Artificial intelligence (AI) is now a major driver of innovation in many sectors: healthcare, digital, food industry, automotive, aviation, etc. Protecting the intellectual property resulting from innovations using AI is a sensitive issue, particularly when it comes to patents.
In Europe and France, the rules governing the qualification of an innovation as an invention are strict and well-defined. In addition to the fact that, to be considered a patentable invention, an innovation must be new, inventive and susceptible of industrial application, it must not be included in the list of exclusions to patentability, which includes mathematical methods and computer programs. AI fits directly into this list, since it is based on algorithms and statistical models.
Exceptions to patentability have been introduced into European (article 52(2), (3) EPC) and French (L.611 - 10 CPI) legislation to protect the general interest. In particular, mathematical methods, considered as universal tools, should remain accessible to all. Also, general algorithmic concepts should not be allowed to become private monopolies. This framework also guarantees legal consistency in the handling of disputes.
However, as many technologies make use of mathematical methods or software, these exclusions are not absolute, but are directed towards innovations claimed as such.
It is precisely this distinction that has enabled case law to move towards relative recognition of the patentability of AI-based innovations.
Among these decisions, T208/84 (VICOM) concerned a method of applying convolution filters to an image, improving the sharpness and contrast of digital images. This image processing algorithm was recognized as patentable on the grounds that it produced a tangible technical effect. This decision marked a turning point in the assessment of the non-excludability of mathematical methods and computer programs. Indeed, it was concluded that, if a computer-implemented mathematical process or software program produces a tangible technical effect, then it makes a technical contribution to a technical problem and is no longer a mathematical process or software program as such. In other words, this decision formalized the fact that inventions involving algorithms or mathematical steps are assessed according to their technical contribution and no longer according to their software nature.
In case T1173/97 (IBM), the Board of Appeal examined whether a claim relating to "a computer program product" could be patented. It concluded that a computer program was not excluded from patentability as such if, when implemented on a computer, it produces an "additional technical effect", going beyond the usual physical interactions between software and a processor. With this decision, the Board of Appeal has changed the criteria for assessing exceptions to patentability by giving priority to the presence of an additional technical effect over the contribution of a claim to the state of the art. This decision has become established case law.
The T641/00 (COMVIK) case, which will be the subject of a separate article, supplements this decision by adding that a claim comprising technical and non-technical features (mathematical or algorithmic process steps) may be accepted if the technical features meet the criteria of novelty and inventive step.
More recently, case T2330/13 (SAP) validated a patent for artificial intelligence in the field of medical diagnostics. The patent application concerned a machine-learning algorithm used to analyze medical images and detect anomalies. During the examination procedure, the application was rejected on the grounds that the invention was based on a mathematical method, excluded from patentability as such. The Board of Appeal overturned this rejection, finding that the algorithm produced a tangible technical effect by improving the detection of medical anomalies, and that its application to the medical field made a specific contribution, distinguishing it from a simple mathematical algorithm. This judgment thus confirms that inventions based on artificial intelligence can escape the exclusion provided for in Article 52(3) EPC when they make a tangible technical improvement. This ruling now serves as the basis for examining the patentability of inventions involving artificial intelligence.
These developments have led the EPO to adjust its Examination Guidelines, which provide a better definition of the conditions for patentability of innovations implementing AI, so that they are no longer automatically considered as mathematical methods or computer programs as such. In particular, the EPO Guidelines provide a (non-exhaustive) list of examples of mathematical methods that do not fall within the scope of exclusions because they produce a technical effect. These include a method for enhancing or analyzing audio signals, digital images or videos, e.g. denoising, a method for separating speech signals and providing speech recognition or controlling a system or a specific technical process.
The EPO and WIPO have also entered a dialogue on the categorization of so-called "generative" AIs, which create images, text or other artificial intelligence. In a report published in 2023, WIPO states that "generative AI models, such as generative adversarial networks, are often excluded from patentability due to their purely algorithmic nature". Currently, AI innovators are encouraged protect their innovations through other means, such as copyright or trade secrets.
The above tends to show that the EPO is trying to adapt its practices to the emergence of these technologies. However, in comparison with US practice, which has become more flexible in recent years, EPO guidelines remain strict on the exclusion of computer programs and mathematical methods. In the USA, AI-based algorithms can be protected by patent if they provide a significant functional advance. In China too, the position of the CNIPA Office is to encourage the protection of inventions using artificial intelligence, including generative artificial intelligence.
This divergence between Europe, on the one hand, and China and the USA, on the other, could encourage innovators to file their patents outside European territory. This consequence can already be seen by comparing the number of patent applications for AI registered in these territories. This could encourage the EPO to progressively relax its rules, particularly if WIPO decides to harmonize the patentability criteria for AIs worldwide. It is to be expected, however, that this relaxation would probably only apply to certain specific technical fields, and that the criterion of tangible technical effect would still have to be met for inventions not to fall into the exclusionary categories of mathematical methods or computer programs as such.
Credit: Brunhilde GREIN
Comentários