Artificial Intelligence Patent 101
Who is creating patents and should you get one too?
Patent Publication Trends
Patent publications related to AI and machine learning have continued to increase at a dramatic rate, now reaching close to 100K applications a year at the US Patent Office. Only three years ago the rate was closer to 70K applications a year. Using patentguru.com, I reviewed the frequency of U.S. patent publications and patent application publications for the following terms in their title: “machine-learning” or “machine learning” or “data analysis” or “analytics” or “computer program product” or “artificial intelligence” or “data science” or “neural network.” The more general terms of “analytics” and “computer program product” were included as only the patent titles were reviewed not the body of the patent text.
It is possible a small amount of non-ML/AI patents are included but manual inspection of the resulting patents determined these search terms gave good results. The top assignee for the patents was IBM, followed by Qualcomm, Samsung, Intel, Microsoft, Google, Huawei, and Apple.
How are Patents Organized?
Patents need to be filed into a particular category. These categories were created a long time ago, before computer work products were something that needed a category. For this reason you will find most US Patent Office ML/AI type patents in category G: Physics, but they can be found in any category depending on the domain the invention is applied on.
A: Human Necessities
B: Performing Operations, Transporting
C: Chemistry, Metallurgy
D: Textiles, Paper
E: Fixed Constructions
F: Mechanical Engineering, Lighting, Heating, Weapons
G: Physics
H: Electricity
How to Get More Information On A Particular Class of Patents
Within each class there are subclasses, with G06F the most common for the US Patent Office ML/AI patents.
Using www.patbase.com, I reviewed information on the category G06N. This is a category where I recently filed a machine learning patent application. The results here are for all patents and applications in the class, not just the machine learning related ones.
The top assignee for the category of computer systems based on specific computational models is IBM, which has twice the number of patents as the next closest assignees Microsoft, Google, and Samsung. In this evaluation I did not limit to U.S. patents and so you will see the distribution of assignees globally.
Many of these applications are for Chinese patents, followed by the United States.
The number of patent applications in 2020 and 2021 for G06N, across all countries, is experiencing dramatic growth. This is just one of the multiple categories where machine learning and AI patents can be filed.
How do I write a patent of my very own?
You don’t. Your patent attorney does it for you. Only a patent attorney or a very dedicated inventor would attempt to write a patent. Your company can hire a patent attorney to help you, if they are dedicated to the concept of patenting their intellectual property and want to protect your project. The cost is not insignificant, and it is entirely possible that your company may prefer to keep your work as a “trade secret” rather than patent it. Assuming your patent is for work associated with an employer, your name will appear on the patent, but the “assignee” of the patent will be your employer.
Publication of your work, rather than pursuing a patent, is another path that can be taken. This may be preferable if the goal is to establish a reputation as an expert, as well as the dissemination of general knowledge, rather than specific protection of the work.
In order to proceed with patenting your work, you need to be able to demonstrate that your work is novel, non-obvious, and useful. You should create documents to explain your invention in layman’s terms. It should clearly explain your work, and what aspect of the work you are trying to protect. You are not able to patent abstract concepts, there must be a practical application to your invention. Your attorneys will be able to search for “prior art” to confirm your claims. While you can also search for “prior art”, keep in mind that you are required to disclose any findings that you have from such a search. When the patent is written, it will contain exactly that — “claims”. These claims describe exactly what your invention is composed of and are the most important part of the patent. They will also decide which class and subclass your patent should be submitted to, and the type of patent you will file.
It can be a bit disorienting to see your invention parsed into legalese that is difficult for even you to understand, and accompanied by unfamiliar diagrams. Apparently this is exactly what the patent office is looking for, so at this point you need to trust your attorneys. You also need to ensure that they have clearly understood your invention as they wrote it up and that the claims do in fact represent what you are trying to patent. Different claims can be associated with different inventors, so this is where you can designate which inventor contributed to the various aspects of the overall invention.
Can I make money from my patent?
Probably not. Also, your employer will probably not make any money from it either. Patents can be used to increase the intrinsic value of a company, and demonstrate their position as an innovative leader, even where no actual monetary value is associated with a specific patent. Also, the patent protects your product from being patented by someone else. Many patents are filed without a specific plan to monetize the invention being protected.
New AI Data From The USPTO
The US Patent Office has recently released a dataset of AI related patents. This dataset was compiled based upon a more comprehensive method of classifying patents than was used in this article. In an upcoming article I will discuss this new data and the related findings released by the patent office.
