The field of artificial intelligence (AI) has progressed rapidly over the last few decades, resulting in billions of people using AI of some form or another in their daily lives. Such widespread adoption of the technology has prompted experts to suggest that the AI market could grow to as large as $15.7 trillion within the next two decades. Accordingly, rewards clearly await innovators who can invent, build, sell, license, or otherwise leverage AI-related technologies. However, in light of Alice Corp. Pty. Ltd. v. CLS Bank Int’l and other more recent developments with regard to patentable subject matter, individual inventors and companies alike may be uncertain how to secure AI-related intellectual property (IP) assets.
Here, the authors utilize various patent search and analytical tools, including Google Advanced Patent Search and Juristat, to obtain quantitative information on published AI-related patent filings. Based on such data, this article illustrates interesting trends in global patent filings, art unit assignment by claim term usage, and recent outcomes for AI-related patent applications, all of which may assist patent practitioners develop effective strategies for protecting innovations in this burgeoning technology area.
First, research indicates, perhaps as expected, that AI-related patent application filings have been increasing throughout the world at growing annualized rates. Figure 1 illustrates the number of AI-related patent application filings in various jurisdictions between the years 2006 and 2016.
Figure 1: Number of AI-related Patent Applications Filed by Year and Jurisdiction
Notably, in 2016, AI-related patent application filings in China outpaced those of other popular jurisdictions, including the United States (U.S.), Patent Cooperation Treaty (PCT), Europe, Japan, and Korea. Other reports confirm this trend continuing through at least 2017. The recent trend of increased funding for Chinese AI startups might explain this growth; those startups reportedly received nearly 50% of total global AI startup funds in 2017.
While China is becoming a leader in the AI patent space, the U.S. has also recently seen tremendous growth in this technology area. For example, in 2016, applicants filed 9,605 AI-related patent applications in the U.S., a decade-over-decade increase of almost 500%.
Second, research suggests that those seeking AI patent protection in the U.S. should carefully assess the particular claim language used to describe their inventions, as different AI‑related claim terms could lead to vastly different patent examination outcomes. More specifically, the U.S. Patent and Trademark Office (USPTO) assigns each U.S. patent application to one of many art units, which are organizational units of patent examiners. Each USPTO art unit is responsible for a set of technology subclasses in the U.S. Patent Classification System (USPC). According to the USPTO, classification of “invention information” for U.S. patent documents is mandatory, and such invention information “is almost always in the claims.” Therefore, the USPTO considers claim language to be a key factor when assigning an application to an art unit.
In this regard, different art units may have different examination outcomes, such as with respect to application allowance rates, examination periods, and types of rejections issued. Consequently, applicants may find it worthwhile to evaluate the relationship between common AI-related claim terms and art unit assignments. Table 1 illustrates the most popular USPTO art unit assignments for U.S. patent applications that recite various AI-related claim terms.
Table 1: Popular USPTO Art Unit Assignments By AI-related Claim Term
As shown, for each claim term, the five most-popular art units are ordered from top to bottom according to the number of patent applications assigned thereto that include the claim term at issue. For example, patent applications including the claim term “artificial intelligence” were most likely to be assigned to art unit 2129. Generally, this result is expected, as the art unit 2129 examines patent applications related to “Miscellaneous Computer Applications.”
However, a review of Table 1 indicates that use of certain AI-related claim terms could lead to some unexpected results. For example, patent applications that include the terms “regression” or “clustering” are most commonly assigned to the art unit 1631, which examines patent applications related to “Molecular Biology, Bioinformatics, Nucleic Acids, Recombinant DNA and RNA, Gene Regulation, Nucleic Acid Amplification, Animals and Plants, Combinatorial/ Computational Chemistry”. This result could stem from frequent use of the terms “regression” or “clustering” in the fields of biology and chemistry, or may be due to other reasons. However, in any case, those seeking AI patent protection in the U.S. should consider the results in Table 1, in an effort to avoid assignment to unintended art units.
Furthermore, in addition to evaluating the relationship between common AI-related claim terms and art unit assignments, it is beneficial to review the historical examination outcomes for those art units. This way, those seeking AI-related patent protection in the U.S. may be able to adjust their claim term usage in an effort to obtain more favorable examination outcomes. For example, applicants could try to have their applications assigned to an art unit that has, on average, a higher historical allowance rate and/or a faster time to allowance.
Table 2 includes relevant outcome-based statistics for the five most common art units associated with the claim terms shown in Table 1.
Table 2: Patent Prosecution Outcomes for the Five Most-Common AI-related Art Units
Table 2 indicates that art unit 2129 is not only the most common art unit associated with the AI-related claim terms in Table 1, but it is also one of the more favorable art units in terms of patent prosecution outcomes. From among the top five art units listed in Table 2, art unit 2129 has the highest allowance rate and the shortest average time from filing to allowance. Additionally, although art unit 2129 may not seem to have the lowest rejection percentages, especially in comparison to art unit 2122, the average number of office actions of 1.5 indicates that, in many cases, the prosecution phase in art unit 2129 may be relatively short.
On the other hand, research indicates that art unit 1631 is historically the least favorable art unit among those listed in Table 2. Namely, among the top five art units, art unit 1631 has the lowest allowance rate, the highest rejection percentage average, the highest average number of office actions, and the longest average time to allowance.
Table 2 illustrates other stark differences between art units 2129 and 1631. First, a review of the rejection statistics (i.e., % Alice Rejections, % 101 Rejections, % 102 Rejections, and % 103 Rejections) indicates that examiners in art unit 1631 issue, on average, 8% more rejections than art unit 2129 in each rejection type. Perhaps more significantly, examiners in art unit 1631 issue 22.62% more Alice rejections than those in art unit 2129 and 10.92% more 101 rejections than those in art unit 2129. The difference in Alice rejections is important to note, as Alice-type rejections are common for software patents.
Significant differences between art units 2129 and 1631 also exist with respect to the average number of office actions and time to allowance. In particular, patent applications assigned to art unit 2129 receive about 1.5 office actions on average. This indicates that art unit 2129 may issue perhaps one or two office actions, and could issue a notice of allowance without even issuing a final rejection. Contrast this with the average of 3.4 office actions issued by art unit 1631, more than twice that of art unit 2129. Such data indicate that applicants having applications assigned to art unit 1631 may end up filing at least one request for continued examination (RCE). Finally, patent applications in art unit 1631 take more than 14 months longer to be allowed than those of art unit 2129. As such, not only is it much less likely for a patent application to be allowed if assigned to art unit 1631, but the likelihood of an RCE may significantly increase the costs and length of patent prosecution.
In conclusion, research involving publicly available patent publication records confirms that AI innovation is expanding rapidly throughout the world, particularly in China. In the United States, AI-related patent applications are being assigned to a relatively small set of examination art units, based, in large part, on claim language. The vastly different outcomes across these art units should prompt patent practitioners to think carefully – before filing – about the specific claim terms recited in their patent applications. Such considerations could result in better applicant outcomes, including higher allowance rates and faster, more cost-effective patent prosecution.
Alexandra E. MacKenzie was a 2018 MBHB summer associate.
© 2018 McDonnell Boehnen Hulbert & Berghoff LLP
snippets is a trademark of McDonnell Boehnen Hulbert & Berghoff LLP. All rights reserved. The information contained in this newsletter reflects the understanding and opinions of the author(s) and is provided to you for informational purposes only. It is not intended to and does not represent legal advice. MBHB LLP does not intend to create an attorney–client relationship by providing this information to you. The information in this publication is not a substitute for obtaining legal advice from an attorney licensed in your particular state. snippets may be considered attorney advertising in some states.
 Darrel M. West & John R. Allen, How Artificial Intelligence is Transforming the World, Brookings (Apr. 24, 2018), https://www.brookings.edu/research/how-artificial-intelligence-is-transforming-the-world/.
 Nations Will Spar Over AI, PwC, https://www.pwc.com/us/en/services/consulting/library/artificial-intelligence-predictions/ai-arms-race.html (last visited Aug. 9, 2018).
 See Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 134 S. Ct. 2347 (2014).
 The authors obtained the data using Google Advanced Patent Search on June 4, 2018. With respect to Figure 1, an AI-related patent application was defined as any patent application that included the keywords “artificial intelligence” and/or “machine learning.”
 Echo Huang, China Has Shot Far Ahead of the US on Deep-Learning Patents, Quartz (Mar. 1, 2018), https://qz.com/1217798/china-has-shot-far-ahead-of-the-us-on-ai-patents/.
 Artificial Intelligence Trends To Watch In 2018, CBINSIGHTS, https://www.cbinsights.com/research/report/artificial-intelligence-trends-2018/, (last visited Aug. 9, 2018).
 The authors obtained the data using Google Advanced Patent Search on June 4, 2018.
 Overview of the U.S. Patent Classification System (USPC), USPTO (Dec. 2012), https://www.uspto.gov/sites/default/files/patents/resources/classification/overview.pdf.
 Information in Table 1 is based on data the authors obtained from Juristat, a patent analytics platform, on June 6, 2018 and June 14, 2018. The top five art units were determined by summing the number of applications in each art unit from the twenty claim terms shown in Table 1. Art unit 2129 had the most applications appearing at 7,472, and art unit1631 was fifth at 3,782.
 TC 2100 Management Roster, USPTO, https://www.uspto.gov/patent/contact-patents/tc-2100-management-roster (last visited Aug. 9, 2018).
 TC 1600 Management Roster, USPTO, https://www.uspto.gov/patent/contact-patents/tc-1600-management-roster (last visited Aug. 9, 2018).
 Table 2 indicates (on a per art unit basis): the allowance rate, the percentage of applications given Alice, 101, 102, and 103 rejections, the average number of office actions, the average time from filing to the first office action, and the average time to allowance. “Alice Rejections” are § 101 rejections that cite Alice Corp. Pty. Ltd. v. CLS Bank Int’l.
 James Cosgrove, Alice: Three Years On, The Juristat Blog (Jul. 19, 2017), https://blog.juristat.com/2017/7/19/alice-three-years-on.