Gunjan Agarwal of Fox Rothschild (Photo: Courtesy Photo)

Sundar Pichai, the CEO of Alphabet, Inc., recently pronounced artificial intelligence (AI) as being "more profound than fire or electricity." While AI is rising as a key commercial player at the global scale with an expected market size of almost $400 billion by 2025, are patent laws around the world equipped to incentivize this revolution?

From Siri to self-driving cars to computer-assisted drug discovery, AI is progressing rapidly, and it has provided competitive intelligence for solving some of the world's greatest challenges. As AI systems advance, they have also started emerging as independent developers of better cars, drugs, personalized products and various research tools. Specifically, AI systems can evolve over time to autonomously contribute to the conception and development of an idea outside of their creator's expertise or original intent. For example, a drug designed and discovered solely by AI systems, without any human interaction, is about to enter clinical human trials for the treatment of obsessive-compulsive disorder, see Rachel England, "AI-Formulated medicine to be tested on humans for the first time," Engadget (2020). Similarly, AI systems have independently developed new artificial computer languages that are more efficient in performing the assigned tasks (e.g., translation), eschewing human languages, see Sam Wong, " Google Translate AI Invents its own language to translate with new scientist" (2016). The role of AI as a creator of new inventions again brings to attention the controversy surrounding ownership and inventorship rights relating to patents filed to protect such AI-generated inventions. It is particularly important to address these issues because key components of an AI system—datasets and ever-evolving AI algorithms—often do not qualify for patent protection, leaving only the output of the AI system as a potentially patent eligible invention.

Earlier this year, the European Patent Office and the U.K. Intellectual Property Office rejected two patent applications in which an AI system (DABUS) was named as the inventor. Both patent offices considered the inventions themselves worthy of patents, but rejected the applications because the "inventor" was not a human. While the definition of an inventor may differ slightly, most patent offices and courts around the globe interpret the term "inventor" to refer to human beings. While this traditional paradigm easily conforms to the typical scenario when AI is a tool that helps human inventors by, for example, creating models for decision making in self-driving cars or discovering new drugs, it creates a grey area when an AI system is fully responsible for the invention without human involvement. The U.S. Patent and Trademark Office (USPTO) recognizes this conundrum, and sought public opinion on whether current patent laws and regulations regarding inventorship need to be revised to account for situations when a non-human entity qualifies as an inventor. (See Request for Comments on Intellectual Property Protection for Artificial Intelligence Innovation, 84 Fed. Reg. 210 (Oct. 30, 2019)).

The U.S. Constitution empowers Congress "to promote the progress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries." Unlike many other countries, inventorship of a patent is central to the U.S. patent system, and ownership of patent applications initially vests in the inventors. A patent application cannot be filed without naming at least one inventor—who can only be a natural person. Given these requirements, entities who have invested considerable money and resources in developing an AI system, which can invent without human intervention, can either forego patent protection in lieu of other types of intellectual property protections, or improperly name a human as an inventor.

Development of an AI system with humanoid intelligence that can create and develop inventions by itself is a financially expensive endeavor requiring hundreds of thousands dollars. (See, e.g., David Ferrucci et al., " Building Watson: An Overview of the DeepQA Project," AI MAG., Fall 2010, at 59 (development of IBM's Watson required "three years of intense research and development by a core team of about 20 researchers")). Patent rights can play a crucial role in promoting investment in this nascent field by providing exclusive competitive advantages as well as direct financial incentives such as licensing revenues, which are unavailable through other types of IP protections. In certain fields, pursuing other avenues of IP protection is often not feasible, and foregoing patent protection would prevent breakthroughs in important areas due to lack of funding.  For example, in highly regulated fields like pharmaceuticals, it is impossible to keep an invention a trade secret. Unavailability of patent rights is also counter to the basic Constitutional rationale for granting patents—dissemination of knowledge and promotion of public access to new technology, and may lead to delays in key inventions reaching the public.

Under U.S. patent laws, failure to correctly name the inventor(s) of patented subject matter can result in a patent being held invalid, change in the ownership of the patent, or even fine or imprisonment for an inventor who incorrectly signs a formal declaration of inventorship. (See e.g., Ethicon, Inc. v. U.S. Surgical, 135 F.3d 1456, 1465 (Fed. Cir. 1998); See 35 U.S.C. Section 115 (2012)).  As such, if a human is improperly named as an inventor on a patent application for which an AI system satisfies the criterion for inventorship, the patent application may at first be rejected by the USPTO, or if granted the patent may eventually be held invalid. (See In re Verhoef, No. 2017-1976, 2018 (Fed. Cir. May 3, 2018)). Similarly, omitting an inventor— when an AI system qualifies as an inventor but is not identified on the patent—can render a patent enforceable. (Frank's Casing Crew & Rental Tools, v. PMR Technologies, 292 F.3d 1363, 1376 (Fed. Cir. 2002)). Finally, the requirement of correctly naming inventors also raises questions about who should be named as the human inventor when the AI system is not merely a general tool—a computer scientist who initially codes and/or trains a model for the AI system, a user who recognizes a real-world application of the AI system, or the end user of the AI system who recognizes the output from an AI system as an invention, or an entity that owns the AI system? For example, the AI-generated invention may be so far removed from the initial conception of the AI system, in time and structure, that it would not be fair to name the initial computer scientist as an inventor.

As it stands, there are no clear options for securing patents for AI-generated inventions. For incentivizing development of creative machinery that can independently generate patent eligible technologies, there is a need for either an expansive interpretation of existing patent laws and/or revision of patent laws. Such reforms may either recognize AI systems as inventors, separate and apart from human operators; or change the definition of inventorship to allow naming of humans operating in a defined relationship with the AI system as inventors (which will also provide oversight and control of AI). It is clear that the question regarding which human should be named as an inventor cannot have a one-size fits all solution.

Practical Considerations: While lawmakers resolve issues regarding inventorship, there is an interim need for protecting inventions created by AI systems. As patent practitioners, consider the following while drafting patent applications for enabling inclusion of human inventors, even if part of the claimed processes or inventions are created by the AI system:

• Draft patent applications to not only describe the invention generated by an AI system, but also the structure, training, and execution of the AI system itself (i.e., describe structures and processes that requires human intervention).
• Strategically draft claims directed to the AI-generated invention to emphasize how a human- engineered AI system structure generates the invention. This requires drafting each independent claim to include a patentable hook that relates back to a human involved with the AI system. For example, relate the final structure of the AI system to the human coded model. Similarly, emphasize human contribution to claims by including parameters, features, weights, etc. provided by a human operator that modify or improve the AI-generated invention output. It may well be the case that changing a parameter in the AI system resulted in significant changes in efficacy.
• Separately include claims directed to the training and modeling phase processes of the AI system that often require a human developer.

Finally, a cloud on the inventorship of a patent also calls into question ownership. Avoid ownership pitfalls by:

• Counselling client corporations to modify employment agreements such that employees who design and train AI systems assign patent rights to any AI invention invented by these AI systems to the company that owns the AI systems; and
• Drafting licensing and purchase agreements relating to AI systems that include provisions for ownership of inventions generated by the AI systems.

Gunjan Agarwal is an associate in the intellectual property department of Fox Rothschild. She can be reached at [email protected].