In Australia, we are observing a noticeable increase in provisional patent filings by self-represented applicants who are filing applications that appear to have been generated with the assistance of AI tools. Some tell tale signs of this are applications with unusual and convoluted titles as well as individual applicants filing multiple applications in a rapid burst (e.g. 150+ applications over the span of a week).
AI is likely writing much of these specifications, what we do not know is the degree to which AI is also inventing. People are often surprised to learn that AI models are capable of inventing new things since by definition they are trained on pre-existing data which is usually obtained from publicly available sources. But these models have indeed shown a capability to invent (an early example was Dr Stephen Thaler’s AI system DABUS (Device for the Autonomous Bootstrapping of Unified Sentience).
The Thaler/DABUS Litigation
The question of AI inventorship reached Australian courts in a landmark case involving Dr Thaler and his DABUS system. Dr Thaler filed patent applications naming his AI system DABUS as the inventor, arguing that DABUS had autonomously generated the inventions, namely a food container constructed using fractal geometry and 'neural flame' for search-and-rescue operations.
In a world-first decision, Justice Beach of the Federal Court in Thaler v Commissioner of Patents [2021] FCA 879 initially ruled in July 2021 that an AI system could be named as an inventor, reasoning that an inventor is an agent noun (and an agent can be a person or a thing) and nothing in the Patents Act or Regulations precluded an AI 'inventor'.
However, this decision was overturned on appeal. In April 2022, the Full Court of the Federal Court ruled in favour of the Commissioner of Patents in Commissioner of Patents v Thaler [2022] FCAFC 62, unanimously agreeing that "the origin of entitlement to the grant of a patent lies in human endeavour". The Full Court concluded that, under the Act, the grant of a patent is premised upon an invention arising from the mind of a natural person(s) and that entitlement to a patent must ultimately derive from such an inventor.
The High Court of Australia rejected Dr Thaler's application for special leave to appeal in November 2022, ruling that it "is not the appropriate vehicle to consider the questions of principle sought to be agitated by the applicant". This ended the litigation in Australia.
The "Material Contribution" Requirement for Inventorship
Under Australian law, inventorship requires identifying the natural person or persons who made a material contribution to the inventive concept. In Polwood Pty Ltd v Foxworth Pty Ltd (2008) [2008] FCAFC 9, the Full Court set out what is effectively a two-part test for determining inventorship:
(1) Discern the inventive concept(s) (the “heart” of the invention) of a patent or patent application from the whole of the specification including the claims; and,
(2) Determine the person or persons who materially contributed to the inventive concept or concepts.
The criteria for inventorship includes identifying humans who have made a material or significant contribution to the invention, beyond merely assisting in the verification and reduction of the inventive concept into practice.
Section 7.2.8.5 of IP Australia’s Patent Manual of Practice and Procedure provides relevant guidance on non-exhaustive principles that may be useful in determining inventorship of an AI-assisted invention:
• The use of AI as a tool in the inventive process does not negate the otherwise material contribution of a person to an invention’s conception. Where AI is used as a significant aspect of the inventing process, but a human has a material effect on the final invention then there may be appropriate inventorship.
• A person merely identifying a problem to be solved, or research to be conducted does not necessarily rise to the level of a material contribution, unless invention lies in recognition of the problem to be solved. In this context merely feeding a known problem into an AI system may not give rise to inventorship. If a person constructs prompts in order to elicit a particular solution then a contribution may become sufficiently material.
• A person who takes output of an AI and analyses and refines it in view of a particular problem to devise a final inventive concept may be an inventor.
• Mere verification and reduction to practice of the output of an AI system is distinct from a material effect on the final invention. If an AI system could produce an output that may merely be appreciated as an invention by a person this appreciation would not appear to amount to a material effect. A final invention must be arrived at because of someone’s involvement for inventorship to be found.
• A general idea or foundational aspect from which the inventive concept is derived may amount to a material effect on the final invention. Building, designing, or training an AI system in view of specific problem in order to obtain a particular solution could lead to inventorship.
• Mere oversight or ownership of an AI system does not alone appear to amount to providing a material effect on any invention generated. This is analogous to mere oversight and ownership of any tool that may be useful in the inventive process.
Practical Takeaways for People Using AI to Invent
If you're using AI tools in your inventive process, here's what you need to know and do:
1. Document Everything
Establishing a material contribution to an invention requires documented evidence, including documentation that records each step in the inventive process from initial idea to concept including a description of the invention, the contribution made by each inventor, relevant dates and times and details of any AI contribution.
Keep detailed records of:
• The problem you identified and your approach to solving it - Document your analysis of the technical problem, why existing solutions were inadequate, and your hypothesis for a solution
• Your prompts and the conversation flow - Save complete chat transcripts showing how you directed the AI, what questions you asked, and how you iteratively refined your queries and directed the inventive process
• Your evaluation and selection process - Record which AI-generated suggestions you rejected and why, demonstrating your technical judgment rather than blindly implementing outputs
• Your modifications and refinements - Document how you adapted, combined, or improved the AI's outputs based on your domain expertise
• Your testing and validation - Keep records of experiments you designed to test AI-generated concepts
• Key technical decisions - Note where you made key decisions about implementation, design choices, or technical trade-offs
• Timestamps showing the development timeline - This helps establish the iterative, human-guided nature of the invention process
2. Understand the risks
AI tools available to the general consumer have come a long way since DABUS, and now sophisticated models are readily accessible to the public. Whilst modern AI tools can be of great assistance to engineers, designers and researchers in design and development processes, care must be taken to ensure that humans remain in the loop and are meaningfully involved in the invention process should patent protection be desired.
Whilst patent offices typically take inventorship and entitlement to patent grant at face value, meaning a patent may be accepted or granted even with incorrect entitlement, this is certain to become a fertile ground for invalidity attacks during opposition procedures or litigation in coming years. Applicants that have simply prompted a model to “design me a widget that does X” and then claimed the output of the model as an “invention” and named themselves as the inventor are going to end up in a precarious position.
3. Consider Trade Secret Protection as an Alternative
If you cannot demonstrate sufficient human contribution to satisfy patent inventorship requirements, trade secret protection may be a viable alternative. If there is a risk of not being able to establish material human contribution in AI-generated inventions for patent protection, then trade secret protection should be considered.
Unlike patents, trade secrets don't require you to identify an inventor or prove human contribution - you simply need to keep the information confidential and take reasonable measures to protect it. This can be particularly attractive for AI-generated innovations where the inventive process was heavily automated.
The key is making a strategic choice early: if patent protection seems uncertain, taking steps to protect your AI-assisted invention as a trade secret may be the more secure path forward.
Photo by Markus Spiske on Unsplash
.jpg)
