Could my AI beat me to the patent races?
Ronald Ker Wei Yu – Faculty of Law, The Chinese University of Hong Kong
-- Artificial intelligence (AI) and intellectual property (IP) has become a hot topic of late.
In December 2019 the World Intellectual Property Organization (WIPO) launched a public consultation process on AI and IP policy, inviting feedback to help define the most-pressing questions likely to face IP policy makers as AI increases in importance. This followed an August 2019 consultation by the United States Patent and Trademark Office (USPTO) about the impact of AI on intellectual property law and policy.
The questions of whether an AI could invent something on its own and whether such an invention would be patentable recently arose when a team from the University of Surrey recently filed two patent applications in the UK Intellectual Property Office (UKIPO) and the European Patent Office (EPO) naming an AI system called "DABUS" (Device for the Autonomous Bootstrapping of Unified Sentience) as inventor.
The first question, whether an AI could invent something on its own, was apparently answered when DABUS "invented" a fractal container capable of changing shape and a flashlight (a "neural flame") for use in search-and-rescue missions raising the follow up question of whether an AI could get a patent.
To be patentable, an invention must be new, inventive, have utility and not be unpatentable subject matter. There also are disclosure requirements – important because one function of patents is diffuse the knowledge in the invention by requiring the inventor to publicize the technical ideas in sufficient detail to allow a skilled person to reproduce the invention.
Both the EPO and UKIPO eventually rejected the DABUS applications, the EPO noting that the inventor must be a human not a machine, echoing the decision of the US 9th Circuit in Naruto v. Slater, No. 16-15469 (9th Cir. 2018) that denied copyright authorship to a monkey that had taken a photo of itself and US. Copyright Office’s statement that “[to] qualify as a work of ‘authorship’ a work must be created by a human being.”
Yet the EPO’s decision regarding the DABUS patents was criticised as a major obstacle standing in the way of a new era of human endeavour and that the WIPO consultation is still proceeding suggests that questions regarding the patentability of inventions created by AI remain.
Issues of humanity is set aside, what of the other challenges to patentability?
First, consider how AI systems work. Modern AI systems generally fall into two camps: the first are systems where algorithms are input to perform the function, the second group are provided large data sets then find patterns from that data – i.e. machine learning systems – a process that is uncomfortably similar to the process of discovery; a concern given discoveries are not patentable (though useful applications of discoveries are).
Outputs of machine learning can be presentations of information or methods of doing something – i.e. algorithms - which, by themselves, are not patentable and in the case of machine learning generated algorithms can be difficult to understand, particularly with systems employing deep neural networks, leading to questions whether this could be a hurdle to disclosure.
Then there are the questions of inventorship and ownership.
Some patent laws require the naming of the inventors (e.g. the US), but this poses a dilemma. Does one name an AI inventor in a patent application and risk subsequent invalidation of one’s patent because an AI cannot be an ‘inventor’ for patentability purposes because it is not human, or not name the AI and risk later invalidation for not disclosing the inventor?
As for ownership, while copyright law might seem at first to provide some guidance – for example Sec. 11(3) of Hong Kong’s Copyright Ordinance notes that the party using a computer that creates a work would receive copyright rights, no equivalent provision exists in its patent law.
WIPO, as noted earlier, is still pondering the matter and has several additional questions to consider, such as:
· Should specific provisions be introduced for inventions assisted by AI?
· Given the standard is whether the invention obvious to a person skilled in the relevant art, what is the relevant art?
· Is AI generated content prior art?
· Should the standard of a person skilled in the art be changed?
· What are the issues regarding disclosure?
· How should the training data be treated?
· Should human expertise used to select training data be disclosed?
There are also policy questions as to whether allowing AI inventions to be patented widens the technology gap between countries or whether local patent offices have the expertise to prosecute patents for inventions created by AI.
But for now, it looks like AI will not be beating us to the patent races.