AI Pills – AI and Patent law

Patentability of A.I. and the role of machines in authorship are two of the main topics concerning artificial intelligence and patents

«As soon as it works, no one calls it A.I. anymore»


This is the clever way John McCarthy – who in 1955 first used the name artificial intelligence – expressed how the perception of progress is deeply inherent with the concept of A.I.

Indeed, since its first successful application, artificial intelligence has been one of the symbols of modernization and, as such, has crossed paths with patent law, making it yet another field in which the innovation brought by A.I. poses its main questions, some of which still far from being solved.

Patentability of A.I. and the role of machines in authorship are two of the main topics concerning A.I. and patents.

Patentability of A.I.

In Italy, as well as in the rest of the European Union, programs for computers are not regarded as patentable inventions. As a matter of fact, in our intellectual property system, original source code and object code are protected by copyright, as if the software was a work of the arts, while patents are set to protect novel and inventive products as well as inventive processes. This may seem to constitute an obstacle when applying for patent protection for A.I., which is – by its own nature – nothing more than lines of code.

However, when novel products are the result of an inventive process carried out on a computer using software (including A.I. software), such products or processes may be patented as “computer-implemented inventions“. Computer-implemented inventions cover claims which involve computers, computer networks or other programmable apparatus, whereby at least one feature is realised by means of a program (Guidelines for Examination in the EPO, F-IV, 3.9).

In Europe there is a rigorous approach to software patenting, consistent with the European Patent Convention, focused around the discussion on the technical character of the claims and based on an established practice for computer-implemented inventions.

In other words, while abstract machine learning algorithms, as such, are not patentable, patents may be granted for applications to solve a technical problem when they are adapted for a specific technical implementation.


A machine as author?

Whilst no problem nor concern seem to longer exist on the patentability of inventions in which computer programs including A.I. make a technical contribution (as long as they qualify as computer-implemented inventions), the role of A.I. in inventorship is currently under thorough debate.

In fact, when inventorship is in question, A.I. may have different roles, as it could (i) serve the only purpose of verifying the outcome of human-made inventions; (ii) be used to find a solution to a problem identified by a human inventor; (iii) autonomously identify a technical problem and an original solution.

Of course, the first and second scenario pose fewer question, as A.I. merely functions as a medium augmenting capability of human inventors. But what happens when artificial intelligence is not itself the subject of the patent nor a mere instrument to it, but rather the source of the so-called inventive step leading to the original solution to a technical problem?

This question can be summed up as whether a machine (i.e., the machine employing the A.I. software) can be considered the author of an invention.


DABUS: the only precedent.

The main patent offices of the western civilization (USPTO, EPO and UKIPO) had a chance to express their own position when between 2018 and 2019 Mr. Stephen Thaler applied for patent protection for two separate inventions (a beverage container and a device that can help with search and rescue emergencies), allegedly authored by a “Device for the Autonomous Bootstrapping of Unified Sentience” or “DABUS”, a system of neural networks created by Stephen Thaler himself.

The applicant firmly claimed that, while he had acquired the right to the patents from the “inventor” by being its successor in title, no natural person – not even him as the creator of DABUS – could claim authorship for the two inventions, as DABUS was the only entity responsible for the inventive step subject to the invention.

In their decisions handed down in 2019 and 2020, the European Patent Office, the US Patent and Trademark Office and the UK Intellectual Property Office rejected the applications, being very clear in stating that their respective patent laws do not allow a machine to be named as the inventor in a patent application, limiting such title to “natural” persons.

As of this date, the UK High Court confirmed the IPO’s decision, starting a reviewing process UK IPO’s Manual of Patent Practice, which has been updated in order to clarify that only natural persons are allowed to be identified as author of patented inventions. Appeals have also been filed before the EPO and the USPTO and their outcome is expected in by the beginning of next year.

The matter is far from being settled. However, we can say that the DABUS case opened a debate long overdue, as it posed a question which could not be further ignored: will patent practices and procedures worldwide adapt to a reality in which humans are no longer the only entity capable of promoting progress?