Clifford Chance

Artificial Intelligence (AI) is set to transform all aspects of our lives, including our home and health. AI is already widely used in Internet searches, and home devices with speech recognition, but in the near future we will see AI become even more widespread. This will have significant repercussions, as AI performs many tasks that, until now, could only be undertaken by humans. AI will remove human intervention from much of the picture. This will particularly affect intellectual property law.

With automatic execution when ordering products, the consumer is completely removed from the purchasing process. Consequently, this could impact trademark law, as visual, phonetic and conceptual comparisons between trademarks will no longer occur and cause confusion amongst consumers.

From a copyright standpoint, AI is able to generate and perform music or create artworks. Should traditional copyright requirements, such as originality, apply? Could AI be deemed as an author under copyright law?

Likewise, numerous questions are arising regarding the patentability of innovation in and arising from AI that have not been previously addressed. Can AI systems patent their inventions? How does the doctrine of equivalents apply to AI inventions? Will the 'person skilled in the art' change?

There is a vast array of issues related with seeking intellectual property protection for AI and machine learning systems. As French lawyers, it appears that the process of protecting AI is similar to making wine since a good wine requires quality of grape and time to mature. The same requirements apply to AI protection.

Fortunately, many tools, such as patents and copyright, are available to help companies protect their intellectual property rights (IP) in France. The French Patent and Trademark Office (INPI) has recognized the growing need to clarify the rules on how inventions related to and made by AI will be handled, and to determine what patent protection exists for this technology. The recent amendment of the Directives by the INPI is a clear sign that France does not want to lag in dealing with the rapid developments in AI.

One certainty: AI is entitled to IP rights in France.

Usually, AI encompasses computers and computerised devices that exhibit behaviours perceived as intelligent by humans, including learning, reasoning and decision-making. Other types of AI, such as machine learning systems, give the computer an ability to change behaviour according to experience.

In this section, we will focus on the IP rights that could be created and generated by Aaron, Emi (Experiments in Musicial Intelligence) and Adam, three manifestations of AI that are 'live' today. Aaron is an AI application that combines theories and techniques to generate computer programs or computational models; Emi and Adam are AI databases that gather and combine algorithms enabling machines or robots to reproduce a form of artificial intelligence.

When deciding to disclose AI and not choosing the option of trade secrets, how can the above-mentioned AI be protected under French law? Probably:

• Aaron's AI program could be protected by copyright law if legal requirements are met (e.g. originality of the program). It is based on an understanding of (i) the creative choices made early in the creation of the AI, and (ii) the selection made in what is proposed by the program.
• Emi and Adam's AI databases could be protected both by copyright law and sui generis database law. The creative choices made at the initiation of the database, i.e. the content that will define the AI, can be protected. In this case, what is protected is no longer the work itself but the conditions of creation of the work through AI.
• In some circumstances, AI could also be protected by a French patent using specific conditions, as was recently underlined by the INPI. The amended Directives detail what is considered technical or non-technical in regards to AI. There are several existing examples of other similar AI that could be protected under current patent law. These include: (i) a precise selection of datasets for learning; (ii) the architecture of neural networks used for a specific task; (iii) the management of memory during learning; (iv) recognition of the environment of an autonomous vehicle based on data obtained by using sensors; (v) image analysis in medicine; and (vi) the use of AI to analyse human language by a dedicated robot.

Challenges of the protection in France of artificially created works/inventions.

This section aims at evidencing solutions offered to companies to overcome the hurdles to, and obtain, IP protection in France.

AI and French Patent Law: an effective protection

Under French law, programs for computers shall not be regarded as inventions. Applicants will sometimes avoid mentioning AI in their patent application to avoid the exclusion of computer programs. However, the ethos is changing and innovation generated artificially by AI might be protected by a French patent when the following requirements are met: a patent shall be granted for any invention in all fields of technology, on the condition that it is new, involves an inventive step, is capable of industrial application and is not subject to prohibition of patentability:

• The debate surrounding AI inventorship is premature until the existence of an AI truly capable of an inventive act has been proved. However, we would point out that the European Patent Office (EPO) has refused two European patent applications, in both of which a machine was designated as inventor, on the grounds that they did not meet the requirement of the European Patent Convention that the inventor designated in the application has to be a human being, not a machine. Both patent applications indicated “DABUS” as inventor, which is described as “a type of connectionist artificial intelligence”. We can therefore imagine that the INPI will, for the time being, follow the same reasoning for Aaron, Emi and Adam.
• Nonetheless, with the prospect of AI contributing to invention, AI provides an opportunity for more collaborative innovation and more thorough patent filing strategies must be determined and implemented.
• Novelty might be an issue, since AI and machine learning will certainly use information previously obtained or disclosed to generate innovation. In other words, AI may improve but cannot change common general knowledge.
• If novelty is established, examiners will assess from April 2020 whether the claimed subject matter is inventive. To assess an inventive step, the directives revised by the INPI suggest that AI and machine learning inventions would be eligible for patentability if their application in a particular field contributes to the technical character of the claimed invention.
• Thus, inventions relating to AI and machine learning systems must demonstrate that the claimed subject matter serves a technical purpose. At a minimum, the claimed invention should provide a technical effect that is more than simply a way of achieving the solution to a problem more quickly. Achieving the technical effect should not be reliant solely on the AI or machine learning system; both are, arguably, well-known and have expected outcomes, at least in the sense of providing improvements. Nevertheless, where AI or machine learning systems are used, the application should comprehensively describe its detailed implementation, both functionally and structurally.
• The 'person skilled in the art' will be also a huge issue. At first glance, we assume that the INPI will consider that the person skilled in the art of AI and machine learning systems is likely to be multidisciplinary, and therefore involve scientists, engineers and/or computer scientists. Hence, the person skilled in the art is correspondingly likely to be a similarly composed team. The common general knowledge of the person skilled in the art in this field includes pre-processing data, setting parameters (such as for training and selecting validation data) and experimentation.
• The same reasoning could be followed for the sufficiency of disclosure allowing the reproduction of the invention. For a patent to be granted, there is no need to describe how Aaron, Emi or Adam arrived at this invention, or to explain their reasonings. To meet this requirement, it is sufficient to describe how the invention was conceptualized so that the person skilled in the art can reproduce it. These solutions may appear in the medical field, e.g. identification of new molecules that may have a therapeutic effect; in the mechanical field, e.g. identification of a particular profile of an aircraft wing with a strong lift or drag coefficient; or in any other technological field.
• When it comes to enforcement, evidence of the act or acts of infringement is clearly another challenge. Due to the nature of the innovation, it is difficult to demonstrate how exactly the AI worked and how it is reproduced by a competitor.

AI and French Copyright Law: a limited protection

AI can be engaged in a creative activity on two levels:

• Creation of a work undertaken both through human intervention and the assistance of AI.
• Creation of a work autonomously accomplished by AI, fed by millions of data and modelled on the works of human beings. We will focus on these artificially generated works.

Artificially created works could be protected, amongst others, by a copyright law. Under French law, copyright protects any work carrying "the imprint of the author's personality", from the moment of its creation, under the conditions that it is original and made material in any form (e.g. paper, digital media, etc.):

It can be any type of work, be it musical, literary, graphic, digital or plastic. Nowadays, we have tangible examples of paintings and music created by AI, such as The Next Rembrandt,  a painting made via a facial recognition algorithm which "learned" Rembrandt's techniques and was trained on Rembrandt's 346 known paintings, or original music composed by Emi.

• However, copyright requirements seem not to be suitable for AI as a 'creator'. It seems that textual reference to human creation, in the Berne Convention and national laws, might exclude the possibly of construing AI as an author under the current legal framework. More specifically, the essential requirement of "originality" is subjective, attached to the author's deliberate choices. Today's artificially generated creations seem much more the "random result of a set of algorithms". In any event, AI cannot express a personality. That is still the author who uses AI and who defines the framework within which the specific work will be created.
• As it stands, the status of author can only be attributed to a physical person (legal entities being unable to claim it). It is important to point out that French law does not recognize AI – such as Aaron, Emi and Adam – as a legal person. In the absence of property, they cannot receive the moral and economic rights arising from the IP protection of 'their' works.
• Leaving the domain of author's rights, it can be highlighted that AI-generated works may be eligible for protection through existing related rights, such as phonograph, audio-visual, communication companies, or database producer, even where there is no human intervention and as long as they meet the required criteria.

Key take-away points

• AI is everywhere but is only at the beginning of its development. From the music industry, through medical innovation and security/defence, to aeronautics, there is a real need for building trusted AI for industry. In order to avoid a "black box" AI legal regime, the criteria for protection must be precisely defined, both for AI instruments and for AI results.
• As of now, there is a legal framework under French law available to protect some inventions and/or creations generated by AI. It is therefore important to companies to determine the value of their IP assets. French patent and copyright law must evolve to catch up with the evolution of AI. Companies should keep in mind that technological advances are often years ahead of legislative change. As a result, when building a portfolio of AI patent assets, companies should take a strategic approach with the certainty that the current legislation will certainly change.
• Likewise, it is important to identify if the innovation relates to (i) core AI, where the challenge is that it could refer only to algorithms (mathematical methods), or (ii) AI as a tool with technical effects.
• A French patent lasts 20 years from its filing date, so that substantial changes may occur during that term of its enforceability. Consequently, it is advisable that companies have a long-term AI divisional filing strategy which would focus on protecting the implementation of the generic AI algorithm and its purpose.