Artificial intelligent patents and generative AI in the EU: false myths and real challenges.

This study will address the patenting of AI inventions by clustering them around three main groups. The author argues that we can indeed distinguish, on the one hand, inventions consisting in a technological refinement of existing AI methods or the design of new techniques in machine learning, deep learning, and related areas; and secondly, inventions that consist in the customization of general AI models in order to obtain an intelligent system with a specific application, typically in a different field (recall the example of the invention detecting ir-regular heartbeats and producing predictions regarding potential cardiac dis-ease risk). These first two categories of inventions can be analyzed together, as they share the feature that the claimed invention consists either of: (i) a new inferential technology capable of being applied across a plurality of intelligent systems, or (ii) an “intelligent” system or product endowed with inferential capabilities and valuable as such. By contrast, when the invention consists in the output produced by human–machine interaction, including those cases in which the result may appear to be solely the product of the AI system or product, the invention is not an “intelligent” product in the sense just described. It is a “traditional” product (e.g. a fractal container, a protein, a chemical compound or, more often, a new use of an old one, or a fractal container). The co-creation of a product with the assistance of the intelligent system gives rise to a distinct set of questions for patent law, primarily concerning inventorship and the right to file the patent application, as well as how to measure the inventive step (or non-obviousness) of an invention whose development was significantly accelerated by the employment of AI. These issues therefore will be dealt separately, as it is contended that many of the complexities regarding patentability of AI technologies, systems/products and AI by-products, like in the case of the DABUS’ fractal container, can be overcome if properly framed within the correct categorization proposed here: by keeping in mind whether the invention consists of an “intelligent” component or product or, rather, of an output created using such a tool.