Daniel Y. Lee is a J.D. Candidate, 2021 at NYU School of Law.

The age of AI-inventors may already be here.  This year saw the first patent applications filed in the name of an AI tool named “DABUS” disclosing designs for a beverage container and a lighting device.  In filing the applications, the people behind DABUS raised a provocative question of whether a nonhuman entity (DABUS in this case) can be named an inventor on a patent application.  The question is one of many AI-related issues on which the USPTO is seeking comment at the time of this writing, though there are some textual bases for it to be decided in the negative such as the statutory definition of an inventor as an “individual” under 35 U.S.C. § 100 (f), and a “person shall be entitled to a patent” under 35 U.S.C. § 102.  Beyond the inventorship issue, however, the prospect of an inventive AI also raises interesting questions about (1) the concept of a Person Having Ordinary Skill In The Art (PHOSITA) and (2) the scope of prior art knowledge deemed to be available to the PHOSITA.  

A PHOSITA is a legal fiction, similar to the reasonable person in tort law, who serves as the judge of whether a new development is obvious and therefore not deserving of a patent.  As such, PHOSITA plays a central role in granting and determining the validity of a patent.  Because the level of skill of a PHOSITA depends on the technology at issue, “[r]esolving the level of ordinary skill in the pertinent art” is a key step in the obviousness inquiry under Graham.  Graham v. John Deere Co., 383 U.S. 1, 17 (1966).  Under current case law, the level of ordinary skill is resolved by considering multiple factors that include, among others, the educational level of the inventor and others in the field, type of problems encountered in the art, and prior art solutions to those problems.  See Envtl. Designs, Ltd. v. Union Oil Co., 713 F.2d 693, 696 (Fed. Cir. 1983). 

However, as potential inventors start using AI not only as a tool to help them design and simulate potential inventions, but as a means to generate novel designs with minimal human intervention, the role of inventors may shift from generation of ideas from the ground up to directing and providing feedback to the design-generating AI and determining which of the AI-generated designs satisfy the inventors’ criteria.  DABUS-generated patent-pending beverage container and lighting device designs showcase such inventive paradigm where DABUS, with human-selected training data, generated the designs, which were in turn selected by the human operators for patenting.

In such AI-driven inventive paradigm, the level of sophistication of AI may become just as important as the education level of the inventor and others in the relevant field.  For example, the people behind DABUS were experts in AI, not people of ordinary skill in the art of beverage container and lighting device design.  Even under the prevailing view of AI as an inventive tool and not an inventive entity, the fact that the non-subject matter experts behind DABUS were able to ‘invent’ by selecting two designs for patenting hints at potential wrinkles to the PHOSITA standard in light of AI.  Should a PHOSITA then be a person having ordinary AI?  Or perhaps the question that should be asked is whether such AI-generated or AI-assisted invention deserves the same level of protection as human-generated invention in a Lockean labor theory sense, and whether such invention needs to be incentivized under the patent regime.  For the time being, the current statutory regime under 35 U.S.C. § 103 treats all manner of inventing the same, dictating that “[p]atentability shall not be negated by the manner in which the invention was made.”    

Another implication of the AI-driven inventive paradigm is the breadth of knowledge presumed to be available to a PHOSITA.  A PHOSITA is famously “envisioned as working in his shop with all the prior art references — which he is presumed to know — hanging on the walls around him.”  Standard Oil Co. v. Am. Cyanamid Co., 774 F.2d 448, 454 (Fed. Cir. 1985).  This alluring image has been tempered to limit the presumption of knowledge to analogous prior art given that no human can possibly consider the entire universe of prior art including references completely unrelated to the invention at hand.  However, an AI on the other hand can parse much more prior art than a human inventor can ever hope to consider in generating a new design.  Not bound by the limited volume (and thereby breadth) of prior art that a human inventor can consider in the inventive process, it may be possible for an AI to consider completely disparate, non-analogous prior art in generating new designs.  In fact, it may soon be possible (or perhaps already is) with advanced image and natural language processing to train an AI to consider all of the 10 million US patents granted to date in generating potentially new designs, without being cabined to particular technology classifications.  When such AIs find widespread use among inventors, the limiting of obviousness analysis to analogous prior art may no longer be needed, or even appropriate.

Only time will tell whether and when truly inventive AI will arrive, and the full extent of the impact it will have on the inventive paradigm and the PHOSITA standard.  When it arrives, however, we will surely have to retire the oft-cited phrase from KSR that a PHOSITA is “a person of ordinary creativity, not an automaton.”  KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398, 421 (2007). 

If this post piqued your interest in the implications of inventive AIs on patent law, the article “Everything is Obvious” by Professor Abbott published in the UCLA Law Review may be of interest.