AI Bulletin – December 15, 2020
Want the AI Bulletin mailed to you?
- USPTO report on AI patents and applications shows rapid growth
- Patent infringement complaint against social media companies over AI solutions for content surfacing and moderation
- Sentencing of former Uber executive Anthony Levandowski over misappropriation of autonomous driving trade secrets
- Regulatory updates from the National Highway Traffic Safety Administration and Department of Veteran Affairs’s proposed framework for Automated Driving Systems
- Update on the Planner5D Litigation
- QE victory for AI software start-up in trade secret and unfair competition case against Facebook
SPOTLIGHT – USPTO HIGHLIGHTS GROWTH IN AI PATENTS
Report Uses AI Methodology to Find that AI Is Diffusing Broadly Across the Patent Landscape
Background. In October 2020, the USPTO’s Office of the Chief Economist released a report showing AI-related patents are growing across a broad range of technologies, organizations, inventors, and geographic areas. To study the acceleration of the importance of AI, the USPTO developed an AI algorithm to gauge the impact that AI has had on patents and, by extension, technological progress more generally. The key findings of the report show the broad diffusion of AI across several metrics:
- Volume of AI patent applications. From 2002 to 2018, the number of AI patent applications more than doubled, from approximately 30,000 to more than 60,000 annually. Not only are absolute numbers up, but AI patents applications as a portion of the total number of patent applications also grew from 9% to nearly 16% during the same period.
- Technologies involving AI. In 1976, patents containing AI appeared in approximately 9% of all technology subclasses used by the USPTO. In 2018, that portion more than quadrupled to approximately 42%.
- Diversity of owners-at-grant of AI patents. Although the majority of top-30 U.S. AI patent owners-at-grant are in the information and communications sectors, there are several companies among this group in other business areas: global multinationals involved in aviation and aerospace (Boeing), building technologies, healthcare, energy, and manufacturing (General Electric and Honeywell), photography and imaging (Eastman Kodak), and venture capital and finance (Bank of America).
- Inventors active in AI. The portion of inventors filing patent applications with the USPTO and active in the AI space was 25% in 2018.
- Geographic reach. AI patentees have been geographically diverse, with traditional technology hubs such as Silicon Valley and Seattle growing alongside the South (North and South Carolina, Georgia, Florida, and the Austin/Dallas/Houston metropolitan areas in Texas); the Southwest (the Phoenix/Tucson metropolitan areas in Arizona and Albuquerque metropolitan area in New Mexico); and Midwest (Illinois, Wisconsin, Ohio, and Kansas).
Categorizing Growth in AI. The USPTO analysis broke AI down into eight “component technologies,” with the use of one or more of these component technologies leading the algorithm to classify a patent or patent application as “containing AI”:
- Knowledge processing generally involves using information about the world in automated systems, e.g., Intuit’s “system and method for identifying, prioritizing and encapsulating errors in accounting data” by applying a knowledge base to the input data.
- Speech includes the recognition and processing of spoken language, e.g., Google Assistant.
- AI hardware includes the physical computer components used to run AI algorithms, e.g., Nvidia’s computer hardware patent for accelerating a game artificial intelligence process.
- Evolutionary computation uses a set of computational routines that represent aspects of evolution, e.g., Chevron’s patent for “forecasting the production of a petroleum reservoir utilizing genetic programming.”
- Natural language processing includes the recognition and understanding of written language, e.g., Cincinatti Children’s Hospital’s patent to process “text with domain-specific spreading activation methods.”
- Machine learning involves applying computational models to data sets to teach recognition of, e.g., consumer product classifications.
- Vision uses images or video to gather information for, e.g., medical imaging diagnostics.
- Planning/control involves processing information to identify, create, and execute activities to achieve specified goals, e.g., detecting problems in a process plant and adjusting process control parameters to correct the problem.
The USPTO’s analysis found that planning/control and knowledge processing represent the largest share of AI components in patents between 1990 and 2018, each with approximately 40,000 patent applications incorporating one or both of those components. These two components often overlap with each other or with other components, as they are the most general and can be applied to a wide variety of technical areas. In general, the USPTO analysis found significant overlaps, especially among related component technologies such as natural language processing and speech.
As one example of the interplay between component technologies, growth in machine learning, vision, and AI hardware—the latter serving as the underlying computational power for increasing advances in the former two component technologies—has accelerated since 2012. Part of this rapid growth is attributed to the recent rise of autonomous vehicles, which heavily rely on advances in vision and, relatedly, AI hardware.
Diversity of owners-at-grant of AI patents. The USPTO analysis found some counterintuitive AI patent owners-at-grant among the top-30 in the United States. While most owners-at-grant are in the information and communications sectors, there are several companies in other business areas: global multinationals involved in aviation and aerospace (Boeing), building technologies, healthcare, energy, and manufacturing (General Electric and Honeywell), photography and imaging (Eastman Kodak), and venture capital and finance (Bank of America).
Diffusion to Other Technology Areas. The USPTO’s analysis found a broad diffusion of AI component technologies to other areas, represented by the growth in the portion of patentees that receive patents including AI components. Between 2000 and 2020, the percentage of all patentees that had at least one AI patent more than doubled, from about 10% to nearly 25%. Moreover, the USPTO found that the percentage of inventor-patentees with at least one AI patent has outpaced the percentage of owners-at-grant with at least one AI patent since 2009, indicating that the uptake of AI takes place not only across organizations, but within them as more inventors adopt AI component technologies.
Geographic Diffusion. The USPTO found that typical technology hubs such as Silicon Valley have large concentrations of AI patent inventors, as do areas with major research universities, likely due to their resource advantage with respect to specialized knowledge relating to AI. However, other areas of the United States have become specialized in certain AI sub disciplines. Inventors from Oregon, home to Nike, have been active in AI patents relating to fitness training and equipment. States in the Midwest, including Wisconsin, Ohio, and Kansas, are home to inventors applying AI technologies to medical imaging and diagnostics.
AI Abroad. The USPTO’s conclusion that AI technologies are rapidly becoming common in a variety of contexts is confirmed by international studies on the rise of AI patents. The Canadian Intellectual Property Office concluded that the “number of worldwide AI patented inventions has increased exponentially over the past 20 years.” The China Institute for Science and Technology Policy at Tsinghua University found “an overall upward trend on an application number consolidation basis over the last nearly 20 years.” The European Patent Office found that AI is among the “fastest-growing fields” in Europe. The World Intellectual Property Organization found that since the emergence of AI in the 1950s, nearly 350,000 AI-related patent applications have been filed, with nearly half published since 2013.
OTHER RECENT DEVELOPMENTS
- On November 25, a Xerox subsidiary, Palo Alto Research Center Inc. (PARC), filed complaints against Facebook, Twitter, and Snapchat alleging infringement of patents related to AI solutions for targeted advertising and content surfacing and moderation. According to the complaints, some of the asserted patents are practiced by the social media defendants in their AI-based systems used to restrict misinformation and exploitative advertising.
- The Patent Public Advisory Committee released its Annual Report for 2020 noting the challenges that AI presents to the USPTO. The Committee provided an update on the developing AI policies at the USPTO, including the published proceedings from the USPTO’s AI Policy conference and an update on the AI tools the USPTO is evaluating for patent examination.
- Former Uber executive Anthony Levandowski pleaded guilty to trade secret theft and was sentenced to 18 months in prison by Judge Alsup in the Northern District of California for his role in taking trade secrets from Google’s self-driving car program just prior to his departure from the company. Mr. Levandowski was also ordered to pay a $95,000 fine and $756,499.22 in restitution. Mr. Levandowski was originally referred for criminal investigation by Judge Alsup in 2017 in connection with the Waymo v. Uber llitigation over which he was presiding (and in which Quinn Emanuel represented Waymo).
GOVERNMENT & REGULATION
- The Department of Veterans Affairs reported a threefold improvement in the number of successfully automated claims after it implemented an AI-assisted API last year. The API was implemented in line with the Department of Defense’s AI ethics principles, which were released earlier this year in February.
- The National Highway Traffic Safety Administration published an advance notice of proposed rulemaking requesting comment on the development of a framework for Automated Driving System (“ADS”) safety on November 19, 2020. While the NHTSA stated that it is still too early to make any decisions on federal safety standards needed to address ADS safety due to a lack of technological maturity, this is the beginning of the process for a Federal Motor Vehicle Safety Standard that could apply to autonomous vehicles.
UPDATE ON THE PLANNER5D LITIGATION
- Prior AI Bulletins covered developments in the Planner5D litigation against Facebook and Princeton University (April 2020 edition and October 2020 edition) related to digital objects and design scenes for use as AI training data. Most recently, Judge Orrick had granted the motion to dismiss Planner5D’s amended copyright claims because the Copyright Office issued registration certificates indicating the year of completion of the allegedly infringed objects was 2019, while the year in which Princeton allegedly scraped the objects from Planner5D’s website was 2016. Judge Orrick granted Planner 5D leave for Planner5D to file registration applications for all its work up to 2016 and to then amend its copyright claims.
- On November 23, Planner5D filed a separate complaint for copyright infringement against Facebook and Princeton University, which will be consolidated with the original suit and its surviving trade secret misappropriation claims. The newly-filed complaint alleges that Planner5D has now submitted corrected applications for its work dating back to 2016. Facebook and Princeton have not yet responded to the complaint.
- Quinn Emanuel successfully defeated a motion to dismiss brought by Facebook against client Neural Magic in a trade secret, unfair competition, unjust enrichment, and tortious interference dispute. Neural Magic is an AI software startup founded by researchers at MIT. Facebook moved to dismiss several of Neural Magic’s non-trade secret state law claims, arguing that these claims overlap with and are preempted by the Massachusetts Uniform Trade Secrets Act (“MUTSA”). Judge Denise Casper of the District of Massachusetts denied Facebook’s motion, holding that these claims are not preempted by MUTSA. The Order noted the lack of authority under Massachusetts law and addressed the split in authority on this issue, ultimately adopting the no-preemption view taken in Arizona, Virginia, and Pennsylvania. Quinn Emanuel associate Kaitlyn O’Connor argued for Neural Magic at the hearing with support from partners Steven Cherny and Patrick Curran./li>