28 October 2019

Will AI be recognised as an inventor?

The notion of whether or not artificial intelligence (AI) can be deemed eligible as an inventor or not has been challenged recently in the press. In this article for Robotics Law Journal, managing associate Andrew White, an expert in our specialist AI team, explores its implications for patent law.

An artificially intelligent system called DABUS (Device for the Autonomous Bootstrapping of Unified Sentience), developed as part of ‘The Artificial Inventor Project’, is challenging the notion of whether AI can be listed as an ‘inventor’ of patent applications. DABUS has been listed as inventor on two different patent families, pending at the United States Patent and Trademark Office (USPTO), European Patent  Office (EPO), UK Intellectual Property Office (UKIPO) and Patent Cooperation Treaty (PCT).

The AI system, described as a type of ‘Creativity Machine’ by its creator Dr Stephen Thaler, an AI expert based in Missouri, contains a first artificial neural network, made up of a series of smaller neural networks, that has been trained with general information from various knowledge domains, and a second ‘critic’ artificial neural network that acts as a monitor to identify new ideas that are different from its knowledge base.

Can AI be an inventor under current patent laws?

Under the UK Patents Act 1977 and the European Patent Convention, inventorship is restricted to ‘natural persons’. Likewise, in the US, patent laws refer only to ‘individuals’ as eligible inventors. The intention behind this statute was allegedly to stop companies or corporations being listed as inventors (but who can instead own the patent applications) and instead provide recognition to those who developed the inventive concept central to such patents.

In the present case, the Artificial Intelligence Project is keen to emphasise the distinction between inventorship and ownership – while an inventor may often be the owner of an invention, this may not always be the case. For example, employees of a company may have the right to be listed as an inventor, but the patent may belong to the company if the invention was made as part of the inventors’ normal course of duties as part of their employment.

The Artificial Intelligence Project states that its view is that machines should not own patents (but rather the owners of the machines should be the default owner), but that nevertheless the machine should be acknowledged as the inventor if it truly devised the inventive contribution. This is because AI does not have legal personality and, therefore, cannot own property.

What did the AI create?

The inventions contained in the patent applications are described by the Artificial Inventor Project as being the result of an extensive neural system that combines the memories of various learned elements into potential inventions that are then evaluated through the equivalent of affective responses. These affective responses effectively act to either amplify these concepts, or instead general new concepts by generating ‘noise’. The system was previously developed to create surreal art.

One of these inventions is a ‘neural flame’ device, relative to a light-emitting element that flashes at a prescribed frequency and fractal dimension. The other invention relates to a ‘fractal container’, which is a container for beverages that has a wall with a fractal profile which provides a series of fractal elements on the interior and exterior surfaces, forming pits and bulges in the profile which enables multiple containers to be coupled together by inter-engagement of pits and bulges on corresponding ones of the containers. It is claimed that the profile also improves grip, as well as heat transfer into and out of the container.

Is AI likely to be listed as inventors in these cases?

When assessing who should be listed as inventor, while the approach to assessing inventorship is a matter of national law that varies from jurisdiction to jurisdiction and is not entirely uniform, many jurisdictions require the inventive contribution to be assessed, and then a determination of who came up with the creative or intelligent conception of that inventive contribution – in other words, who came up with the real inventive ‘spark’, i.e. that which distinguishes the invention from that which came before it. This is consistent with the Oxford English Dictionary definition of ‘conception’, which is defined as ‘forming or devising an idea or plan in the mind’.

However the nature of the ‘invention’ is defined, what is generally required is some actual creative or intelligent contribution to the conception phase of the invention, which goes beyond the provision of abstract ideas and the mere execution of ideas on the other. It will be interesting to see where the ideas contained in these patent applications fall on this spectrum.

In the present case, at least, the patent applications are likely to struggle for being formally deficient – for example at the European Patent Office, the applications are likely to fall foul of Article 81 and Rule 19 of the European Patent Convention for not designating the inventor, and if not remedied, refused.

In addition, many people consider that AI is still so primitive at present that it is considered merely a ‘tool’ – it is written by humans, and trained by humans. As such, because it is just a tool, the user of that tool should be listed as the inventor. Many commentators, such as Dr Noam Shemtov, of Queen Mary University of London, who authored a study on inventorship in inventions involving AI activity commissioned by the European Patent Office in February 2019, consider that AI systems that have currently been developed have not yet progressed beyond this stage.

However, with the increasing speed at which AI is developing, it seems that even if in these two cases DABUS is not credited as being the inventor, sooner or later this question will have to be addressed by lawmakers around the world.

Furthermore, it should be noted that aside from the issue of whether or not AI may be listed as inventor in the present case, AI-related inventions (i.e. inventions that make use of AI, for example in drug discovery) are being filed at patent offices around the world at an increasing rate. Many patent offices, such as the European Patent Office, are keen to explain how such inventions are patentable, and recently held a conference in May 2018 on patenting AI inventions.

What are the implications if AI can be listed as an inventor?

Dr Shemtov hypothesises in his study that for an AI to be listed as an inventor under patent laws worldwide, this will require a change to the legal systems and legal frameworks that will involve the creation of a ‘legal or electronic personhood status’. This may be not dissimilar to a system relating to the legal personality of corporations. It may also require a fundamental reassessment of the notion of inventorship with respect to the chain of creation in that it can no longer be based on the idea of creative or intelligent conception, but one which could be described as a ‘functionalist’ approach that looks at the system’s output and contribution. 

Although the Artificial Inventor Project states that it considers AI should not own patents because AI does not have legal personality, if indeed AI can be held eligible as inventor, then the question that naturally follows is: why can’t AI then have legal personality also? This would lead to a far-reaching reassessment of ourselves as humans, and society’s relationship with technology. It would have implications for copyright law, civil liability and data protection, to name but a few. For example, with the advent of autonomous vehicles, it may have implications for the question of liability in an accident involving such autonomous vehicles.

Interestingly, the question of whether or not AI may have a legal personality is one that has already been considered by the EU Parliament, who considered that such an assessment of the nature of AI being considered as a legal person was a plausible future development, but one that is currently not warranted by current technical or legal considerations.

This article was originally published in Robotics Law Journal in October 2019.