Ai - representation of human face and synaptic imagery on blue background

A Place for Patent Protection in Australia’s AI Ecosystem

The National Artificial Intelligence Centre, a branch of CSIRO, has recently released the AI Ecosystems Report 2023. The report highlights the key artificial intelligence areas that Australian innovators should pursue in order to become globally competitive, including the livestock, horticulture and medical technologies industries. Also highlighted is the need for inventors to pursue intellectual property protection regarding their AI inventions – patents in particular.

Australian AI Patents come from Overseas

The report analyses data regarding patented AI inventions in Australia from 2000-2023. The report reveals that the bulk of AI-based patent applications filed in Australia come from applicants based overseas. 32% of these applications come from just ten companies, with LG, Sony, iRobot, Accenture, and Apple rounding out the top five. Interestingly, the number of AI-based inventors residing in Australia is higher than that of Australian applicants, indicating that many Australian inventors are working for overseas-based companies. This may not be a surprise for those operating in tech industries, but it is an indicator that Australian companies could be more competitive when attracting local talent.

Research to Commercialisation Pipeline Blockage

In the first half of 2023, Australia contributed 1.6% of the world’s AI-based research publications, but just 0.24% of patent applications in the same area. While acknowledging that patents are not the only form of intellectual property protection available and that this is admittedly a narrow sample time, Australia lags in commercialising research output across the board. AI is no exception.

Potential Patentability Problem

In Australia, getting a patent granted for a computer-implemented invention is often challenging. Current Australian law holds that an invention that is implemented on a computer must consist of a ‘technical solution to a technical problem’. If said invention does not provide a tangible outcome (such as an improvement in the operation of the computer or some physical effect), then it is not patentable. This is a potential roadblock for CSIRO’s stated goals of increasing the rate of patent applications for AI technologies stemming from Australia due to the substantially computer-based nature of AI technologies.

However, a practical solution may already have been discovered, and it shows in the data. The OECD has identified a group of phrases relating to AI technologies. Of these, ‘robot’, ‘image processing’, ‘machine learning’, ‘neural network’ and ‘learning model’ are the most found in patent applications for AI technologies with an Australian inventor over the past 23 years. Other phrases with fewer entries are things like ‘computer vision’, ‘unmanned aerial vehicle’ or ‘autonomous vehicle’. It is not reasonable to assume that all these applications will become granted patents. However, it is still possible to learn from this observation. The connection is that most of these are technologies that have a tangible implementation. The AI is controlling a physical object or producing an altered state of affairs beyond the merely digital. The message is clear: to patent an AI technology, you should apply it to the real world.

News from the UK

The England and Wales High Court recently (21 November 2023) decided that an invention based on an artificial neural network (ANN) was a patentable invention in Emotional Perception AI Ltd v Comptroller-General of Patents, Designs and Trade Marks. The judgement compared the ANN, as might be implemented via hardware, to that exact implementation via software. Very briefly, an ANN is a prominent AI technology capable of learning. It works by learning what it should do from various sets of training data. Crucially, when the learning is complete, the ANN is ‘frozen’, and no more software manipulation occurs. The network could conceivably then be manufactured purely from electronic components and would not be ‘programmed’. Hence the ANN is not necessarily software. So, the judgement says, it is a patentable invention.

The law in the UK regarding the patentability of computer-implemented inventions differs from that in Australia. In the UK, legislation prohibits the patenting of a ‘computer program as such’, whereas in Australia, it is a matter of case law established by precedent. Nonetheless, the practical outcome is very similar. Australian courts do not typically follow precedent as set in the UK (any more) but nonetheless, there are enough similarities in the legal approaches surrounding patent law that this decision will be seriously considered should a dispute surrounding an ANN appear before an Australian court. The approach taken by the Judge in this case could well be applied in an Australian context, opening many additional paths to patentability for AI technologies.

The Future

AI technologies represent an exciting step forward. CSIRO wants Australia to cement its role in an AI-powered world, and a secure base of protected intellectual properties will help.

For additional information or advice relating to patents and AI inventions, please contact one of our experienced Attorneys.