The U.S. Patent and Trademark Office (“USPTO”) is preparing to add to the wave of artificial intelligence (“AI”) and machine learning tools, with its own tool to improve its efficiency and quality of its operations. A subset of these tools focus on the application of AI and machine learning to enhance prior art searches with concept-based search and categorization. The USPTO’s tool is called Unity and is planned for use in carrying out automated prior art searches and presenting the results to examiners before beginning a traditional manual search.1 As the USPTO begins to implement this tool, practitioners should be aware of how these types of tools are used as well as limitations of the technology.
The USPTO Is Testing AI Tools for Examiners
The USPTO is actively investigating AI and machine learning tools to increase efficiency. The USPTO’s Strategic Plan includes initiatives to implement artificial intelligence and machine learning as a prong to improve the USPTO’s effectiveness.2 In particular, the USPTO is considering AI and machine learning tools for several tasks, including search tools, generating templates for office actions, chat bots, and patent classification.3 Similar tools are being considered in foreign IP offices as well.4
Recently, the USPTO developed and began testing Unity, a “cognitive assistant” designed to improve examiners’ prior art search capabilities.5 Unity’s functionality is described as “a single-click, to conduct a “‘federated search’ across patents, publications, non-patent literature and images” to generate a report before an examiner begins searching.6 This functionality is in line with the USPTO’s recent efforts related to automated tools.7
Although the USPTO is yet to release precise details about Unity’s capabilities, the USPTO issued a request for information (RFI) regarding AI tools in 2018.8 This RFI provides some insight into the type of tools sought by the USPTO and likely indicates what Unity may be capable of. Several objectives for the USPTO’s AI tools are outlined in the RFI, including i) broadening prior art search capabilities beyond curated sources; ii) improving classification taxonomy and consistency; iii) identifying more relevant search results; iv) accelerating examiner reviews of search results; v) recording search activities and analyzing those records for areas of improvement; and vi) determining when a search has been exhaustive and is ripe for conclusion.9 Regarding prior art searches, the USPTO is seeking “better results, not more results” to improve search coverage and “influenc[e] examiners to look under different rocks.”10 Other USPTO documents cite to similar goals for AI tools, noting that AI-based search tools can address some of the limitations of current keyword-based search tools.11
A number of artificial intelligence tools, designed to assist with patent-related search, are commercially available. These tools range from assistive search, such as natural-language or term-expansion tools, to advanced predictive analytics suites.12 Some of these tools focus on searching prior art to generate technology landscapes and identify relevant references in large datasets.13 Other tools are designed to analyze patent applications and automated prior art search results to make predictions regarding its patentability and potential rejections.14 Some tools even provide predictive analytics from an idea, before a patent application is drafted, claiming to model whether an invention constitutes patentable subject matter and its novelty.15
Since Unity provides references to examiners prior to starting a manual search, it will likely be capable of analyzing the application, identifying key concepts, and searching for prior art references. This level of automation is similar to the more advanced assistive search tools currently available commercially. The USPTO’s goal for “better” references suggests that Unity may perform some selection or ranking function to direct examiners to specific results, presumably based on relevance to the ideas or concepts the tool identified in the application text.16 Ideally, Unity-generated pre-search reports will give examiners a strong foundation guiding them to relevant references, while weeding out inapplicable ones.
Presumably, Unity will search across the databases already used by the USPTO to create the pre-search report, albeit with a broader scope than a human searcher could accomplish alone. This will help examiners identify material from voluminous sources. If Unity’s capabilities expand its search to resources beyond the USPTO’s databases, examiners may cite references from wider pool of potential prior art (not previously available). In either scenario, the examiner will rely on Unity’s determinations of top references to cope with the volume of results. Although examiners will conduct a manual search after the automated pre-search report, that search may be influenced by the references and concepts identified by Unity. Like other big data analytics and AI tools, those used by the USPTO may be susceptible to bias from the data or applied models. Bias in the tool could negatively impact the examiner’s manual search results, especially if examiners rely heavily on the automated tool. This risk may be addressed if the USPTO includes the pre-search report in the prosecution file history, perhaps with the examiner’s search information. Disclosing the report would allow practitioners to identify which references originate with Unity and related data provided to the examiner by the tool. With larger data sets, examiners will likely rely on Unity and similar tools to enhance their search capabilities, but examiners must still make the decision whether a reference applies to an application.
Although adopting new AI-related tools provides an opportunity for significant improvement of USPTO operations, it is not without risk. The USPTO will need to disclose sufficient information regarding the tools to satisfy transparency and government accountability obligations.17 As noted above, bias, false-positives, and other artifacts of automated search tools may impact the functionality of AI tools and tasks that rely on them. Notably, the USPTO has already indicated that it cannot implement black-box tools, i.e., those where the factors and process determining an output are not identifiable, and appears to be proceeding with some awareness to other issues.18 Practitioners should also be aware of potential limitations of AI tools, both in their own use and when used by other entities, and take steps to eliminate or reduce them.
As the USPTO implements Unity, it is important to understand the tool’s capabilities and limits. Unity’s automated pre-search reports will assist examiners with the ever-growing volume of prior art, as will similar tools in development by foreign patent offices. Practitioners should be aware of how these pre-search reports are created and understand potential limitations of the technology.
*Ryan Dowell, a Baker Botts law clerk, assisted in the preparation of this article.
1 Andrei Iancu, Dir., U.S. Patent & Trademark Office, Remarks at the 2018 National Lawyers Convention (Nov. 15, 2018), https://www.uspto.gov/about-us/news-updates/remarks-director-iancu-2018-national-lawyers-convention.
2 U.S. Patent & Trademark Office, 2018–2022 Strategic Plan 6 (2018), https://www.uspto.gov/sites/default/files/documents/USPTO_2018-2022_Strategic_Plan.pdf.
3 See, e.g., U.S. Patent & Trademark Office, Patent Public Advisory Committee Quarterly Meeting: IT Update (2018), https://www.uspto.gov/sites/default/files/documents/20180802_PPAC_AI_IT_Update.pdf; Iancu, supra note 1.
4 See Index of AI Initiatives in IP Offices, World Intellectual Prop. Org., https://www.wipo.int/about-ip/en/artificial_intelligence/search.jsp (last visited Jan. 28, 2019) (listing 64 AI initiatives in IP offices across member countries, including several prior art search tools).
5 See Iancu, supra note 1. “Cognitive assistant” is often used to describe AI tools that enhance human productivity with decision-making capabilities, such as sorting through large data sets and identifying important pieces to display for the user.
7 See, e.g., Patent Quality, U.S. Patent & Trademark Office, https://www.uspto.gov/patent/patent-quality; U.S. Patent & Trademark Office, Patent Quality UPDATE (2017), https://www.uspto.gov/sites/default/files/documents/patent-quality-update-nc-1-25-presentation.pdf; U.S. Patent & Trademark Office, Patent Quality Conference 5 (2016), https://www.uspto.gov/sites/default/files/documents/PatentQualityConference%20Booklet.pdf (describing efforts beginning in 2015).
8 U.S. Patent & Trademark Office, USPTO’s Challenge to Improve Patent Search With Artificial Intelligence (2018), https://www.fbo.gov/utils/view?id=719670d083e8f92f9c11394b3895ef86.
9 Id. at 3–4.
10 Id. at 3.
11 See, e.g., Patent Public Advisory Committee Quarterly Meeting: IT Update, supra note 3.
12 See, e.g., Ambercite, https://www.ambercite.com/ (last visited Jan. 28, 2019) (non-keyword search to “find, score and rank patents”); PatentField, https://en.patentfield.com/ (last visited Jan. 28, 2019) (semantic and fuzzy search tools); TEQMINE, https://teqmine.com/ (last visited Jan. 28, 2019) (“expert” analysis of full text patents).
13 See, e.g., Innovation Landscape and Competitive Intelligence, Clarivate Analytics, https://clarivate.com/specialty/innovation-landscape-competitive-intelligence/ (last visited Jan. 28, 2019).
14 See, e.g., RoboReview, TurboPatent, https://turbopatent.com/roboreview/ (last visited Jan. 28, 2019) (providing predictive analytics for §§ 101, 102, 103, and 112 rejections)
15 See, e.g., Invention Hub: Predict, TurboPatent, https://turbopatent.com/inventionhub/ (last visited Jan. 28, 2019) (shows analytics from an “idea” description with novelty and subject matter eligibility predictions).
16 See id.
17 The USPTO noted that it needs to be able to explain its prosecution decisions. USPTO’s Challenge to Improve Patent Search With Artificial Intelligence (2018), supra note 8, at 2. For example, if a USPTO obviousness rejection is under review by a court, the agency would need to be able to provide a record of substantial evidence supporting why it determined knowledge to be known within the art. See In re Zurko, 258 F.3d 1379 (Fed. Cir. 2001) (holding that a Board of Patent Appeals decision is subject to substantial evidence review); MPEP § 2144.03.
18 See USPTO’s Challenge to Improve Patent Search With Artificial Intelligence, supra note 8, at 2.
# # #About Baker Botts L.L.P.
Baker Botts is an international law firm of approximately 725 lawyers practicing throughout a network of 14 offices around the globe. Based on our experience and knowledge of our clients’ industries, we are recognized as a leading firm in the energy and technology sectors. Throughout our 179-year history, we have provided creative and effective legal solutions for our clients while demonstrating an unrelenting commitment to excellence. For more information, please visit BakerBotts.com.