The recent refusal of a patent application by PayPal Inc. at the Australian Patent Office sheds light on the challenges surrounding the patentability of AI and machine learning systems (PayPal Inc. [2023] APO 54). The rejected application, which proposed a system for generating more accurate recommendations using AI machine learning, faced scrutiny on the grounds that, while the combination of machine learning models was innovative, it did not offer a substantial technical contribution beyond standard computer usage.

PayPal’s patent application, filed in 2019, detailed a “system and method for obtaining recommendations using scalable cross-domain collaborative filtering.” The proposed invention aimed to enhance recommendation accuracy by incorporating three machine learning models. Firstly, an algorithm-based ML model determined a recommendation score based on user information, establishing a correlation between the user and other entities. Secondly, a cross-domain collaborative filtering ML model generated a recommendation score based on both user and cross-domain information. Finally, the invention applied the first and second scores to train a third ensemble ML model, producing a more accurate total recommendation score.

The rejection of the patent application highlights the ongoing challenge of determining the patentability of machine learning innovations. The Delegate, responsible for the decision, found that, despite the inventive combination of machine learning models, the proposed system did not offer a substantive technical contribution. The crux of the rejection rested on the Delegate’s finding that the invention primarily addressed a business problem, categorizing it more as a business innovation than a technical one.

Notably, the decision reflects the broader uncertainty in Australia concerning the patentability of computer-implemented methods and systems, especially those involving AI and machine learning. The Delegate’s decision quoted extensively from prior Federal Court cases emphasising that inventions must result in something tangible, rather than being purely abstract or conceptual. The key term, “manner of manufacture,” encapsulates the requirement that patents should be granted for practical, useful inventions with real-world applications.

This is another in a series of Australian cases, where decision makers have reinforced that business methods or schemes are potentially patentable, yet found the specific invention before them to be unpatentable. The complexity of machine learning arrangements, as evident in PayPal’s case, poses challenges in determining whether an invention is a scheme or plan, or an improvement in computer technology, a goal post which is challenging to identify in Australia, as highlighted by the 2022 split decision by the High Court in Aristocrat Technologies Australia Pty Ltd v Commissioner of Patents [2022] HCA 29. The Delegate’s decision, while relying on principles from prior cases, highlights the ongoing need for clearer guidance on what constitutes a patentable “manner of manufacture” for machine learning innovations. The delicate balance between technical innovation and business solutions remains a focal point in patent examination, necessitating continued refinement of guidelines to accommodate the dynamic nature of emerging technologies.