Philip J. Foret

Artificial intelligence (AI) is everywhere these days — integrated into our personal and professional lives — whether we realize it or not. AI solutions enhance user interactions, improve daily life and drive technology development. Programmers and developers grasp AI and machine learning, but the intricate workings often remain a mystery to most users who interact with AI-enabled technologies for practical, real-time applications.

For example, with an AI-powered neural network that identifies images, like facial recognition, most people find it challenging to understand exactly how the system identifies images. In the area of intellectual property law, AI serves as a key platform for innovation. Unlike tangible inventions such as, for example, foldable smartphone screens, AI’s abstract nature makes it difficult to patent.

The U.S. Patent and Trademark Office often classifies AI-enabled technologies as abstract and unpatentable. To increase the likelihood of patent approval for an AI-enabled invention, it is essential to lay a clear strategy to prepare for the inevitable view that the invention is abstract. For example, effective strategies highlight specific technical advancements in areas such as AI training methods, data processing, optimization techniques or new applications designed to solve well-known technical problems. Additionally, demonstrating how an AI-enabled network enhances performance, efficiency or accuracy in particular tasks can help establish a concrete use.

One effective strategy involved the patenting of AI-enabled computer vision technology, which used an infinite training loop to train a convolutional neural network with 2D synthetic images captured from a 3D computer-generated synthetic image. The training loop developed models capable of accurately identifying objects shown in real-world photographs. The infinite training process does not exhaust data; instead, it evolves, ensuring that the training loop continuously encounters new data. In this approach, the convolutional neural network is trained continuously with updated data sets of synthetic, computer-generated images, so the deployed model met precision and recall goals of greater than 95%. This result shows that the deployed model successfully and accurately identifies most of the true positives and true negatives of the objects shown in digital images and video.

For patenting with AI technology, it is best to chart a specific path to prepare for eligibility challenges, which will be rigorously assessed under Section 101 of the U.S. Patent Act, particularly in light of significant court decisions such as the U.S. Supreme Court’s 2014 opinion in Alice v. CLS Bank International.