Using AI to invent therapeutics: should artificial intelligence be recognised for inventive activity? – Lexology

Original article was published on artificial intelligence

Two patent applications naming artificial intelligence as the sole inventor were refused by the European Patent Office following oral proceedings on the grounds that an inventor must be a human being. Is a machine capable of truly inventive activity? If so, what are the implications for the protection of the resulting technology?

Artificial intelligence (AI) can be broadly defined as the concept of a machine performing a task that is normally accepted as requiring human intelligence. AI algorithms ‘learn’ from data, information and even from their own decisions, and are capable of extracting concepts and relationships at high speed. AI is increasingly being incorporated into drug discovery pipelines.

The most common applications use deep-learning algorithms – similar to those used in face and image recognition – which are ‘trained’ using experimental results or information on the 3D structure and binding properties of small molecules to recognise target specificities with much greater accuracy than what was previously thought possible. Use of AI in the initial stage of drug development can increase the speed, accuracy and predictability of candidate selection. A slight increase in the reliability of predictions can potentially save vast amounts of money.

Incorporating AI into drug discovery pipelines

Biotechnology companies face numerous considerations when adopting AI in the development of therapeutics. These include access to relevant high-quality data, choice of algorithm, collaborations with other biotechnology companies for data sharing and tech companies for the use of existing AI technology, ethical and privacy issues accompanying the sharing of medical data, and the need for biologists and chemists with some understanding of AI for optimisation of protocols and the evaluation of results. Ownership and inventorship considerations are of particular importance.

Who is the inventor of an AI-generated invention?

When a molecule shows promise as a therapeutic early in the development process, the relevant company generally attempts to patent it to prevent others from copying it. Patent protection is crucial to investment in innovation. A patent application is required to name at least one inventor, but who is the inventor of a therapeutic that is ‘invented’ by AI?

One applicant deliberately tested this question by naming his patented machine, DABUS, as the inventor on patent applications for two separate inventions (EP 18 275 163 and EP 18 275 174). DABUS, or ‘device for the autonomous bootstrapping of unified sentience’ is described by its owner as a type of connectionist AI. DABUS uses artificial neural networks to generate and evaluate concepts, and reinforcement learning to learn from its own successes and failures. Using a separate neural network, the applicant claims that DABUS recognised the novelty and potential inventiveness of its creations, a food container and a pulsed light source for attracting attention, before a human did.

The designation of the inventor was questioned by the European Patent Office (EPO), and in his response the applicant quoted the UK Patents Act, which states that the inventor is the actual deviser of the invention. The applicant argued that it is dishonest not to disclose the actual deviser of the invention; failure to do so might constitute an offence in some jurisdictions; not accepting that AI could be an inventor precludes inventions created by AI from patentability; and no case law exists preventing AI from being named as an inventor.

The applications were refused following oral proceedings in November 2019 due to a failure to satisfy formal requirements; an application must state the family name, given names and full address of the inventor. The EPO argued that European patent law provides for natural and legal persons. An AI system has no rights or legal personality, meaning that it cannot transfer rights. Importantly, the EPO emphasised that the issue with naming AI as an inventor arises from a failure to meet formality requirements rather than any particular patentability issue. The EPO asserted that the concept of an inventor as a natural person was an internationally accepted standard, quoting the laws of several major jurisdictions.

Does the law need to change?

The requirement to name an inventor who is a ‘natural person’ was intended to acknowledge and protect the rights of human inventors, and originated before anyone envisaged machines capable of inventive activity. Inventorship and ownership are separate issues; the former must be determined in order to establish the latter. The inventor does not usually own the patent, but as AI is not a legal person and cannot transfer rights, companies should be mindful of potential challenges to the rights to inventions invented by AI.

Not all involvement of AI in technology development warrants inventorship for the machine. Applications in the biopharmaceutical industry often involve a machine outputting many potential candidates, which a human must then assess and evaluate. It may be relevant to determine who identified a problem to be solved – this could be a biologist, a programmer, or perhaps both. Was the problem already structured and understood by a human? AI can now autonomously generate new concepts, and evaluate their novelty and potential inventiveness. If a human cannot reasonably claim involvement, does naming a human as the inventor devalue genuine human invention?

One patentability requirement is that the invention would not have been ‘obvious’ to a person working in the relevant field before the patent was filed. Access to AI or the potential acknowledgement of AI as a hypothetical ‘person’ skilled in the field could lead to an argument for an increase in the threshold for inventiveness. Nonetheless, the availability of patent protection of AI-generated technology seems vital to incentivise innovation, promote the dissemination of information and enable the commercialisation of socially useful products.

What next?

International applications have been filed for the inventions generated by DABUS. It appears likely that these applications will be used to test the position of other patent offices, potentially prompting legal reviews. The EPO decision may also be appealed. Meanwhile, AI is revolutionising the creation of therapeutics, and its inventive capabilities are likely to increase. Inventorship and ownership of intellectual property should be discussed and preferably decided prior to AI-driven therapeutic development projects, while also looking out for relevant changes in patent law.

This article was originally published in the Australasian Biotechnology Journal in April 2020.