There are some special considerations when patenting AI inventions. To better understand what they are, consider the figure to the right, which shows three types of deep learning invention types.
Type I uses deep learning, such as a deep neural network, to read signals output from some physical device like sensors on an industrial process, machinery, a chemical reaction, or a digital image. This is a common use for deep learning – improving some process or thing by applying deep learning to its form or operation.
We call Type I inventions “closed loop” because they use deep learning to provide feedback into the system or thing being analyzed, to improve upon it. The United States Patent and Trademark Office (USPTO) favors Type I inventions and will grant patents for them as long as they pass the other requirements, such as being sufficiently inventive.
Type II inventions are what we call “open loop.” This describes when deep learning on one system or thing is applied to improve or change a different system or thing. The USPTO also favors this type of invention.
Type III inventions can be problematic to patent. Not only is the invention open loop, but the results of the learning are merely displayed, or stored in files. The USPTO considers the mere display or storage (or communication) of learning results as simply moving data around and not doing anything technologically useful with it. Thus, it’s much more challenging to obtain patents for Type III inventions than Type I or II inventions.