As artificial intelligence (AI) becomes integrated into more and more aspects of our daily lives, startups of all different kinds are incorporating it into their products. However, determining whether any given AI technology is patentable can be a challenging task given the rapidly changing nature of the tech itself and the complex legal framework surrounding patents. In some cases, it might make more sense to keep what goes into the AI a trade secret instead of seeking patent protection at all. In this article, we’ll explore some of the key considerations that startups should keep in mind when assessing whether their AI-based invention is ready for patenting.

First, it is important to understand what is patentable when it comes to AI. Generally speaking, patents are granted for novel and non-obvious inventions. For AI in particular, it can also really help to have a practical application for the AI. In other words, what does the AI do in the real world? While in some cases the invention might be in the AI itself or how it is trained, describing how that AI will be used in the real world can provide a backstop to potential future challenges that the invention is not patentable.

For instance, does your AI improve the functioning of a computer? Does your AI make an inference to then operate an IoT device to take a physical action? Does your AI improve the manufacturing process? These implementation details can be particularly useful down the line when your patent application is examined by the United States Patent and Trademark Office (USPTO). These details can also help prepare the case for filing in Europe and other jurisdictions where a particular focus is placed on the real-world implementation of the AI rather than just the functioning of the AI itself. And if you’re thinking big with your startup, you probably need to be thinking of filing for patents outside the U.S. anyway.

There are other guideposts to consider as well in terms of whether your AI is ready for patenting. For instance, as a best practice, your patent application should disclose (1) the AI architecture and what type of AI model is being used in the first place, (2) the algorithm used with the AI, (3) what training methods are used to train the model, (4) the nature and character of the training data itself, (5) what the inputs and outputs of the model are during deployment, and (6) how those outputs are used. This last one gets to the real-world implementation aspect mentioned above, and the more different implementations and use cases you can disclose, the better. Keep in mind that you don’t actually need a working prototype to pursue patent protection as long as you can articulate how to make and use your invention.

Additionally, if you’re going to spend a decent amount of money trying to procure a patent, it might help to do a little due diligence first. This involves performing your own prior art search or hiring a professional search company to do it for you. A well-done professional search can give you an idea of what’s out there already and help you determine if you need to refine or improve your own invention to get around existing art. But here, remember that the inventors have a duty to disclose relevant prior art to the USPTO so be sure to save any relevant material you find.

Another consideration when assessing whether to patent an AI invention is the potential for actually detecting infringement. Even your AI invention is patentable, if it would be difficult to get inside the “black box” of a potential infringer’s AI to see if they’re really copying you, then patent protection may not be best. This is another reason to focus on the inputs, outputs, and practical applications of the AI in your own patent application – these aspects can be easier to detect in a potential infringer than the internal functioning of the AI itself.

Moreover, if your AI invention has infringement detectability issues, you might consider keeping the technology a trade secret instead, being sure to safeguard the trade secret through employment agreements and other means. However, it’s also important to keep in mind that if a competitor could reverse-engineer the trade secret without the aid of confidential information, or at least come close with a design-around, then the trade secret owner would not have much recourse.

There’s one other thing you might also keep in mind as your startup gets off the ground. Remember that public disclosures and offers for sale can create bars to you ever being able to get patent protection for your AI invention. Therefore, the best practice is to file a patent application before publicly disclosing your invention at trade shows, writing about your invention in a tech journal, offering your invention for sale on the open market, or working with third parties to develop your invention.

Have questions about protecting your invention? Rogitz & Associates is here to help. Specializing in patent protection for high-tech inventions including AI-based inventions, our phone lines are always open. Feel free to call us at (619) 338-8075.