The Rise of 'Centaur' Inventors: How Patent Law Should Adapt to the Challenge to Inventorship Doctrine by Human-AI Inventing Synergies

Abstract

We are at the dawn of a paradigm shift of the inventing process itself. In this new paradigm, human researchers and AI algorithms collaborate and each utilize their own comparative advantages in pursuing a common inventive goal, thus playing complementary roles in a spectrum of Centaur (“human-AI”) Inventing Synergies (“CIS”), achieving an “1+1>2” effect. From an epistemological perspective, human beings have “common sense” and solid understanding of causation, which makes them good at “question framing” and “value assignment”; on the other hand, AI cannot understand causation, yet it has vast capability in weak correlation recognition, which makes it good at “solution prediction”. AlphaFold’s recent breakthrough in predicting 3D protein structures with an accuracy comparable to experimental results is a showcase of AI’s great potential in assisting the inventing process - compared with conventional tools, some AI systems may be capable of “independently” generating technological solution(s) that are novel, “surprising” and potentially valuable, once the specific problem has been properly framed by a human researcher.

This paradigm shift poses practical challenges to various aspects of patent law, including the inventorship doctrine. Under US patent law, the touchstone of inventorship is conception, which is defined as the formation in the mind of the inventor, of a definite and permanent idea of the complete and operative invention. Meanwhile, case law held that one does not conceive by merely “posing the problem to be solved”, no matter how specific it is. Therefore, in some CIS cases no human inventor can be actually identified, for it is the AI algorithm that comes up with the idea of the “complete and operative invention”. In other words, inventorless inventions (“IIs”) do exist.

Faced with this doctrinal challenge, policy makers may have several choices at hand. One, leave the IIs in the public domain (or subject them to non-IP protection only, e.g. trade secret). After all, no inventor, no patent. This choice is not practical, particularly in light of the ever-increasing commercial values of a range of IIs, as shown in recent AI-assisted drug design and repurpose cases. Two, make AI systems qualify for inventorship through judicial interpretation or legislation, as argued by the advocates in the recent DABUS case. Such a radical approach however, is in fundamental conflict with the deontological foundation of the IP system. In a holistic view of the well-acknowledged IP foundations as well as IP’s operative principles of efficiency and dignity (identified by Robert Merges in his Justifying IP), this Article proposes a third approach, arguing that the definition of “conception” should be updated so that a human researcher in a CIS regime may be deemed as having constructively conceived of the idea, if three pre-conditions are fulfilled simultaneously: (1) She is the one that posed the specific problem to which the AI algorithm eventually predicted the solution; (2) She is the first human being that grasps the non-obviousness of the specific and settled idea of the solution; and (3) She makes an adequate disclosure of the algorithm’s specific role in the inventing process. On a more normative level, this Article further argues that the dignity principle may become even more relevant in the AI-enabled age, and doctrinal rules should be adapted only in ways consistent with it.

Keywords: artificial intelligence, inventorship, patent, human-AI collaboration, centaur inventing synergy