In its Patent Statistics and Analysis Q1 2022 report, GlobalData noted patent filings for AI-related inventions grew an average of 28% from the first quarter of 2018 to the first quarter of 2022, making it the fastest-growing tech area for global patent filings. The opportunities available with AI-related patents will only continue to grow given the ongoing advancements in the hardware and software powering AI and machine learning (ML) systems as well as the real-world applicability of AI systems for problem solving. For example, Moderna used AI to develop the first FDA-approved COVID-19 vaccine, and the company’s CEO credits AI with Moderna’s ability to develop the vaccine as quickly as it did.
Protecting your AI innovations with patents is one of the best ways to preserve and maximize the value of your software assets. As with all software patents, filing a patent for an AI innovation isn’t so straightforward. There are many considerations to take into account, including all the different claims you may want to include in your application to ensure maximum coverage, enforceability, and licensing value. It takes a very experienced software patent attorney, such as those at Blueshift IP, to comprehensively understand the patentability of artificial intelligence patent claims.
In this article, the attorneys at Blueshift IP identify seven components and processes you may want to claim in an artificial intelligence patent application, including some you may not have considered. If you have questions about patenting your AI innovations, do not hesitate to contact us here: https://blueshiftip.com/contact/.
Mixing and Matching to Achieve Greater Patent Protection
Our graphic below depicts the basic framework of a machine learning system. Based on what is valuable from a commercial perspective and patentable from a legal perspective, you may be able to mix and match components and processes to achieve greater patent protection.
1: Training Module/Process
The training module in the graphic above represents the method or process used to train a model, as highlighted in gray above. If you claim the model training process, a competitor will infringe on your claim if the competitor uses the same process as you to train their model. Patentability depends on what is novel (new) about the process you use to train your model and can be derived from:
- The algorithm used by the training module
- The training parameters
- The training data, or
- A combination of any of the above
If you’ve developed a new process to train the model and if that method is particularly novel or broadly applicable to many different models or types of data, a claim to protect that training module could be especially effective against competitors.
2: The Trained Model
You might not think to patent a model itself. However, if the model itself is novel, useful, and non-obvious, it makes sense to pursue protection for that model, at least in the United States. (Data structures may not be patentable outside the U.S.) This is especially true if there is commercial value in selling or licensing trained models independently of the process used to train those models.
3: Process for Executing the Model
Again, this might not be immediately apparent, but if your model generates something novel, useful, and non-obvious when it is executed, consider claiming the process for executing the model, regardless of whether the process for training the model itself is patentable.
4: Model Output, Either Physical or Digital
If executing the model generates output that is novel, useful, and non-obvious, it may be worthwhile to patent the output of the model. For example, if the model generates a design for a circuit, that circuit might be patentable independently of how it was designed. This type of patent claim can be very easy to overlook if you don’t pay close attention to it.
5: Process for Retraining/Updating the Model
From a licensing perspective, it’s worthwhile to consider claiming the process of retraining or updating the model based on new data. If you claim the retraining/updating process and licensees create new/updated models based on new data they input, you may have a claim to those additional models, and you may be able to realize revenue from those models as a result.
6: Process for Refining the Output (Manually and/or Automatically)
This part of the process may not seem critical, but it is worth considering, especially if the model output must be refined in order for it to be useful as an end product. For example, if the output is refined either by sorting or filtering the contents of the output in order to increase its quality and usability, the refining process could be patentable if it is novel and non-obvious.
7: Final (Refined) Output
As with the other components discussed, if the final, refined output generated by your model results in something novel, useful, and non-obvious, it may be worth patenting that final output, independently of the other components and processes in the framework that came before the final output.
Consult an Expert
Given the complexity of AI-related patent applications, we suggest you consult an experienced software IP attorney, such as those at Blueshift IP, to:
- Evaluate every invention on a case-by-case basis to determine all of the potentially patentable features
- Prioritize artificial intelligence patent claims based on their value
- Evaluate patent versus trade secret protection on a case-by-case basis
- Investigate non-obviousness carefully and build strong arguments for non-obviousness into your AI patent applications
- Investigate disclosure requirements, and disclose all necessary details in your AI patent applications from the beginning
The attorneys at Blueshift IP are well-versed in how to obtain AI patents that are strong, broad, and defensible, and which have maximum value for sale and licensing. If you are interested in learning more about patenting your AI innovations, reach out to the attorneys at Blueshift IP by filling out the form on our contact page: https://blueshiftip.com/contact/.