Big Data Analytics to Automate Patent Disclosure of Artificial Intelligence’s Inventions
Ford, Walker, Haggerty & Behar LLP
The exponential growth of artificial intelligence (AI) will change the future of humanity.[1] As Stephen Hawking once suggested, the short-term impact of AI depends on who controls it; the long-term impact depends on whether it can be controlled at all.[2]
For the time being, when it comes to patent incentivization, an AI’s control ends at the United States Patent and Trademark Office (USPTO).[3] A patent is an exclusive right granted to an inventor — a natural person — in exchange for a full disclosure of the invention.[4] Much like humans, AI machines are able to invent arts and sciences.[5] In the patent application process, the USPTO requires a natural person to be the patent holder, and the holder must provide full disclosure of the invention.[6]
Currently, patents are conditioned on inventors describing their inventions. However, it is impossible to provide a full explanation of an invention created by AI when the inner workings and the use of AI in the inventive process are not adequately understood. Hence, compliance with the current patent requirements has become very challenging when it comes to inventions created by AI. However, patent protection should be available for AI-generated works because it will incentivize further innovation.
The question is how AI can provide a full explanation of the inner workings of an invention it creates. When the inventive process is unknown, this strict requirement may require that an owner of AI reveal the AI machine’s algorithm. But requiring AI owners to disclose the AI’s algorithm not only fails to describe the invention but takes away the protection that the AI owners might have in the ownership of the AI machine itself. There are several solutions to satisfy the USPTO’s patenting requirements concerning AI inventions while protecting ownership of the AI itself.
Scholars have introduced four possible Artificial Invention’s Patent Disclosure Scenarios , all of which are a combination of the following four distinct requirements: (1) Un-disclosed AI-generated output, (2) Disclosed AI-generated output, (3) Disclosed AI-based tool, and (4) Undisclosed AI-based tool.[7]
The patent system serves to provide more knowledge to society than would otherwise be the case.[8] If, however, the disclosure is truly ancillary and often comes into tension with the more important goal of incentivizing innovation, could social welfare conceivably be improved by jettisoning the requirements of United States Code section 112? Arguably, no.[9] At its core, the scope of the disclosure to have the patent monopoly rights granted is one of the challenges for AI inventions. The way we satisfy the second scenario’s requirements is by allowing the AI machine to explain its invention. After all, if the AI machine can create the invention, it can leave a footprint that guides the third party to reproduce that invention. It is fair to claim that the second scenario mentioned above is in line with public policy and satisfies the primary aim of patent law, which is to balance the interests of inventors on one side and the interests of the public on the other. Thus, the second scenario protects social welfare and utilitarian theories.[10]
If the AI machine can invent, it can also log its steps while the invention is in process. After the completion of the invention process, a guided AI with a supervised method of log analysis and deep learning can filter, compile, and optimize the log created during the invention process. It can be seen as reverse engineering the AI invention using the log file that recorded all related steps during the invention process. Finally, this AI machine can produce a detailed disclosure regarding the subject invention without opening the BlackBox of the inventor. Thus, to satisfy the USPTO’s disclosure requirement, we can allow the guided AI to define what steps and specifications are required to disclose and accordingly generate the disclosure. By using this method, the AI machine owner will produce the AI-generated disclosure to the patent office but can keep algorithms of the AI machine undisclosed. This approach will keep the algorithm of the AI machine itself in a BlackBox, and it will simultaneously produce the required disclosure, including a detailed description of the invention and reproduction process. Accordingly, the solution mentioned above will provide the intended social benefit and be in line with the public policy behind patent law.[11]
[1] Joe McKendrick, How Fast Is Artificial Intelligence Growing? Look At The Key Bellwethers, Forbes (Dec. 19, 2018, 12:26 AM), www.forbes.com/sites/joemckendrick/2018/12/19/how-fast-isartificial-intell…-bellwethers/?sh=2e5f8977474a.
[2] See Fiona Macdonald, Stephen Hawking Says Most of Our History Is “The History of Stupidity”, Science Alert (Oct. 21, 2016), www.sciencealert.com/stephen-hawking-says-most-of-our-history-isthe-history-of-stupidity.
[3] AJ Willingham, Artificial Intelligence can’t technically invent things, says patent office, CNN, (Apr. 30, 2020), www.cnn.com/2020/04/30/us/artificial-intelligence-inventing-patent-office-trnd/index.html.
[4] Patents Basics, Confidentiality, World Intellectual Property Organization,
www.wipo.int/patents/en/faq_patents.html.
[5] Bernard Marr, Can Machines And Artificial Intelligence Be Creative?, Forbes, (Feb 28. 2020), www.forbes.com/sites/bernardmarr/2020/02/28/can-machines-and-artificial-intelligence-be-creative/?sh=32b538eb4580
[6] See Thaler v. Vidal, 43 F.4th 1207, 1209 (Fed. Cir. 2022), “When a statute unambiguously and directly answers the question before the court, the court’s analysis does not stray beyond the plain text. Here, Congress has determined that only a natural person can be an inventor, so artificial intelligence cannot be.” (The case, Thaler v. Vidal, arose from two patent applications filed in 2019 by Stephen Thaler, naming an AI system (DABUS) (for “Device for the Autonomous Bootstrapping of Unified Sentience”) as the inventor.)
[7] Ebrahim, supra note 8, (categorizing the four possible combinations of the requirements as (1) provides an undisclosed AI-based tool, which refers to the claiming of an AI-based tool along with disclosure of the AI itself and how the AI operates; (2) provides an AI-based tool without disclosure of how the AI operates; (3) provides an application of an AI-based tool to generate output that is claimed in a patent application, where there is disclosure of the AI-generated output; and (4) provides an application of an AI-based tool to generate output that is claimed in a patent application, but there is no disclosure of the application of AI.
[8] See Rebecca S. Eisenberg & Arti K. Rai, Harnessing and Sharing the Benefits of State-Sponsored Research: Intellectual Property Rights and Data Sharing in California’s Stem Cell Initiative, 21 Berkeley Tech. L.J. 1187, 1194–95 (2006); Anthony J. Mahajan, Note, Intellectual Property, Contracts, and Reverse Engineering After ProCD: A Proposed Compromise for Computer Software, 67 Fordham L. Rev. 3297, 3317 (1999).
[9] Id.
[10] See Jay P. Kesan, Economic Rationale for the Patent System in the Current Context, 22 GEO. Mason L. Rev. 897, 898–99 (2015) (explaining the utilitarian economic theory for patents is based on incentives, such that innovation would be at suboptimal levels absent the incentives.)
[11] Omid Valinasab, Big Data Analytics to Automate Patent Disclosure of Artificial Intelligence’s Inventions, 27 U.S.F. Intell. Prop. & Tech. L.J. 133 (2023).