Artificial Intelligence (or AI) raises many issues in relation to industrial property rights, both in the creation of a work or an invention, and in the protection of IP by IP rights.
Chapter 1. What is #8217; IA?
Section 1.1. Definition
But to speak well of the AI, it will have to explain a little what this is (and especially what it is not #8230;).
Artificial Intelligence is in fact only a computer program, no more, no less. But a computer program a bit special: a human has not specifically programmed its operation, but this program has been learned from #8221; its operation by analyzing a large number of learning data (we will explain it below).
You must not see a & ptc; #8220; intelligence & #8221; especially behind these AIs: AIs only do what #8217 they learned, without reflection.
Many people have sought to give a definition to RNs:
- AIs are #8220;computer programs that perform tasks that are, for the time being, more satisfactorily performed by human beings, because they require high-level mental processes such as: perceptual learning, organization of memory and critical reasoning” (Marvin Lee Minsky) ;
- AIs are #8220;computer programs to which tasks are assigned to which even the human being can not really define the underlying rules& #8221; (Me)
Section 1.2. Some examples of #8217; IA
IA includes a very large number of concepts that I have tried to group in the following schema (which is clearly not complete):
All this to say that #8217; we should not mix everything: l & #8217; IA this n & #8217; is not deep-learning and vice versa & #8230;
Section 1.3. How some AI work
1.3.1. How the regression methods work
Regression methods are mainly used to try to determine relationships between variables.
For example, suppose my variables are:
- l & #8217; age of #8217; one person;
- the percentage of his white hair.
In this very simple example, just take a large number of people, analyze the hair and note their age and we will have a curve of this type.
Of course, this is not enough: you only have a cloud of points. Even if a human being sees globally the relationship that emerges, we must not forget that we can generalize the problem to 800 variables and, in this situation, it is a little more complex to visualize #8230;
Therefore, we use so-called regression methods (linear, polynomial, logarithmic, etc.) in order to determine the curve or the surface that best fits the scatterplot.
In the end, the red curve will give us the relation between the different variables (and even the uncertainty about this relation, c & #8217; is magic).
1.3.2. How does a network of neurons work?
The underlying idea of #8217 is a neural network of & lt; TTP220; mimer & #8221; the functioning of our neurons, but in their most basic operation.
In a brain, each neuron has an axon that acts like an electrical wire, driving nerve impulses (in the form of an action potential) to the neighboring neuron, thus ensuring the functional activity of the brain.
Nevertheless, if the sum of the nerve impulses arriving on the axon does not exceed a certain value (threshold of excitability of the neuron), the axon does not relay the nervous message: we thus have to deal with a thresholding mechanism.
Computer scientists did exactly the same thing, but replacing neurons in the brain with logical blocks & #8221;
Thus, each neuron transmits values to its neighbors. The neuron receiving values will then sum these by weighting them with weights depending on the links (w1, w2, etc.). This sum will then be thresholded to the #8217; help of #8217; a threshold function (as here the sigma function).
Of course, we do not have 3 neurons, but much more in practice. There are many neural network architectures, but here is what an implementation might look like (5 input parameters and 1 output parameter).
All #8220; l & #8217; intelligence & #8221; contained in the neural networks lies in the good determination of said weights. Thus, these weights will be determined by attempting to solve the following optimization problem: Given a large number of input and output data, what are the weight values for maximizing the response of the network #8221;
To solve this problem of optimization, the weights are often initialized in some way and we will try to modify them gradually to try to optimize the output.
As long as the result of the network is not that expected (or its error rate is not less than a predetermined threshold), one continues to look for values for the weights.
(Yes, I know & #8230; my example is formally simplistic and therefore wrong for purists, but that's just to explain).
1.3.3. Operation of #8217; a neural network of type & #8220; Deep-learning & #8221;
In reality, a network of type & #8220; Deep-learning & #8221; is very close conceptually to a network of neurons, but the number of layers is much higher.
It is important to understand that networks of type & #8220; Deep-learning & #8221; have a very large number of weights or parameters, and it can be very complex to make them converge during learning (ie solving the problem of #8217; optimization can be difficult).
Networks of type & #8220; Deep Learning & #8221; have emerged in recent years, however, as computing power has increased significantly in recent years, especially with the use of graphics cards.
Chapter 2. Technical Achievements in Relation to the #8217; IA
Section 2.1. Patentability
2.1.1. The patentability of inventions made by IA
It can happen that 1P38217; IA & #8220; finds #8221; technical solutions to problems that the man has been laying for a long time.
These solutions may be in the medical field (eg identification of new molecules that may have a therapeutic effect) in the field of mechanics (eg identification of a specific profile of a wing with a strong lift) or in n & #8217 any other technological field.
2) Problem with the notion of "invention"
The question that one can immediately ask is: Is an invention made by a computer an invention?
Fortunately, the #8217;A52 (1) EPC proposes a definition of the invention:
European patents are issued for any invention in any technological field, provided that it is new, that it involves an inventive step and that it is susceptible of industrial application.
Thus, we saw clearly that there is no difficulty: invention is defined by its application (ie new and inventive thing in a technological field) and not by its origin or genesis.
Therefore, an invention can very well be described as #8217; & #8220; invention & #8221; if its application is technological.
3) Problem with respect to IPT, inventive step
Some were able to say that there was no inventive step for an invention made from AI because the effort of the inventor was nil.
I can only disagree.
Indeed, the effort of the inventor or the difficulty that he has had in inventing are not relevant criteria for assessing the inventive step: the requirement of IPPT217, inventive step is based on the difficulty for a person skilled in the art. to arrive at the invention from the documents of the prior art and not about the difficulty which the applicant has encountered in practice the inventor.
For example, it would not come to anyone's mind to say that an inventor can not protect his invention because he found it by chance (and this happens quite often).
Of course, it would be possible to argue that the person skilled in the art is an AI (or a man using AI) but this approach would inevitably lead to the conclusion that all inventions are obvious (see & #8220; Everything is obvious & #8221; from Ryan Abbott).
4) Problem regarding the sufficiency of description
Some have argued that an invention made by an AI would be insufficiently described because the inventive process that led to the invention is not known.
C & #8217; is, it is true, a problem of AI: most often, they give a result, but struggle to explain why such a result is given.
I definitely disagree with this position because the sufficiency of description (A83 CBEis not intended to describe how the invention has been conceptualized, but is intended to ensure that a third party, in light of the description, can achieve the invention.
Therefore, there is no need to know why the invention works. It is sufficient for the person skilled in the art to verify that the invention works by carrying it out.
5) Problem with the notion of inventor
CBE n & #8217 has no definition of the notion of inventor.
However, the generally accepted interpretation is that the inventor, within the meaning of #8217;A60 (1) EPC, is a natural person & #8220; or a human (ie natural person).
Indeed, how to accept that & #8220; rights & #8221; The invention belongs to a machine or an algorithm, since no national law (ie, member states) provides a property attached to an entity other than a human.
But is it a real problem?
First and foremost, the EPC only requires that #8217 be designated an inventor. There is no penalty if the inventor is not the right one or it does not exist (see the instructions below). filing requirements in Europe).
But in truth, this is not even the subject for me: the real question is what is the inventive act and who is doing it.
Indeed, it is quite clear that an AI is no more than a computer program helping a human. The human is going to choose how to use the IA, will choose the dataset to train it, will look at the output data to assess their relevance and applicability to the problem it is facing.
Therefore, there is not really any difference with the use of this human of a simulator / calculator that will help this person. It would not be in the minds of anyone to say that the inventor did not realize the invention because he used a computer to carry out his invention.
Thus, it is always possible to argue that the inventor is the person handling the IP in order to obtain the desired result.
6) Problem regarding ownership
With respect to the ownership of the invention, we may ask whether the holder of the IPA may file a patent application for an invention exited from the IPT.
In my opinion, and in European law, this poses little difficulty because theA60 (1) EPC provides that the invention belongs to the inventor or to his successor.
For all that (and even if we were to consider that the inventor is really a machine), the lack of the right to file a patent (see inventions filed by an unauthorized person) can only be invoked by the real owner of the rights.
So, to put it simply, no one can question a patent on the grounds that the holder of the title l & #8217; would have stolen from a machine & #8230;
7) Some examples
In order to show you that this is not a purely theoretical subject, we have recently seen several patent applications filed with EPO for the purpose of inventions made by an AI.
Some of these inventions were made by DABUS (which itself is filed at #8230; EP2360629 (A3)).
The first invention aims a container with fractal walls and to secure two containers in a simple manner.
The second invention is aimed at an alarm system (via the flashing of a diode) having a fractal repeat sequence and thus allowing better recognition by a human eye.
In another technical area, we can evoke this Gillette invention (EP1284621B1) concerning toothbrushes whose #8220; tufts & #8221; s & #8217; intertwine in a particular way. And this, thanks to the #8220;Creativity Machine& #8221; from S. Thaler & #8230;
2.1.2. The patentability of inventions implementing an AI
Here, we are in the case where the invention is really in the implementation of an AI (eg the recognition of an image with an IA).
The inventive concept can be located in several places:
- in the particular selection of the dataset for #8217;
- in the architecture of the neural networks used for a specific task,
- in memory management when learning,
For this very different subject, I refer you to my article on mixed type inventions in Europe.
Section 2.2. Protecting AI models
We have seen that inventions made by an AI or implementing an AI could be protected by patent law.
However, IA involves other entities such as the model (eg the configuration of neural networks).
The AI model is often very complex to obtain, because it requires a very particular selection of training data and requires a lot of effort (eg in terms of computing power or in terms of human effort to intelligently format the data. input data).
So, how to protect them?
In my opinion, there are two tracks:
- the protection of computer programs by the right of the author;
- the protection of databases by an ad-hoc right.
Indeed, the fact that the #8217; IA remains a program of #8217; computer motivates the #8217; analogy possible with the programs of #8217; computer (ie software). What looks most like the #8217; 1P38217 training; an AI is the compilation of source code in a compiled code: l & #8217; 1P38217 training; an AI is a kind of compilation of #8217; a system whose purpose is to make this system conform to expectations of the user.
Therefore, it is possible to consider that the IPA model is protected by the copyright law provisions of the software author (1).L112-2 CPI).
I remember the definition The software given by the French Academy in its dictionary (9th edition) is & #8220;Structured set of programs performing a specific function, allowing the accomplishment of a given task& #8220 ;. L & #8217; IA seems to fit well in this definition.
But it is also possible to make an analogy with the databases.
Indeed, as I said earlier, the model of an AI is a set of configuration / weight / etc. In other words, c & #8217; is a kind of configuration parameter database.
According to this directive (and its interpretation by the CJEU), a database must have the following characteristics:
I think that it is possible to consider that the parameters of a model have all of these characteristics (the weights have a meaning even taken individually, the weights are arranged in such a way as to know which nodes (or links) they belong to). apply, it is possible to navigate in the model to know the parameters).
The difficulty remains the protection of this database. Indeed, Article 7 of the Directive 96/9 / EC provides that #8217 protection of the contents of the database can be granted to #8220;when obtaining, verifying or presenting such content attest to a substantial investment from a qualitative or quantitative point of view“.
By obtaining, we speak not of the creation of the data but of its acquisition. It must be recognized that in the model training, what is complicated is the creation of the data and not the acquisition of #8221;
Nevertheless, if I had to argue, I will say that the training of the model is a kind of & ptc; #8220; verification & #8221; the validity of the values of the configuration weights & #8230; Therefore, the protection of Article 7 of the Directive 96/9 / EC s & #8217; applies t & #8217; it?
Of course, I have no answer about the protection applicable to the model. The previous elements are only tracks. It will be necessary to wait for the case law.
Chapter 3. Works Generated by AIs
Section 3.1. Introduction
I am not a specialist in the protection of works by the right of the author, but it is true that it would be a pity not to mention this point.
So I'm sorry to the experts of the subject about my approximations.
Section 3.2. Some examples
Today, there are a number of works from #8217; art & #8221; produced from #8217; IA:
But we also have music generated by artificial intelligence (example of #8217; piece généré par Spotify Research) :
Section 3.3. Protection by right of the author?
Une fois que nous avons vu ces exemples, il est possible de se demander si ces “œuvres” peuvent bénéficier d’une protection par droit d’auteur.
En droit français, pour bénéficier d’un droit d’auteur (L111-1 CPI), il est nécessaire de vérifier que l’œuvre est “original” (critère jurisprudentiel).
En gros, pour être original, une œuvre doit comporter :
- l’empreinte de la personnalité de l’auteur,
- la marque de l’apport intellectuel de l’auteur et
- l’expression des choix libres et créatifs de l’auteur.
A mon sens, pour les exemples que l’on a vu précédemment, cela ne pose pas vraiment de question : en effet, ces exemples ont été créées par des IA qui ont été configurées (i.e. qui ont subi un apprentissage) précis afin d’obtenir un résultat voulu par l’auteur.
Dès lors, les exemples précédents marquent clairement la volonté de l’auteur d’obtenir un tel résultat.
Nous pouvons considérer que l’IA joue un rôle similaire à un logiciel comme Photoshop ou comme une table de mixage. L’IA simplifie simplement le processus de création.
Bien entendu, il faut regarder au cas par cas et je ne saurais faire une réponse unique de type “toute image ou son sortant d’une IA bénéficie d’un droit d’auteur”, mais l’usage d’une IA ne me semble pas exclure, de ce simple fait, l’application du droit d’auteur.
Pour conclure, je voudrais simplement rappeler que, lors de l’invention de la photographie, certains voulaient exclure celle-ci de “l’art”. En effet, ils considéraient que la photographie appartenait au domaine technique.
Ne pensez-vous pas que nous sommes dans la même situation concernant l’IA ?