LOVE, LAW & ASSETS
November 10, 2023
The introduction of ChatGPT has revolutionized the way people interact with computers forever. We are only at the very beginning of what is being called the next industrial revolution, and the AI hype can already be seen everywhere. As we navigate this new industrial era, the question of AI patentability arises. This article explores whether AI inventions can be patented and outlines the key requirements.
Table of contents
- South Africa’s Patent Law and AI Inventions
- Patentability of Computer Programs in South Africa
- Requirements for Patentability of AI Inventions
- Patentability of Automated Processes
- The Role of Hardware in AI Patentability.
- Technical Effect and Training Data for AI Patents
- Conclusion: Preparing for the Future of AI Patentability
- Ready to Navigate AI Patentability?
South Africa’s Patent Law and AI Inventions
South Africa is currently a non-examining jurisdiction which means that a complete patent application that fulfils all the formal requirements will be granted irrespective of whether the invention is patentable or not. This could inevitably lead to a situation where an invalid patent is granted. To ensure, as far as possible, that a patent is valid It is therefore good practice to follow decisions from South African courts and where no decision exists for a certain topic, to follow guidelines and case law developed in examining jurisdictions e.g. Europe.
Patentability of Computer Programs in South Africa
Section 25 of the South African Patent Act states that a program for a computer is not an invention for the purposes of the Act and can therefore not be patented. It is however possible to patent the method/steps employed by the computer program in what is referred to as a method claim provided the basic requirements of patentability are met.
Requirements for Patentability of AI Inventions
Any invention, to be patentable, must be novel and involve an inventive step. These requirements would also be applicable for computer programs. Novelty is judged on a worldwide basis. The novelty of an invention is destroyed if the invention has been disclosed in any form anywhere in the world before the filing date of a patent application. Inventiveness, a key element in AI patentability, means that the invention must not be obvious. A problem-solution approach is followed when determining the inventiveness of an invention i.e. does the invention solve a problem in a non-obvious way.
Patentability of Automated Processes
It should be noted, from an inventiveness point of view, that the automation of a manual process is not patentable e.g. if the method (the steps) that a computer program is employing is known, then the automation thereof will not be patentable, unless the automation produces an unexpected result
The Role of Hardware in AI Patentability.
European case law has determined that for a method claim, describing the steps which are implemented by a computer program, to be patentable, the “any hardware” approach must be followed. When this approach is used, the steps must refer to the hardware that is used when implementing a computer program. AI inventions typically involve the use of a Large Language Model that runs on a neural network by providing instructions in a human language to a computer which then implements the instructions by running code – this is the way ChatGPT functions.
The instructions must be given to the computer by means of a human-computer interface, and this interface together with the hardware required to implement the neural network should therefore be described for an AI invention to be patentable.
Technical Effect and Training Data for AI Patents
Furthermore, AI inventions can be patented provided that they have a technical effect that solves a technical problem across their full scope. Thus, for a neural network to be patentable the training data which produces the technical effect must be disclosed with enough technical information to enable replication, unless the skilled person can determine the training data without undue burden considering the state of the art. It is not necessary to disclose the actual training data – it is sufficient to specify the characteristics of the training data which affect the technical effect.
Conclusion: Preparing for the Future of AI Patentability
Furthermore, AI inventions can be patented provided that they have a technical effect that solves a technical problem across their full scope. Thus, for a neural network to be patentable the training data which produces the technical effect must be disclosed with enough technical information to enable replication, unless the skilled person can determine the training data without undue burden considering the state of the art. It is not necessary to disclose the actual training data – it is sufficient to specify the characteristics of the training data which affect the technical effect.
Ready to Navigate AI Patentability?
Contact us today to discuss how we can assist you in securing patents for your AI inventions or visit our services page to learn more about our expertise in intellectual property law.
