USPTO publishes report on public views on artificial intelligence and IP Policy – US IP law adequate for now, until artificial general intelligence is reached?
11 min read
As artificial intelligence (AI) evolves, it becomes imperative to examine whether the current intellectual property (IP) legal frameworks, in the US and abroad, are adequate to address issues specific to AI. The United Kingdom Intellectual Property Office (UKIPO), European Patent Office (EPO) and the United States Patent and Trademark Office (USPTO) have all recently had the opportunity to weigh in on the issue of whether an AI machine can be named as the inventor on a patent application. In late 2018 and in 2019, Dr. Stephen Thaler filed two patent applications in each of the UKIPO, EPO and USPTO, naming DABUS,1 a patented AI machine, as the inventor of the subject inventions.2 All three offices came to the same conclusion for similar reasons: Current law suggests that an inventor must be a human.3 In January 2019, the USPTO held an AI IP policy conference, which included panel discussions featuring IP specialists to discuss AI and IP policy considerations.4 Following the conference and Thaler's patent applications, the USPTO sought further insight into public opinion on how IP laws and policy should develop as AI technology advances and issued two requests for comment (each an "RFC"), one on August 27, 20195 and a second one on October 30, 2019.6
Earlier this month, the USPTO published the resulting report, entitled "Public Views on Artificial Intelligence and Intellectual Property Policy,"7 which summarizes nearly 200 comments from various stakeholders in response to the RFCs (the "Report"). The Report was divided into two parts, the first covering responses to the August 27, 2019 RFC with respect to patenting AI inventions and the second covering responses to the October 30, 2019 RFC with respect to the impact of AI on IP policy areas other than patent law (including copyright, trademark, database protection and trade secret law). Below is a brief summary of some of the key findings and considerations of the Report.
Part 1 – Patenting artificial intelligence inventions
The USPTO received 99 responses to the RFC corresponding to this section of the Report including from foreign patent offices, bar associations, industry associations, academia and various stakeholders, both national and international, from the electronics, software, automobile, medical and pharmaceutical industries. Notably, the Report discusses:
- Elements of an AI invention: The Report commences by considering the different elements of AI inventions (from the structure of the database on which the AI is trained, to the training data, the algorithm itself and the results of the AI invention). The Report states that AI inventions can be categorized as (a) inventions that embody an advancement in the field of AI; (b) inventions that apply AI (to a field other than AI); and (c) inventions that may be produced by AI itself.8 The Report would have benefited, however, from more clearly articulating the differences between these first two categories and the last category, namely, protecting AI inventions themselves (i.e., categories (a) and (b)) and protecting the output of AI inventions (i.e., category (c)), and considering each type in later parts of the Report (that sometimes implicitly focused on one category or another).
- Eligibility of AI to be a named inventor: The Report acknowledges that US patent law is replete with language indicating that an inventor must be a natural person. It then examines the notion of "conception" (the touchstone of inventorship) and notes that if a natural person contributed to the conception of a claimed invention, "the use of an AI system as a tool would not preclude the natural person from qualifying as an inventor (or joint inventor)."9 These views, however, appear to be based on a general consensus that AI technology is not capable of conceiving an invention itself (e.g., "conception is inherently a human activity").10 Most responses appear to assume that AI has not yet reached artificial general intelligence (AGI) (i.e., when AI has the capability to think and invent on its own).11 Most commenters believe AGI is still in the distant future, but a minority of comments suggest that this is a present-day reality that should be addressed.12
- Ownership of AI-produced inventions: Concerning patent ownership, a vast majority of commenters stated that only natural persons or a company, via assignment, should be able to claim ownership of patents and inventions (and that this should not extend to machines).13 Once again, these conclusions appear to be influenced by the view that AI is not capable of autonomous inventing. In addition, a minority of responses stated such ownership rights should be considered for a person who trains an AI process or who owns or controls an AI system.
- Other considerations: The Report also considers patent eligibility considerations, the disclosure and enablement requirements that apply to all US patent applications, the potential to affect the standard for nonobviousness, and prior art considerations.
In addition to the above, the Report discusses the commenters' views on the necessity of new intellectual property rights to protect AI inventions and commenters were nearly equally divided between the views that new IP rights were necessary and that the current IP framework in the US is adequate.14 The majority of commenters who requested new IP rights focused on the need to protect data associated with AI. However, commenters did not provide concrete proposals on how such new IP rights should function, and many encouraged the USPTO to request further comment from the public on the issue.15
Part 2 – IP protection for AI innovation in areas other than patent law, including copyrights, trademarks, database protections and trade secret law
The USPTO received 98 responses to the RFC corresponding to this section of the Report including from bar associations, industry associations, academia and various stakeholders, both national and international, from the electronics, software, media and pharmaceutical industries. The Report discusses, among other things:
- Copyright law: Similar to the responses regarding AI and patent inventorship, the vast majority of commenters acknowledged that current copyright law does not permit a non-human to be an author (outside of the work-for-hire doctrine).16 Some considered AI as a tool, noting other tools have been used in the past to create copyrightable works (e.g., Photoshop and Garage Band). On the other hand, a minority argued that a sufficiently creative work made by AI without human intervention should obtain copyright protection and that the owner or controller of the AI system who fixes the work in its final form should be able to obtain authorship.17 The Report also discusses potential copyright infringement concerns related to AI algorithms or processes. Commenters noted that use of copyrighted material to train AI, including machine learning (ML) processes, may violate reproduction rights of a copyright owner and such use may or may not be a non-infringing fair use.18 Other commenters expressed the view that use of copyrighted works for ML should be permissible and compensated.19 Another small subset of commenters emphasized the applicability of the fair use doctrine, stating that allowing AI to use copyrighted material for training should be presumed fair use and as a result, would promote further innovation.20 Most commenters also noted that existing laws governing copyright infringement are sufficient, while some commenters emphasized the lack of clarity with regard to AI-generated works.21
- Database protections and trade secret law: Databases and datasets used to train algorithms already have some protection under copyright law as compilations of data as long as the originality requirement is met. One commenter argued, however, that sui generis IP rights may be necessary since copyright protection will not cover the data values in a compilation (i.e., the raw data), and a rearrangement of data would not necessarily be considered a derivative work.22 In the US, databases and datasets are also afforded protection as trade secrets, with criminal remedies under the Economic Espionage Act and civil remedies under the Defend Trade Secrets Act (DTSA) and applicable state laws. One commenter stated that trade secret law may be the only viable protection available in the realm of AI.23 Others argued that issues may arise and a sui generis form of protection may be required in the future (e.g., where trade secret protection is impractical or impossible as data is distributed in a product or where AI makes it easier for certain trade secrets to be uncovered (e.g., like making data more discoverable) without defeating secrecy measures).24 Contract law is another alternative of protection for data. Most commenters stated that existing laws are adequate to continue to protect AI-related databases and dataset and there is no need to reconsider a sui generis database protection law.25 Some commenters even noted that areas other than IP law might be implicated, including privacy and product safety.26
In summary, the majority of the stakeholders that provided responses to the USPTO's RFCs largely agreed that existing US IP laws are sufficiently robust and flexible to address AI-related issues. However, commenters also stressed the need to revisit these issues and consider new IP rights in the future as AI continues to evolve, including when AGI is achieved. The USPTO stated that it will use this Report to further understand how IP policy is impacted by AI and other emerging technologies, by engaging further with the public, providing additional guidance and by continuously training the examiners.27 The full Report is available on the USPTO's website.
1 See "What is Dabus?"
2 See "Patent Applications"
3 See Decision on Petition filed by Stephen Thaler re Patent App'n Nos. GB1816909.4 and GB1818161.0 (United Kingdom Intellectual Property Office, December 4, 2019); see also Press Release, European Patent Office, EPO publishes grounds for its decision to refuse two patent applications naming a machine as inventor (January 28, 2020); see also Decision on Petition filed by DABUS c/o Stephen Thaler re Patent App'n No. 16/524,350 (United States Patent and Trademark Office, April 27, 2020). On August 6, 2020, Thaler filed a complaint for declaratory and injunctive relief in the U.S. District Court for the Eastern District of Virginia, claiming that the USPTO violated the Administrative Procedure Act and improperly added a new requirement for patentability when it said an AI machine cannot be named as an inventor. See Complaint, Thaler v. Iancu, et al, (E.D. Va. 2020) (No. 1:20-cv-00903).
4 See USPTO, Artificial Intelligence: Intellectual Property Policy Considerations. Other iternational intellectual property offices have also hosted conferences, conducted research, published papers, or called for comments on the impact of AI on IP policy: See, for example, World Intellectual Property Organization, Impact of Artificial Intelligence on IP Policy: Call for Comments; UKIPO, Artificial Intelligence and intellectual property: call for views; EPO, Artificial Intelligence; and the Japan Patent Office, Recent Trends in AI-related Inventions.
5 See Federal Register: Request for Comments on Patenting Artificial Intelligence Inventions
6 See Federal Register: Request for Comments on Intellectual Property Protection for Artificial Intelligence Innovation
7 See USPTO, Public Views on Artificial Intelligence and Intellectual Property Policy, (October 7, 2020)
8 Id. at 1-2.
9 Id. at 5.
10 See id.
11 See id. at 6.
13 See id. at 7.
14 See id. at 15.
15 See id. at 16.
16 17 U.S.C. §201(b).
17 See USPTO, supra note 7, at 21.
18 See id. at 26.
20 See id. at 27.
21 See id. at 28.
22 See id. at 38.
23 See id. at 39.
24 See id. at 40.
25 See id. at 37.
26 See id. at 38.
27 See id. at iv.
This publication is provided for your convenience and does not constitute legal advice. This publication is protected by copyright.
© 2020 White & Case LLP