The challenges to patentability posed by Artificial Intelligence

As you probably know, unless you’ve just landed on this planet, Artificial Intelligence (AI) has become a buzzword and is omnipresent in the media.

Something similar has happened in the world of patents and industrial property, where all kinds of AI-related events, seminars, courses, etc. have taken place over the last year: European Patent Office, USPTO, AI-Decoding IP (UKIPO-WIPO), to name a few examples.

A new document on AI and patentability SCP/30/5 has been recently published by the Secretariat of the Standing Committee on the Law of Patents (SCP) following a proposal by the Spanish Delegation SCP/28/7. This document is especially recommended for those who are not experts in AI.

When talking about AI and the challenges it poses to patentability, it is possible to identify the following issues:

Subject matter eligible for patent protection.

The right to the patent in inventions generated autonomously by AI.

The prior art generated by AI.

The impact of AI on the assessment of inventive step.

Sufficiency of disclosure

The responsibility derived from the infringement of patent rights by AI.

I will try to provide a brief introduction to all those topics, though it would be possible to write quite a thick book on any of them.

Subject matter eligible for patent protection

When talking about AI-related inventions and patentability, three types of inventions can be distinguished: AI algorithms , use of AI for obtaining inventions and inventions autonomously generated by AI.

AI software or algorithms can be patented in a large number of countries and their patentability has been thoroughly studied. The patentability of these inventions is usually more strictly examined. In Europe, in order to be patented, they must have technical features, i.e. a technical solution to a technical problem. Inventive step will only be evaluated in relation to those technical characteristics.

The use of AI as a tool for obtaining inventions is well known, for example, in the screening and toxicity analysis stages of procedures for inventing new drugs. That should not pose any problems regarding the patentability of those inventions.

The last question with regard to patentability concerns inventions generated autonomously by AI. According to experts in the field, this is still science fiction, but it is something that is sure to happen in the following decades and something we must be prepared to address.

The right to the patent in inventions generated autonomously by AI.

Currently, most patent laws implicitly or explicitly establish that the inventor must be human.

In the future, companies should be ready to face the moment when AI is capable of generating inventions autonomously . There is a European Parliament resolution about the relationship between civil law, robotics and AI that recognises the need to adapt IP laws to current technological progress. It suggests creating the concept of “electronic personality” with its own rights and obligations.

Finally, another question is whether AI generated inventions should be patented. There are two opinions: some experts say yes, because that would encourage research into new AI algorithms and more powerful computers but others are not in favour of patenting them because inventions would be generated without being incentivised. Lincoln’s famous quote; “patents add the fuel of interest to the fire of genius” would cease to be valid since the fire of genius would no longer need fuel to be lit.

The prior art generated by AI

There are already “deep learning” machines that generate “prior art” which could be used at patent examination, oppositions, third party observations, judicial procedures, etc.

For the time being, the drawback is the lack of sufficiency of this “prior art”, but it will probably be solved sooner or later. We should also be ready to decide whether this “prior art” will be admitted and if not, how would it be possible to detect AI generated prior art?

The impact of AI on the assessment of inventive step.

Since inventions are bound to be generated autonomously by AI sometime in the future , it will be necessary to ponder whether the methods for the assessment of inventive step currently used such as the “Problem Solution Approach” will be valid or whether it will be necessary to elaborate new ones.

Legislators will have to work as well on the definition of the person skilled in the art. if it is an invention obtained through the use of AI, the hypothetical figure of the person skilled in the art will probably have to be equipped with that AI tool, whereas if the invention has been autonomously generated by AI, the person skilled in the art would probably have to be an AI algorithm.

Sufficiency of disclosure

The requirement of “sufficiency of disclosure” states that the patent document must describe the invention in such a way that it can be implemented by any person skilled in the art. Nevertheless, in some cases, especially when it comes to artificial neural networks, AI functions as a “black box” with input and output data with something in between very difficulty to describe.

Sometimes not even the inventors themselves know the details of what is going on. Perhaps, the definition of this requirement should be changed regarding AI inventions.

The responsibility derived from the infringement of patent rights by AI.

Who will be liable for the infringement of patent rights carried out autonomously by AI? The end user, the developer, the trainer, the algorithm itself? the most sensible answer would probably be the developer or the trainer.

In most patent laws, things cannot be liable for anything. The European Parliament resolution previously mentioned proposes the creation of a compulsory insurance system and an insurance compensation fund.

Patent professionals already have this as food for thought. I hope to return to this topic sometime soon.

Leopoldo Belda Soriano

In Spanish

Proofread by Ben Roadway