MHSc in Translational Research

Artificial Intelligence (AI), more commonly known as machine learning is the next holy grail of technology. Its advent is bringing about a paradigm shift in our lives. Dr. Gabriella Chan, our inspiring faculty member discusses AI and its numerous facets. 

Gabriella Chan| TRP | January 15, 2020

AI ‘outperforms’ doctors diagnosing breast cancer” read the headline on BBC’s Health page just two days into this new decade. Highlining a paper published in Nature, the article described how an Artificial Intelligence (“AI”) model trained on 29,000 X-ray images bested 6 radiologists in reading mammograms.
Andrew Ng,  CEO of Landing AI and refers to AI as “the new electricity.” While noting that virtually all industries will be transformed by AI, he sees AI’s biggest and most imminent untapped opportunities in agriculture, healthcare, and manufacturing.
Most references to AI, including the model in the Nature study featured by the BBC article, are directed to deep learning algorithms that train on structured data. But for Andrew Ng, the holy grail of AI is “effective unsupervised learning,” meaning algorithms that learning organically, without labelled data.
As we await the dawning of the AI holy grail, meet DARBUS (Device for the Autonomous Bootstrapping of Unified Sentience), and its creator, Dr. Stephen Thaler, from the University of Surrey in the United Kingdom (UK). Dubbed a “Creativity Machine” by its inventor, DARBUS is a patented neural network trained on general information (NN1). NN1 generates novel ideas which are then monitored and assessed by a second “critic” neural net (NN2) for their novelty, utility, and value.Dr. Ryan Abbott likens interactions as those between NN1 and NN2 to the human brain’s cognitive circuit embodied by the thalamocortical loop. The Artificial Inventor Project, headed by Dr. Thaler, field a number of patent applications through the Patent Cooperation treaty (PCT), the United States Patent and Trademark Office (USPTO), European Patent Office (EPO), and United Kingdom Intellectual Property Office (UKIPO) for inventions created by  DARBUS. In one instance, DARBUS invented an environmentally-friendly fractal food container and in another a device for attracting enhanced attention with applications in search-and-rescue operations. Both patent applications named DARBUS as the sole inventor and Dr. Thaler as the assignee (owner).
In early December 2019, the UKIPO refused both applications explaining that by naming the machine as an inventor, the application does not meet the inventor identification requirement of the UK Patents Act. Although not cited by the examiner, but brought to his attention by Dr. Thaler’s attorney as part of his argument in support of a sole AI inventor, the recently amended Paragraph 3.05 of the Formalities Manual also states that:
“Where the stated inventor is an ‘AI Inventor’, the Formalities Examiner [should]request a replacement. An ‘AI Inventor’ is not acceptable as this does not identify ‘a person’ which is required by law. The consequence of failing to supply this is that the application is taken to be withdrawn under s.13(2)”. (Emphasis is mine)
In late December 2019, the EPO also refused the applications, noting on its website that it had rejected the applications because the inventor designated in the applications is not human. A more fulsome explanation from the EPO may be forthcoming this month.
At law, a reference to a “person” may mean a natural person (i.e. a human) or an artificial person (i.e. a body corporate, partnership, government body, etc.) or both. To avoid the ambiguity of “person” when referring to a human, lawyers are careful to use “individual”. Since patents were intended to issue to natural persons to prevent corporate inventorship, an assumption has always been made that when patent laws do refer to a “person” (as in the above quote) or to an “individual”, the target is a human. The academic debate about whether a non-human inventor or creator can be classified as an individual or a person, particularly under patent and copyright law, has been raging for some time. Listing DARBUS as a sole inventor laid the debate on the doorstep of patent offices.
The need for policy and then legislative action in the face of a technological tsunami, which our current IP laws were not designed to withstand, is palpable in these debates. Change is inevitable. In fact, just weeks after the UKIPO’s rejection and just 7 days prior to the EPO’s rejection of the DARBUS applications, the World Intellectual Property Office (WIPO)  launched a public consultation on AI and Intellectual Property (IP) policy. Comments are due by February 14, 2020.
The consultation paper identifies and invites discussion on 7 IP policy questions and issues for AI-generated IP. In summary:
Issue 1: Inventorship and Ownership:

  • Should the law permit or require that an AI application be named as the [sole or joint][1] inventor or should it require that a human being be named as the inventor?
  • Should inventions that have been generated autonomously by an AI application be eligible for patent protection?

Issue 2: Patentable Subject Matter and Patentability Guidelines

  • Should specific provisions be introduced for inventions assisted [or created solely][2] by AI or should such inventions be treated as other computer-assisted inventions are treated?

Issue 3: Inventive Step or Non-Obviousness

  • Should the standard of a person skilled in the art be maintained where the invention is autonomously generated by an AI application or should that standard be replaced by an algorithm trained with data from a designated field of art?
  • What implications will have an AI replacing a person skilled in the art have on the determination of the prior art base?
  • Should AI-generated content qualify as prior art?

Issue 4: Disclosure [assuming one would seek to patent the AI rather than keeping it confidential][3]

  • In the case of machine learning, where the algorithm changes over time with access to data, is the disclosure of the initial algorithm sufficient?
  • Should the data use to train an algorithm to be disclosed or described in the patent application?
  • Should the human expertise used to select data and to train the algorithm be disclosed?

Issue 5: General Policy Considerations for the Patent System

  • Should a separate system of IP rights for AI-generated inventions be developed to adjust innovation incentives for AI?

Issue 6: Authorship and Ownership

  • Should copyright be attributed to original literary and artistic works that are autonomously generated by AI or should a human creator be required? In whom should copyright vest?

Issue 7: Infringement and Exceptions

  • Should the use of copyrighted data and works without authorization for machine learning constitute an infringement of copyright?
  • How would the unauthorized use of data subsisting in-copyright works for machine learning be detected and enforced, when a large number of copyright works are created by AI?

Keeping in mind that IP law is territorial and only humans and legal entities can own property, including intellectual property, these discussion issues are sure to spark meaningful, and hopefully impactful, debate. The IP and AI community will await the results of this consultation with bated breath.
In the meantime, for an in-depth argument in support of the equivalency of artificial inventors and human inventors under the law see Ryan Abbott, I Think, Therefore I Invent: Creative Computers and the Future of Patent Law, 57B.C.L. Rev. 1079 (2016).
[1]My addition.