Have you developed a new software code and are wondering if you can patent an algorithm? This is a common question with machine learning developers, as they often want to know if their machine learning algorithm can be protected under patent law.
Can You Patent an Algorithm?
You might be surprised to know, it’s not as simple of a question as you might think. Before we can answer if you can get patent protection for an algorithm, it’s best to start with “What are the general patent requirements?“, and “How do the courts view algorithms?”
The answer to these two questions may create a unique opportunity for anyone who has developed a machine learning algorithm and is looking for software patents. Keep reading…
Algorithms in Today’s Information Age
Artificial intelligence and machine learning developments have skyrocketed in recent years. As we continue deeper into the information age, machine learning and software developments have become increasingly important as our culture becomes more and more dependent on computer devices and machine learning software.
Software developers are increasingly developing and relying on machine learning algorithms to make advances in this digital age. As these software developers have sought to have their ideas patented to protect their intellectual property rights, related patent case law has been changing in this area in recent years.
General Patent Requirements
Do you have software that can be patented? In general, for something to be patentable it must be useful. To be considered useful, it must fall into one of four categories:
- Manufacture; and
- Composition of matter.
Moreover, the invention you are looking to patent must also not be:
- An abstract idea;
- A natural phenomenon; or
- A law of nature.
Can You Patent An Algorithm Based On These Criteria?
The problem machine learning algorithms run into in meeting these patent criteria is that, on their own, algorithms are abstract ideas. When standing on their own, courts consider algorithms foundational tools for scientific work… rather than patentable ideas.
The good news is… If you can break down your machine learning algorithm and software patents into a series of mathematical steps and procedures that mechanize a process, then the algorithm shifts from “abstract idea” into the patentable “process” category.
So, while an algorithm cannot be patented, you can patent the series of steps that lead to your algorithm.
Eligibility Criteria for Patents Related to Algorithms
A machine learning algorithm is not patentable if it falls into one of three categories:
- Mathematical concepts;
- Methods of organizing human activity; or
- Mental processes.
So to have patentable software, if you can break down your machine learning algorithm into a series of steps and explain how it solves a real-world problem, then it will probably meet the eligibility criteria.
The Alice / Mayo Test
Whether you can patent software has been increasingly fought over in the courts. In Alice Corp. v. CLS Bank, the Supreme Court ruled that because abstract ideas, natural phenomena, and laws of nature “are the basic tools of scientific and technological work”, it was concerned that granting patent rights for these types of tools might impede innovation rather than promote it. Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 573 U.S. 208, 216 (2014).
Alice, and another related legal opinion Mayo Collaborative Servs. v. Prometheus Labs., Inc., 566 U.S. 66, 71 (2012), established a test referred to as the Alice/Mayo test. This patent law test is as follows:
- Is the claim at issue directed to a “judicial exception,” such as an abstract idea? If so then:
- Do the claims contain an element or combination of elements to ensure that the patent in practice amounts to significantly more than a patent upon the [ineligible concept] itself?
Alice and the lower court cases that followed it have established the standard for subject matter eligibility for the Patent and Trademark Office in the U.S. Different countries will have different standards
The United States Patent and Trademark Office has put out some useful guidance on patent subject matter eligibility. This explains this test in greater detail to help understand if you can patent a machine learning algorithm.
Examples Of Patents Related to Algorithms
An example of a recent algorithm-related patent can help explain the type of software code that could be eligible for software patents. For example, Amazon is seeking to patent an algorithm for :
“a system for capturing and processing portions of a spoken utterance command that may occur before a wakeword.”
Essentially, Amazon is hoping to patent the algorithm underlying the software idea that Alexa, its ‘smart’ speaker, will be able to recognize specific spoken words such as “love” or “hate” during speech taking place around it, even if the wakeword “hey Alexa” has not been used before making the statements.
Who Is Patenting Algorithms?
Everyone from large tech giants like Google and Apple to small start-up companies are having algorithms patented. Artificial intelligence is developed across different countries. A majority of algorithm and software patents are being assigned to large tech companies located in the United States and Japan.
In terms of specific companies, IBM is the current leader of artificial intelligence and software patents. IBM currently has over 8,000 patent applications, followed by Microsoft with over 5,000 patent applications. Toshiba and Samsung are companies that are also high on the list of total artificial intelligence patents.
Types of Artificial Intelligence Patents
The WIPO recently released a report based on a detailed study of recent artificial intelligence patents. It found that:
- Machine learning is the dominant artificial intelligence technique disclosed in patents;
- For AI-related patent applications, computer vision (which includes image recognition) is the most popular to be patented; and
- Transportation and telecommunication are the top fields for artificial intelligence applications.
Patent Rights Vs. Copyright Rights for Algorithms
Patent rights are not the only form of intellectual property rights. Even if you cannot patent your software because you cannot distill it down to a series of steps, you may be able to copyright aspects of it, such as the source code. Copyrights gives original works of authorship patent protection.
Computer programs are copyrightable. Keep in mind that copyright rights actually last longer than patent rights, and so can preserve your control over portions of your software longer than a patent might be able to. These rights are different from patent rights, however, so it’s important to get legal advice to understand your various intellectual property rights.
Who Owns An Algorithm?
Since algorithms are usually proprietary intellectual property, the developers or companies they work for reserve the right to modify, repurpose or sell them.
Is Google’s Algorithm Patented?
Yes, Google’s algorithm is patented. PageRank was developed by Lawrence Page, one of the co-founders of Google (with Sergei Brin), in 1997. Brin filed a patent application on January 9, 1998, and the patent was granted on September 4, 2001.
Can Artificial Intelligence Own Copyright?
No, AI cannot own a copyright. The copyright protection requirements do not apply to AI-generated works since they are not created by humans.