or even among qualified experts in the field.

Allowing machine inventors could have numerous consequences, both foreseeable and unforeseeable, which were really not explored at the hearing.  Even if there are potential benefits in some fields (e.g. pharmaceuticals) – which is far from certain – there could be significant disadvantages in others.  As I have noted elsewhere, allowing patents for inventions churned out by tireless machines with virtually unlimited capacity, without the further exercise of any human ingenuity, judgment, or intellectual effort, may simply incentivise large corporations to build ‘patent thicket generators’ that could only serve to stifle, rather than encourage, innovation overall.  It is ironic that the objects clause in section 2A of the Act, added on a recommendation of the Productivity Commission – which was decidedly in favour of limiting, rather than expanding, patent rights – is now being deployed in a bid to open up Australia’s patent system to automated inventors.

The potential consequences of allowing machine inventors do not end with the entitlement to obtain a patent.  This could also present challenges for the substantive requirements for validity – most notably the requirement for an inventive step.  The established test requires an assessment of whether a claimed invention would be obvious to a ‘person skilled in the art’ in possession of the ‘common general knowledge’.  In essence, obviousness is assessed by comparing what the inventor has contributed to what a non-inventive person would have done in the same circumstances.  This test might cease to make sense if the inventor can be a machine.  Should the test be revised to encompass what machines are capable of doing?  Should both human and machine inventors be assessed against the same standard, or should there be different standards?  These are the kinds of questions that have been raised in consultations – such as the USPTO and WIPO processes mentioned above – and which currently have no clear answers.

Conclusion – Could the Court Favour DABUS?

The fact is that there is currently no real evidence for or against allowing patents on machine inventions, although it is not hard to devise scenarios pointing either way.  There are, however, certainly many open questions about the consequences of AI and machine learning for the patent system, on which discussion and research is only just getting started.  If a change is to be made, I do not believe that it should occur by judicial development of the law.  Serious consideration of all the issues is required, which is something that can only happen through consultation and government action.

On balance, however, I formed the impression that Justice Beach was quite sympathetic to, and even enthusiastic about, the case for allowing DABUS to be named as inventor.  He was not presented at the hearing with any submissions suggesting that this might be a contentious development requiring careful consideration more appropriate for legislative, rather than judicial, action.  I fear that this may lead to an unfortunate result.

To be honest, I was surprised that the Commissioner did not mount a stronger defence, given that this may be all that stands between us and a potential change to the law with unknown and untested consequences.  On the other hand, I imagine that the decision will be appealed regardless of the outcome, so perhaps the Commissioner is simply minimising the costs incurred at first instance.


Post a Comment

Copyright © 2014
Creative Commons License
The Patentology Blog by Dr Mark A Summerfield is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Australia License.