The USPTO issued a Report in October 2020 titled Inventing AI: Tracing the diffusion of artificial intelligence with U.S. patents, along with supplementary material that describes the methodology and scope of patent related data used in the Report. Following a first report also issued in October 2020 that pertains to AI and IP policy, Inventing AI chronicles the USPTO’s increase in the number of filed AI-related patent applications and issued patents in recent years, and the attendant proliferation of AI technologies.
As new technologies are developed, it takes time for them to be understood, adopted, and effectively utilized. Technologies that do become broadly adopted may have potentially large effects on innovation, productivity, economic growth, and the volume and variety of goods and services provided. The economic impact of AI-related technology is said to be larger when a growing number of inventors, companies, and other organizations use AI in their invention and production processes.
The Inventing AI Report utilizes U.S. patents as a proxy to gauge the extent to which AI technologies are growing in volume and diffusing across a broad spectrum of technical areas, inventors, companies, and geographies. To gauge the potential impact of AI, the USPTO developed a machine learning patent landscape algorithm that was used to determine the volume, nature, and evolution of AI technologies contained in U.S. patents from 1976 through 2018.
AI Component Technologies
The Report defines an AI patent application or issued patent to include one or more of eight component technologies that span software, hardware, and applications, as listed below:
- Planning/control — contains processes to identify, create, and execute activities to achieve specified goals
- Knowledge processing — involves representing and deriving facts about the world and using this information in automated systems
- AI hardware — includes physical computer components designed to meet performance requirements through increased processing efficiency and/or speed
- Computer vision — extracts and understands information from images and videos
- Machine learning — contains a broad class of computational models that learn from data
- Natural language processing — relates to understanding and using data encoded in written language
- Speech recognition — relates to techniques to understand a sequence of words given an acoustic signal, including answering articulated questions and responding to spoken commands
- Evolutionary computation — contains a set of computational routines using aspects of nature and, specifically, evolution
(1) AI Patent Applications More Than Doubled From 2012 to 2018
At the end of 1999, the American Inventors Protection Act (AIPA) provided for publication of patent applications 18 months after filing, which subsequently increased the volume of publicly available patent applications. Because of the implementation period associated with publishing patent applications after the expiration of 18 months from the earliest claimed filing date, the trends that pertain to the volume and share of public AI patent applications are most informative after 2002.
In 2000, there were approximately 10,000 published AI patent applications, which constituted approximately 6% of all published applications. From 2002 to 2018, AI patent applications increased by more than 100%, from 30,000 to more than 60,000 annually. Over the same period, the share of all patent applications that included some form of AI grew from 9% to almost 16%.
Relatedly, on a global scale, a report by the World Intellectual Property Organization (“WIPO Report”) found that nearly 340,000 AI-related patent families were published from 1960 until early 2018. The number of AI patent applications filed annually grew by a factor of 6.5 between 2011 and 2017, with over half of the identified inventions being published since 2013.
(2) AI Patent Applications in Several Component Technologies More Than Doubled From 2012 to 2018
In 2018 there were slightly more than 40,000 planning/control applications and slightly less than 40,000 knowledge processing applications, more than doubling their respective numbers in 2002. Also in 2018, the number of public patent applications in computer vision (approximately 17,000) and machine learning (approximately 10,000) more than doubled since 2002.
The number of applications in AI hardware increased to about 19,000 in 2018 from approximately 10,000 in 2002. The close association between the increase in number of AI hardware and computer vision applications may reflect the interplay between advances in image recognition and the need for attendant increases in computational power and performance.
(3) The Percentage of Patent Groupings With AI Patents Quadrupled From 1976 to 2018
Patents containing AI appeared in about 10% of all technology groupings used by the USPTO in 1976 and about 35% of all technology groupings in 2002. In 2018, AI patents appear in more than 42% of groupings, in three distinct clusters with different diffusion rates. Planning/control and knowledge processing and are diffusing the fastest across patent technology groups, which reflects the general applicability of these AI components to a wide variety of technical areas.
The diffusion rate is slower, but still increasing, for a second cluster that includes computer vision, machine learning, and AI hardware, which individually appear in a range from approximately 17–20% of the technology groupings. Diffusion for the third cluster (evolutionary computation, speech recognition, and natural language processing) is the slowest, individually hovering around 5% in 2002 and between approximately 6–10% of technology groupings in 2018.
Relatedly, the WIPO Report found that patent literature focuses most extensively on machine learning, followed by logic programming (expert systems) and fuzzy logic. The most predominant AI functional applications are computer vision, natural language processing and speech processing.
(4) The Percentage of AI Inventors and Patent Owners Increased Substantially From 1976 to 2018
The percentage of inventor-patentees who are active in AI started at 1% in 1976 and increased to 25% by 2018. Growth in the percentage of organizations patenting in AI has been similar, from approximately 2% in 1976 to 22% in 2018.
From 1976 to 2000, AI inventor-patentees tended to be concentrated in larger cities and established technology hubs, such as Silicon Valley. Since 2001, AI inventor‑patentees have been increasing states such as Maine, South Carolina, Oregon, and Montana, as well as in several Midwest Region states.
Most of the top 30 AI companies are in the information and communications technology sector. These companies held 29% of all AI patents granted by the USPTO from 1976 to 2018.
Globally, companies from Japan, the U.S. and China dominate patenting activity. AI-related patent filings were made in the early 1980s in Japan, which was subsequently overtaken by the U.S. and China. Since 2014, China has led the world in the number of first patent filings in AI, followed by the U.S. Together, China, the U.S., and Japan account for 78 percent of total patent filings. See WIPO Report.
Key Takeaways and Considerations
- National and global AI-related patent filings have soared in recent years, and appear to be poised for continued growth.
- AI technology is being used beyond cutting-edge, high-tech fields and is diffusing broadly to sectors such as banking and airlines and technologies such as fitness equipment and agriculture. Organizations should ensure that IP policies and invention disclosure procedures facilitate disclosure and adequate protection of all AI-related technology and inventions.
- Global considerations such as patent eligibility requirements and the scope of patent protection available in key jurisdictions of interest, as well as other forms of possible IP protection should not be overlooked.