24 September 2021

The Ghost in the Machine

Regular readers of our biannual publication Inside IP will be aware of the efforts by a team led by Professor Ryan Abbott at the University of Surrey to obtain patent applications naming an AI system called DABUS (Device for the Autonomous Bootstrapping of Unified Sentience) as inventor. The inventions in question are a ‘Fractal Container’ and a ‘Neural Flame’. Applications have been filed in a number of jurisdictions, testing the response of the patent system to AI-created inventions.

Recent months have seen significant milestones in this process. In a judgment handed down on 21 September 2021, the UK Court of Appeal affirmed (albeit in a split decision) a lower court’s decision that the patent applications could not proceed as the requirements to name an inventor had not been met. Contrastingly, on 30 July 2021, a decision of an Australian court decided that an AI could be an inventor and that the applications should therefore progress.

The big questions, and the questions that matter

The concept of AI inventorship gets to a core issue about our relationship with technology and our understanding of ourselves. There is an association in the public mind between “invention” and moments of creative inspiration. If a machine can invent then are we saying something about its similarity to human minds? These are big questions, and fascinating ones. Moreover, the DABUS AI is itself an impressive piece of technology, running with a sufficient degree of autonomy that these questions come into view.

However, when it comes to the relationship between patent law and inventorship, it’s not altogether clear that we should expect answers to the big questions of consciousness and creativity. As the UK Court of Appeal notes, the phrase “first and true inventor” historically encompassed the first “importer”. Today, UK law refers to an inventor as the “actual deviser” of the invention. The concept of “invention” in patent law is not coterminous with the lightbulb-moment of popular imagination; the philosophical nature of inspiration and the role of the conscious mind can be left to one side for these purposes.

For all that, the technology does demand answers to questions that may have far-reaching effects on patent law. Can AI create an invention, and, if so, is that AI to be understood as an “inventor”? If such inventions are accepted to exist, who should own the patents to them?

Harmony and disharmony

Today’s international patent system is harmonised to an impressive degree; from the Paris Convention of 1883, through the TRIPS agreement of 1994 and with many other initiatives and agreements in between, the fundamentals of the patent system are remarkably consistent across jurisdictions. However, the contrast between the UK and Australian decisions (albeit that this may not be the end of the road in Australia at least) suggests that there may be further work to do if that consistency is to be maintained.

All three judges in the Court of Appeal found that according to the UK Patents Act an “inventor” must be a person, and as such cannot be an AI. This analysis predominantly focused on the wording of the relevant statutory provisions, which do not appear to contemplate or allow for a non-human “inventor”. This conclusion is not the same as excluding the possibility that AI might create an invention – it is simply to say that such an invention would have come into effect without an “inventor” as the law understands it. We return to the consequences for such inventor-less inventions below, as this is the heart of the split between the judges at the Court of Appeal.

Australian law is worded differently, and the Australian judge therefore found more room to consider whether an AI could in theory be an “inventor”. That analysis was intriguing, with the judge finding that while the term “inventor” has historically implied an individual, there is no reason that this meaning could not evolve over time to encompass machines capable of performing the task of invention. One might argue that there is a distinction between recognising that the meaning of a word may evolve, and actively introducing a new definition. However, the judge here was firm in the conclusion that AI can create inventions, and appeared to derive from this starting point a necessity to fit the law to allow such inventions to be patented – AI’s ability to be an “inventor” was almost a by-product of this.

Finders keepers

If both the UK and Australian courts accept at least the possibility of an AI-created invention, then what do we know about who might own a patent for such an invention? Here, one thing is clear: the owner will not be the AI itself. This creates a tension with the conventional understanding of the chain of title of patent ownership, which has often been thought of as some form of transfer from the inventor to the ultimate proprietor.

The split between the UK judges was on whether the absence of an inventor should necessarily matter. The majority view was that in the absence of an inventor a patent application will fail, if for no other reason than the technical ground that a patent applicant must declare the identity of the inventor and the manner in which the applicant obtained ownership. The minority view was that while an AI system cannot be truly an inventor, there is nothing to prevent an application proceeding without accurate inventor details and, further, that it is conceivable that the owner/operator of an AI system might obtain ownership in its creations. While couched in technical language, this divide may have significant consequences – as Lady Justice Elisabeth Laing put it in deciding with the majority:

If the inventor is not a person, but a machine, no person has, as a matter of law, a right to apply for a patent in respect of the machine’s inventions. It is as simple as that… If patents are to be granted in respect of inventions by machines, the 1977 Act will have to be amended.

This was the heart of the issue in Australia, too, where the judge felt that such an outcome would be an anathema to the “object clause” of the Australian patent act. That is, leaving such inventions unprotected was felt to go against the fundamental aims of the patent system; this was enough for the judge to justify the movement away from a dictionary definition of “inventor”. The Australian judge also felt that the inability of the inventor to own property did not necessarily prevent another party from deriving a right to the property as a result of their relationship with that inventor.

That is to say, while an AI cannot assign its property (because it has none), intellectual property arising from the AI’s activity could nevertheless be owned by another. The Australian judge saw no absolute requirement for formal assignment from one party to another for this to happen – arguing that ownership/control of the machine may be sufficient to claim ownership to that machine’s output. The minority opinion in the UK worked on similar lines – arguing that there might be rights that come from owning an AI system (and also that it was not the UKIPO’s place to judge whether this was true or not). This was a bridge too far for the majority at the UK Court of Appeal, who felt that the applicant had not shown any just reason in law why any rights arising from the AI’s activity would belong to them.

The object of the game

The patent system is designed to motivate and reward innovation. In those terms, is there a problem if AI-created inventions cannot be patented? Will this reduce the rate of human progress?

One might argue that without patent protection, there will be less incentive for AI-machines to be built and operated in such a way as to obtain these potential innovations. However, if the innovation is contingent upon this human endeavour, then perhaps the question falls away – those who put the conditions into effect for the AI to generate its output might be rightly considered the effective “inventors”, notwithstanding the autonomy in the AI process. It’s the humans that need to be motivated, not the machines.

However, if invention-creating AI is a technology with the potential to launch a step change in innovation, then the patent system ought to be ready to reward the innovation necessary in creating those machines in the first place. Ultimately, the DABUS case is about whether the children of DABUS are patentable, but perhaps we are trying to run before we can walk. DABUS itself is a technological marvel with a known, human inventor. Yet this does not fit easily with existing precedent at the UKIPO or EPO when assessing what is patentable – policy makers would be wise not to forget the importance of supporting innovation in AI itself as they worry about its creations.