Organizations and pioneers in the rapidly evolving AI field should carefully evaluate the patenting of AI models as a complex and vital aspect of their intellectual property strategy. The patentability of artificial intelligence (AI) models is dependent on many legal and technological criteria relevant to each country, which pertain to the model’s ingenuity, practicality, and novelty.

For an AI model to be considered innovative, it must not be a pre-existing work that is publicly available. Furthermore, it is crucial that it demonstrates a certain level of non-obviousness; it should not be a commonplace or expected answer among experts in the relevant field. As per the 2021 recommendations issued by the European Patent Office (EPO) on AI and machine learning, AI models that demonstrate technical characteristics or address technical problems may qualify for patent registration. For instance, the utilization of predictive algorithms in an artificial intelligence model to improve the energy efficiency of data centers might be considered as both novel and non-obvious.

When it comes to answering how to patent AI models, DeepMind, a subsidiary of Google, is a good case study to consider, wherein Google has acquired patents for numerous cutting-edge AI models, including those utilized in AlphaGo, an artificial intelligence specifically developed for gaming purposes. The patents underscore the technical solutions provided by the artificial intelligence (AI), specifically pertaining to the enhancement of power management in data centers. This exemplifies the successful resolution of a technical issue, hence enhancing the non-obviousness of the applications. The AI model must also demonstrate a clear “technical effect.” This means it should offer a tangible technical solution or improvement over existing technologies. For example, AI models that enhance image processing speeds by reducing computational load could be seen as having a significant technical effect, which makes them eligible for securing AI patent rights. IBM’s Watson Health provides an excellent example of technical effect, where AI models are used to analyze unstructured data in health records to provide personalized treatment recommendations. This not only improves the efficiency of data processing but also contributes to better health management outcomes, aligning with patent criteria under U.S. and European patent laws for answering all queries pertaining to the frequently asked question, how to patent AI inventions.

Patent applications for AI models must include sufficient detail that another expert in the field can replicate the invention without undue experimentation. This is a critical requirement to prevent the patenting of abstract ideas without practical applicability. According to a World Intellectual Property Organization (WIPO) report on intellectual property and artificial intelligence, detailed disclosure is paramount. The report highlights that failing to provide explicit implementation details can lead to the rejection of patent applications, as seen in some cases where generic claims about AI were not backed by specific algorithmic processes or system capabilities. Innovators should aim to clearly demonstrate the novelty, non-obviousness, and technical contribution of their AI models, ensuring detailed and sufficient disclosure in their patent applications. Engaging with case studies of successful patents and adhering to best practices, such as those from IBM and Google, can provide valuable insights into navigating the complexities of AI patents effectively. Furthermore, staying informed with ongoing regulatory developments is crucial to align patent strategies with broader legal and ethical standards.