By: Romaine Marshall and Jennifer Bauer , Polsinelli PC This article addresses the broad scope of artificial intelligence (AI) laws in the United States that focus on mitigating risk, and discusses the...
By: Bijan Ghom , Saxton & Stump This article addresses existing deepfake technology and covers topics such as the available platforms to both create and detect deepfakes and the best practices for...
By: Ellen M. Taylor , SLOAN SAKAI YEUNG & WONG LLP THIS ARTICLE ADDRESSES THE BROAD SCOPE OF artificial intelligence (AI) laws in the United States that focus on mitigating risk. AI-driven employment...
By: Jessica Bishop and Sarah Stothart , GOODMANS LLP This checklist provides an overview of key legal considerations attorneys should review when advising clients on negotiating and drafting contracts...
Review this exciting guide to some of the recent content additions to Practical Guidance, designed to help you find the tools and insights you need to work more efficiently and effectively. Practical Guidance...
Copyright © 2025 LexisNexis and/or its Licensors.
By: Kirk A. Sigmon BANNER WITCOFF
THIS ARTICLE DISCUSSES PATENTING ARTIFICIAL (AI), MACHING LEARNING (ML), AND RELATED INVENTIONS.
It provides a high-level overview of AI and ML, offers tips for drafting a patent application directed to inventions relating to AI and ML, and discusses trends and strategies for handling prosecution of such inventions.
As a preliminary matter, it is important to distinguish among AI in the general sense, ML, deep learning, and other commonly used terms in the field.
AI
The term artificial intelligence generally refers to causing computing devices to perform human-like thinking. The phrase has been used in patent applications for decades, though historically, few computers could do anything remotely approximating human-like thinking. In fact, many patent applications seemed to use the term like a marketing mechanism, extolling the virtues of particular algorithms and/or processes.
ML
ML models, a subset of artificial intelligence, are one of the latest forms of algorithms that enable computers to approximate human-like thinking. ML models are often configured (i.e., trained) through large quantities of data—often referred to as training data—to learn, through that data, to perform particular tasks. While the term machine learning is also quite old (and was used as early as the 1960s by computer scientist Arthur Samuel), it was historically somewhat infeasible, and modern computing devices permit ML model implementation on even consumer-grade hardware.
Stated more plainly, the world has been trying to do ML for a long time, but modern hardware makes it significantly easier to do so.
One of the most promising implementations of ML models comprises so-called deep learning, using artificial neural networks that are intentionally designed to mimic the human brain. Such an approach is computationally costly but can result in some amazing results: for example, the famous ChatGPT algorithm uses deep learning in a manner that allows it to answer questions realistically.
Natural Language Processing (NLP)
It is not uncommon for AI and ML to be associated with NLP, which relates to algorithms that process (i.e., understand, output) human communications (e.g., human-written text, conversations, and the like). For example, NLP might be paired with a trained ML model such that a user can provide natural language input, that input can be processed into appropriate input data for a trained ML model, and then the input data can be provided to input nodes of the trained ML model. As another example, many NLP implementations use trained ML models for the purposes of translation, sentiment analysis, and the like. With that said, not all AI is NLP, and not all NLP is AI. For example, one might argue that an algorithm configured to remove stop words (e.g., the, is, are) from text is an NLP algorithm, though such an algorithm does not involve AI.
For practical guidance on trends in patenting AI, drafting patent applications directed to AI, and specific strategies, subscribers may follow this link.
Not yet a Practical Guidance subscriber? Sign up for a free trial of Practical Guidance to read this complete article.
Kirk A. Sigmon is a partner at Banner Witcoff. He counsels clients at all stages of invention, patent prosecution, intellectual property enforcement, and litigation. Kirk routinely works with U.S., Japanese, Korean, Chinese, and European intellectual property matters. Kirk’s cases have involved a broad range of technologies, including computer networking, cellular communications, video gaming, virtual reality, ML/AI, military weapons systems, blockchain technologies, aerospace flight systems, video encoding, petroleum engineering, optoelectronics, data storage, magnetics, agronomy, and toys. Kirk is an IBM-certified Machine Learning Professional and a Government Blockchain Association-certified Blockchain Legal Specialist.
For an introduction to U.S. patent law, see
> PATENT FUNDAMENTALS
For answers to fundamental questions about patents, see
> U.S. PATENTS Q&A CHECKLIST (PATENT FUNDAMENTALS)
> PATENT LAW FUNDAMENTALS RESOURCE KIT
> PATENT-ELIGIBLE SUBJECT MATTER (SECTION 101) STATEMENTS OF LAW
> GENERATIVE ARTIFICIAL INTELLIGENCE (AI) RESOURCE KIT
For an analysis of the patent litigation process, see
> PATENT LITIGATION FUNDAMENTALS
> OBVIOUSNESS REJECTIONS: ATTACKING THE PRIMA FACIE CASE
> OBVIOUSNESS REJECTIONS: REBUTTING THE PRIMA FACIE CASE
For resources that can help a patent prosecutor respond to an office action in a pending patent application in the USPTO, see
> PATENT OFFICE ACTION RESPONSE RESOURCE KIT