Innovations in artificial intelligence (AI), and more specifically machine learning, have been increasing at a rapid rate, both in Canada and internationally. AI has been behind the growth in a number of industries, ranging from online commerce to medical treatments to manufacturing.
Any company with technology related to AI would be wise to seek protection for AI innovations. The following are a few considerations when planning to patent an AI innovation.
Detectability refers to how easy is it to detect whether a patented technology is being copied. In order to enforce a patent, it is necessary to first detect whether someone is infringing the patent.
AI presents some challenges to detectability because often the “brains” behind the AI is a machine learning algorithm that might be thought of as a black box — input data goes in and predicted (or inferred) data comes out. It can be difficult if not impossible to detect whether one AI algorithm is copying another.
One possibility is to define the invention based on its inputs and outputs, with the AI algorithm being described at a higher level. For example, an AI innovation may be defined by the sensor data received into the AI algorithm and the predicted data generated, with the AI algorithm described at a high level. Detectability is then based on the inputs and outputs to the AI algorithm.
If an AI algorithm is implemented in such a way that infringement is very difficult or impossible to detect, it may be worth considering the possibility of maintaining the algorithm as a trade secret.
Which aspect to protect
Innovations in AI are not limited to the AI algorithm itself. In many cases, improvements in AI stem from the technology surrounding the core AI.
While the “brains” behind the AI might be treated as a black box, the usefulness and effectiveness of AI are often dependent on the quality of the input data, and how its outputs are used. There can be innovation in how data is gathered, generated or labeled; there can also be innovation in how predictions generated by AI are interpreted or fine-tuned by non-AI algorithms.
Typically, a technology is developed to address a specific problem in a specific context. This can be particularly true in the case of machine learning-based technology that is trained for a specific scenario. By examining the technology surrounding the machine learning core, it may be possible to identify innovations that can be more broadly applicable and hence more valuable to protect.
Patentability of computer-implemented inventions, including AI technology, continues to be a challenge in many jurisdictions.
The Canadian Intellectual Property Office (CIPO) and the United States Patent and Trademark Office (USPTO) are among the IP offices around the world that have set out specific guidelines for what is considered patentable subject matter, and how patentability is determined. While there can be differences among jurisdictions, general pointers can be found.
At issue is whether a patent is an attempt to protect subject matter that is excluded from patent protection. Describing an invention as purely an abstract idea, mathematical formula, or manipulation of data will likely result in rejection of a patent application.
Examples of how AI technology could be excluded from patentability include attempting to patent the AI algorithm by itself, or attempting to patent manipulation of data using a neural network without a tangible result. It can be easy to become focused on the algorithmic core of the technology, and lose sight of the technical implementation.
Oftentimes the outputs from AI are not used in their “raw” form, but is further interpreted and used as part of a larger system. In such cases, the AI algorithm can be patentable by claiming the algorithm integrated within the system. Drafting the patent application with this in mind can avoid the invention being categorized as unpatentable subject matter.
Guidance for what is considered patentable subject matter continues to evolve around with world, with some countries diverging and others converging in their approaches. However, most jurisdictions recognize the importance of AI in promoting innovation and are working to promote disclosure and protection of AI innovations.
The number of patent applications being filed for AI innovations is on an accelerating upward trend. With proper planning and forethought, a patent for AI technology can be a valuable asset.