[Opinion] Can an AI system be an inventor ?

The AmeriKat braces for the take
over of the machines....or more likely not...

There has been much headline ink spilled on the question of AI-inventorship in the IP press and beyond.  Kat friends Professor Kelvin FK Low (National University of Singapore), Professor Wai Yee Wan (City University of Hong Kong) and Associate Professor Ying-Chieh Wu (Seoul National University) examine the recent Australian and UK cases and critical analyzes the arguments on both sides in an attempt to answer that question and how (or whether) AI-inventorship incentivizes invention.  Over to the Professors:

"There is an increasing influential and bludgeoning legal literature on how artificial intelligence (AI) systems should be treated in law. One question that has recently been in the headlines around the world, thanks to the Artificial Inventor Project, is whether or not an AI system can be regarded as an inventor. The Artificial Inventor Project is championed by Ryan Abbott, who has argued in “I Think, Therefore I Invent: Creative Computers and the Future of Patent Law” (2016) 57 Boston College L Rev 1079 that while an AI system is not yet a legal person, it should nevertheless be acknowledged as an inventor, with any patent it produces being allocated to its owner. The headlines are the result of the Project’s filing of parallel applications to patent offices in several jurisdictions over a number of inventions generated by an AI system named DABUS (Device for the Autonomous Boot-strapping of Unified Sentience), invented by Dr Stephen Thaler. In our recent paper, we critique Abbott’s proposal whilst contemplating AI’s status as property or person.

It is perhaps most interesting to compare the contrasting fortunes of the Project’s filings in Australia (Thaler v Commissioner of Patents [2021] FCA 879) and England and Wales (Thaler v The Comptroller-General of Patents, Designs And Trade Marks [2020] EWHC 2412 (Pat)/Thaler v Comptroller General of Patents Trade Marks And Designs [2021] EWCA Civ 1374). In Australia, Beach J accepted that Dr Stephen Thaler, the inventor of DABUS, could be said to either be entitled to have the patent assigned to him or be a person who derived title from the inventor, thus overcoming the difficulty that even if DABUS could be regarded as an inventor, it could not itself own a patent owing to its lack of legal personality. The convoluted contortions that Beach J engaged in to arrive at this conclusion was rejected in England. The argument is perhaps best summed up by Marcus Smith J, who described it as “hopeless” because “DABUS would – by reason of its status as a thing and not a person – be incapable of conveying any property to Dr Thaler. In short, the ability to transfer, which DABUS lacks, is fatal to Dr Thaler’s contentions. The same point can be put in a different way: because DABUS is a thing, it cannot even hold property, let alone transfer it.”

Setting aside issues of statutory interpretation, Abbott’s proposal (and hence the Thaler litigation) rests on two prongs, both deeply flawed. First, an inapt analogy to accessio, an aspect of which deals with “ownership of the progeny of animals or the treatment of fruit or crops produced by the labour and expense of the occupier of the land (fructus industrialis)”. Many non-experts assume that these rules have an air of natural law to them and deduce implications which do not follow. As Arnold LJ clarified, these instances of accessio “all concern new tangible property which is produced by existing tangible property.” It could not apply to intangible property produced with the aid of tangible property else much of our intellectual property law will have to be rewritten. If I take a photo with your camera, I own the copyright to the photo, even if I used your camera without permission (though I would then be liable in trespass).

These rules are not rules of nature but rules of policy full of exceptions. The rule that severed crops belong to the person with the right of exclusive possession at the time of severance is subject to the emblements exception, whereby emblements – crops which are required to be sown – may be taken by a tenant who has sown them even if they are uncut at the termination of the lease. And partus sequitur ventrem, often wrongly regarded as positing that the offspring of an animal belong to the owner of the mother (male seahorses carry their offspring through gestation), does not apply to cygnets (The Case of Swans (1572) 7 Co Rep 15b, 77 ER 435), which offspring are shared by owner of cock and hen supposedly because swans mate for life. This arguably more natural rule for animals that reproduce sexually suggests both that partum sequitur ventrem is one of convenience and that it can potentially be applied to other monogamous species (e.g. seahorses). The rule is based on possession rather than ownership as demonstrated by its application in cases of chattel (animal) hire, where ownership of offspring is assigned to an animal’s hirer rather than its owner (Tucker v Farm and General Investment Trust Ltd [1966] 2 QB 421). While Beach J suggests possession extends to inventions in abstract, i.e. information, this muddles the way possession operates with respect to things and information. Possession of things (rivalrous) entails the exclusion of others. Possession of information (non-rivalrous) does not.

Secondly, Abbott relies upon policy considerations. Before addressing these, however, it is important to dismiss the implicit suggestion by Beach J that a rejection of Abbott’s arguments would lead to vast swathes of inventions assisted by AI being unpatentable. Dr Thaler had previously listed himself as inventor in relation to inventions generated by an earlier iteration of DABUS, an arrangement Abbott acknowledges is accepted practice: “a person can qualify as an inventor simply by being the first individual to recognize and appreciate an existing invention.” This practice enables accidental inventions to be patented, which is significant because accidental inventions often lead to more significant advances than deliberate ones.

Abbott’s suggestion that his proposal incentivises the sharing of AI resources is dubious (contra Michael Schuster, “Artificial Intelligence and Patent Ownership” (2018) 75 Wash & Lee L Rev 1945). He belittles the contribution of end users (who must ask the right questions of the right AI system), grossly underestimates the bargaining position of owners of AI machines (who can effectively dictate terms of access so that it is not necessary for the law to allocate patents to them by default), neglects to consider the impact of such a policy on end users of AI (who may be disincentivised to seek access), and omits consideration of innocent non-consensual access (who arguably continue to deserve being awarded patents but be liable in tort for trespass). Abbott’s examples of “interlopers”, such as interns or visitors intercepting patents are deeply flawed since owners can obviously control access and a duty of confidence will prevent interlopers from doing so, which Birss LJ in the English Court of Appeal noted. Indeed, the status quo is far more efficient than Abbott appreciates even if it may be counterintuitive to laypersons. 

Should an owner of an AI system fail to appreciate the value of an AI-generated output, a subsequent person who happens upon the discarded output can claim inventorship through disclosure. After all, as Margaret Boden explained in “Could a Robot Be Creative – And Would We Know?” in Kenneth M Ford, Clark Glymour, and Patrick J Hayes (eds), Thinking About Android Epistemology (MIT Press) 217, “to recognize the creativity of a creative robot we would need at least to share its conceptual spaces, if not its values too.” Such a person may not be an AI system’s owner. But if patents are by default allocated to the owner of an AI as Abbott suggests, no one would be incentivised to claim inventorship."