Patenting Artificial Intelligence

Artificial intelligence (AI) aims to automate and optimize human decision-making. Today, we interact with AI applications hundreds of times a day. These include, for example, automated recommendation systems, spam filters, speech recognition, and safety technology in the automotive sector. Most of these applications are based on machine learning methods, which typically use statistics to identify patterns in datasets and apply these patterns to new inputs for decision-making. Many AI applications in various technical fields take advantage of this, leading to the development of innovative interdisciplinary technologies and products. The goal of manufacturing companies is to achieve a strong competitive position through their innovations and to maintain it for as long as possible. However, strong protection is also essential for applied research to ensure successful technology transfer and commercialization. Therefore, it is desirable to protect innovations based on or utilizing AI through a legal monopoly.

A granted patent is a means of achieving a “return on investment” in research and development. It is an exclusive right, valid for a defined territory (for example, Germany, the U.S., or China) and for a limited duration, to defend the protected technology against unauthorized third parties. On the basis of a granted

A granted patent is a means of achieving a “return on investment” in research and development. It is an exclusive right, valid for a defined territory (such as Germany, the United States, or China) and for a limited duration, to defend the protected technology against unauthorized third parties. On the basis of a granted patent, the patent holder can take action against patent infringement and defend their technical monopoly against competitors.

“Patents shall be granted for inventions in all fields of technology, provided that they are new, involve an inventive step, and are industrially applicable,” according to Article 52(1) of the European Patent Convention (EPC). However, according to Article 52(2) and (3) of the EPC, mathematical methods, methods for performing mental acts, computer programs, and the presentation of information as such are excluded from patentability.

To overcome the aforementioned hurdles, an invention must be new and must inventively solve a technical problem based on a technical effect. An invention is considered inventive if those skilled in the relevant technical field do not regard the claimed invention as obvious. In the examination conducted by the Office, all information on the prior art published in the relevant technical field prior to the filing date of the patent application is taken into account. If, taking this prior art into account, those skilled in the art would not arrive at the proposed technical solution, a patent is granted for this invention.

AI as such is a method based on mathematical methods and models for mental activities, which are excluded from patentability. To overcome this exclusion from patentability and be recognized as “computer-implemented inventions” within the meaning of patent law, the AI must contribute to the technical solution. The AI itself must produce a technical effect without requiring further human interaction (decisions of the European Patent Office [EPO], including T 1035/18 and T 598/07). Although the accumulation of large datasets, among other factors, enables the development of algorithms and thus the integration of AI across all technical fields—including chemistry, pharmaceuticals, biotechnology, and medical technology—these inventions must

AI as such is a method based on mathematical methods and models for mental activities that are excluded from patentability. In order to overcome this exclusion from patentability and be recognized as “computer-implemented inventions” within the meaning of patent law, AI must contribute to a technical solution. AI itself must produce a technical effect without requiring further human interaction (decisions of the European Patent Office [EPO], including T 1035/18 and T 598/07). Although the accumulation of large datasets, among other factors, enables the development of algorithms and thus the integration of AI across all technical fields—including chemistry, pharmaceuticals, biotechnology, and medical technology—these inventions must be described in a clear, precise, and reproducible manner in order to obtain a granted patent (e.g., EPO, T 0161/18, T 1191/19).

In general, a distinction must be made between the AI-based inventions described above and AI-generated inventions. The latter are created de novo using AI; however, the inventive technical solution is functional even without the use of AI. An inventor is a person with legal capacity who can assert their right to be named as the inventor and to the patent. AI does not have legal capacity and therefore cannot be named as an inventor (e.g., EPO J 8/20; “DABUS” decision).

AI-supported predictive analytics are used particularly in the fields of biotechnology and biomedicine. For example, AI-optimized screening methods identify new proteins, antibodies, or receptors for therapeutic applications. The biomolecule subsequently synthesized is a tangible entity, such as a vaccine. This raises the question of who should be credited as the inventor of an AI-generated invention.

In practice, new challenges arise in determining which inventors contributed to the invention. This issue is particularly significant for employers who must compensate employee inventors based on their contribution to the development of the invention. This is because only those who have made a creative contribution to the invention are entitled to be named as inventors and are entitled to appropriate compensation. To date, there is no known decision by the Arbitration Board for Employee Inventions of the German Patent and Trademark Office or by a court of general jurisdiction in which the determination of the rightful inventors of an AI-generated invention has been discussed. However, in the future, more detailed questions—such as those regarding the identification of the technical problem, the respective creative contribution to the AI architecture, and the necessary training data and training methods—will need to be examined.