Continuing rejection of computer implemented business methods in Australia - Paypal’s AI patent refused

IP Australia (the Australian Patent Office) has recently refused a patent application filed by PayPal directed at machine learning models that tailor recommendations to consumers at the point of completing an online purchase with the aim of boosting consumer purchases and donations.

Paypal argued that the invention was technical – in that it solved the problem of applying machine learning to generate more accurate recommendations for customers, and that it was not a mere business idea. However, the application was refused on the basis that the system was not a ‘manner of manufacture’ within the meaning of section 18(1) of the Patents Act 1990 (Cth).

The test for whether computer implemented inventions meet the ‘manner of manufacture’ requirement is in a continuing state of flux following the split 3-3 decision of the High Court in 2022 (Aristocrat Technologies Australia Pty Ltd v Commissioner of Patents [2022] HCA 29). The split decision of Australia’s highest court means that the Full Court’s decision in that case stands, and that a single judge of the Federal Court is bound to follow it (but this is not the case in relation to other Full Federal Courts that may consider this issue).

The starting point in Australia is that a business method, or mere scheme, is not, per se, patentable. Where there is a computer involved, the computer must be integral to the invention and not simply the mechanism by which the invention is performed (Commissioner of Patents v RPL Central Pty Ltd [2015] FCAFC 177).

In this decision, the Delegate agreed with the Examiner and found that the invention was not technical in nature. While it utilised machine learning models to determine a personalised recommendation to the customer, the Delegate said that the invention did not relate to an improvement in the machine learning techniques, but to known techniques, utilised in a conventional way, to operate on specific data sets to product the recommendations.

The PayPal system

Conventionally, to increase customer engagement, it was common at the priority date (December 2018) for additional purchase suggestions to be made to consumers when they were completing an online purchase. For example, a pop-up notification suggesting that the customer make a donation to a charity. However, these recommendations are not tailored to the specific consumer – but are made based on the popularity of the product or service being recommended.

PayPal’s system feeds information regarding a consumer who is completing a transaction into two machine learning models:

1. An algorithm based model that uses information regarding the consumer’s previous transactions; and
2. A cross-domain collaboration filtering model that uses information regarding the merchant and other entities (including charities).

The recommendation scores generated by these models are then used as inputs to train an ensemble model that calculates a total recommendation score. This score is unique to each consumer.

The system presents the recommendation generated by the combination of the three machine learning models to the consumer at the checkout.

Manner of manufacture requirements

In determining whether an invention is a mere scheme or business method, or potentially patentable subject matter, consideration will be made of…