December 17th, 2020
In recent years, artificial intelligence (“AI”) has had a transformative effect on a broad variety of industries, including natural language processing, drug discovery, finance, computer vision, and even transportation logistics. To protect these advancements, many companies have substantially increased their AI-related patent filings, or in some cases, have begun to consider AI-related filings for the first time.
In October 2020, the United States Patent and Trademark Office (“USPTO”) released a report entitled Inventing AI – Tracing the diffusion of artificial intelligence with U.S. patents (hereinafter the “Report”). The Report primarily details the current status of AI-related patent filings. In particular, the Report focuses on the historical growth of AI-related filings, current industry-specific trends, and future protections for innovation in AI.
What is AI?
Broadly, AI generally refers to a computer system (e.g., including software) that performs a task typically performed by humans, such as prediction, cognition, planning, communication, etc. Many of the more recent, high-performing AI systems utilize machine learning, in which the computing system is trained or learns to perform the task on its own via exposure to data. See Report, p. 3.
However, as the USPTO notes, a generic definition of AI lacks the specificity required for classification of patents. See Report, p. 3. Therefore, the Report defines “artificial intelligence” as including one (or more) of these eight component technologies:
- Knowledge Processing: representing and deriving facts about the world and using this information in automated systems (e.g., detection of accounting errors with an algorithm that utilizes a “knowledge base”).
- Speech Recognition: understanding of spoken words (e.g., a virtual assistant application).
- AI Hardware: hardware that is specialized to facilitate AI processing (e.g., tensor processing units).
- Evolutionary Computation: an algorithm or technique that utilizes an evolutionary approach to problem solving.
- Natural Language Processing: understanding written language (e.g., optical character recognition).
- Machine Learning: computational models that learn from data.
- Computer Vision: extraction and understanding of information from images and videos.
- Planning and Control: AI-centric improvement of business operations and logistics.
See Report, p. 3. There are a number of key understandings to be gleaned from the taxonomical system set forth in the Report. First, it should be understood that “artificial intelligence” is not necessarily synonymous with “machine learning.” Id. Namely, categories such as evolutionary computation, or planning and control, may be classified by the USPTO as being “AI” technologies without the explicit inclusion of a machine-learned model in the claim (See e.g., U.S. Patent No. 10,031,490). Second, an invention is not always clearly classifiable under only a single component technology. See Report, p. 3. Rather, some inventions may be classified under multiple component technologies. For example, an invention that trains a machine-learned model to improve logistics could be classified as being both a “machine learning” component technology and a “planning and control” component technology.
Diffusion of AI Across Technologies
According to the Report, the share of patent filings directed to AI has grown substantially in the past few decades – from 9% in 2002 to almost 16% in 2018. See Report, p. 6. Much of this growth has been driven by diffusion of AI technology across industries that are conventionally disconnected from the cutting edge of software innovation. Specifically, technologies in the “planning and control” and “knowledge processing” component technologies have seen substantial growth, with filings in both fields increasing from 5,000 to over 40,000 in the past sixteen years. See Report, p. 6. AI hardware and machine learning filings have also seen a sizeable increase over the same time frame i.e., from 5000 filings to nearly 20,000 filings. Id.
However, measuring growth by component technology does not paint a complete picture. Rather, to provide a more detailed overview, the Report analyses the diffusion of AI technologies across technology subclasses. See Report, p. 7. The Report defines diffusion as “the spread and adoption of a new technology by inventors, companies, and other innovators.” Report, p.6. Broadly, the USPTO reports that the percentage of technology subclasses with at least two AI filings in a year has increased from 20% in 1990 to over 40% in 2018. Id. In particular, the component technologies “planning and control” and “knowledge processing” were the primary drivers in increasing AI diffusion, with both technologies increasing in diffusion from 15% in 1990 to 35% in 2018. Id. Other component technologies, such as “computer vision”, “machine learning”, and “AI hardware”, also increased substantially by 15% in the same time frame. Id.
One of the strongest indicators of AI diffusion is the total percentage of U.S. patentees and owners that own AI patents. While only 5% of patentees owned an AI-related patent in 1990, nearly 25% of patentees own an AI-related patent in 2018. See Report, pp. 9-10. As an example, among U.S. companies, IBM is by far the largest owner of AI-related patents, with 46,752, which is more than the next four companies combined (e.g., Microsoft, Google, Hewlett-Packard, and Intel). See Report, p. 10.
Although the list of the top 30 largest AI patent owners is primarily populated by the largest U.S. information technology companies (e.g., Amazon, Apple, Adobe, etc.), there are a number of other companies represented in the list. Id. In particular, the inclusion of companies such as Eastman Kodak, Boeing, and Bank of America serves to demonstrate the diffusion of AI-related technologies across a broad range of industries. Id.
How should business organizations respond to the rapid diffusion of AI-related technology? One takeaway from the Report is that organizations should identify areas of their respective industry in which AI-related technologies may be utilized in the future and then develop and protect their intellectual property in the identified areas of AI-related growth. For example, organizations may actively seek patentable subject matter from inventors to identify early adoption and filing opportunities and/or may develop robust trade secret protection measures. Organizations that fail to maintain effective channels of communication with inventors risk losing valuable patent protection (e.g., due to a public disclosure of AI-related subject matter). As such, organizations should identify and actively communicate with key inventors and product groups utilizing AI technologies to identify and execute upon opportunities to protect their intellectual property.
In addition, the drafting of valuable and effective AI-related patent applications is best performed by patent counsel with significant experience in AI technologies. Perhaps more so than other software or software-related applications, AI-related applications require specific domain knowledge – both to effectively interview inventors and to draft accurate claims. As such, it is recommended that patent applicants seek counsel that have developed specific focus and experience with AI-related technologies.
Finally, legal practitioners and corporate executives – including those who have not traditionally been involved in computer-related patent filings – should monitor developments in both AI and the law surrounding it. As demonstrated in the Report, the industries that are conventionally tech-adjacent are also the industries that are experiencing the most disruption by AI-related technologies. Therefore, a savvy practitioner and/or organization that leads the charge in actively identifying and generating AI-related filings within their industry may be able to leverage this advantage to develop a strong position over lagging industry competitors.