Dramatic progress by machine learning in fundamental tasks, such as image and language recognition, has triggered an explosion of artificial intelligence (AI) applications from self-driving cars to medical diagnostic tools.
According to a UK IP Office (UKIPO) report, “Artificial Intelligence: A worldwide overview of AI patents and patenting by the UK AI sector”, published patent applications relating to AI have risen 400% over the past decade. Elsewhere, the World Intellectual Property Organisation (WIPO) has begun a series of “conversations” over AI and IP. A recurring theme is whether creations by AI should be protected by copyright, design right, patents, or new sui generis rights—or not at all.
There are well-known “arguable” examples of invention by AI, such as an unusual but effective antenna designed for NASA in 2004 by “evolutionary” software, and at least one patent granted has been attributed to an inventive AI. US patent number 6,847,851, granted in 2005, covers a circuit design that the named inventor, John Koza, later revealed was designed by genetic programming.
These and other examples were collected more than a decade ago in Robert Plotkin’s “The Genie in the Machine: How Computer-automated Inventing is Revolutionizing Law and Business” (2009).
AI can generate inventions humans cannot
It seems AI can generate inventions that a human alone would not. For his book, Plotkin interviewed Gregory Hornby, a scientist at NASA Ames Research Centre, about the antenna design and was told: “No human engineer would have thought of an antenna that looked so crazy, yet the antenna works better than previous human designs.”
There has been no exploration so far in case law of the boundary between AI-assisted invention and invention by AI, and there is debate over whether autonomous inventions by AI are a realistic prospect.
For example, representatives of IBM and the International Association for the Protection of Intellectual Property raised doubts at the second WIPO conversation on AI and IP. Meanwhile, the WIPO has noted the potential need to agree definitions for “AI-generated”, “autonomously generated by AI”, “AI-assisted”, etc.
“Applying established procedure, the UK examiner did not question the applicant’s statement that Dabus was the inventor.”
Matt Hervey, Gowling WLG
The Artificial Inventor Project led by Ryan Abbott at the University of Surrey in the UK is pursuing patent applications naming “Dabus” (Device for the Autonomous Bootstrapping of Unified Sentience) as the sole inventor, an AI application created by physicist Stephen Thaler.
Significantly, care was taken to create circumstances in which the inventions are said to have been autonomously generated by the AI. The applications have been rejected by the US Patent and Trademark Office, the European Patent Office and UKIPO, but these are under appeal, and applications are pending in Canada, China, Germany, India, Israel, Japan, South Korea and Taiwan.
Protecting inventions by AI
The Dabus project is a brilliant attempt to further the debate on inventive AI, driving close scrutiny of patent law, but the scope of the decisions has largely been limited to confirming that patent laws were not intended to protect inventions by AI.
This is not surprising. In the UK, for example, the Patents Act was passed in 1977, when invention by AI was most probably not even considered. In its appeal to the UKIPO, for which judgment is awaited, the Dabus team skilfully traversed a very narrow path through all the legislative requirements to show that none expressly precludes patents for inventions by AI—but that is not the same as showing the law allows them.
Whether to protect such inventions is a matter of policy and needs broad consideration, including economic analysis of whether inventions by AI should be protected to encourage investment, how long protection should last and what thresholds should apply for, eg, novelty, inventiveness, investment, etc. As the UKIPO hearing officer observed, protection for inventions by AI should not be “not shoehorned arbitrarily into existing legislation”.
- 400%: The global increase in published patent applications relating to AI over the last decade (source: UK Intellectual Property Office)
Applying established procedure, the UK examiner did not question the applicant’s statement that Dabus was the inventor so there was not need to consider the legal principles governing the border between “AI-assisted” and “AI-generated” inventions.
The approach is yet to be determined, although the principles of entitlement to be a joint inventor might be applied. Establishing whether there was a human inventor is likely to involve complex issues of fact.
Moreover, the status of patents wrongly naming a human inventor warrants examination. In the US, such a patent might be unenforceable but there may be no mechanism to challenge such a patent in the UK.
Consider a full suite of IP to protect inventions
There are already plausible circumstances in which the human contribution to an invention may not attract patent protection. For example, if an AI makes a drug candidate plausible, further human investigations such as clinical trials may not amount to “devising” the invention.
Pending further clarity, those seeking to recoup R&D investments involving AI should consider the full suite of IP rights, trade secrets, contractual measures, technological safeguards and sector-specific protections, such as data exclusivity for clinical trial data.
Dabus may accelerate any appropriate legislative reform, although changes tend to lag technological developments by many years. However, it must be suspected that the potential negative impact of such delays to those seeking patent protection for inventions arguably or actually by AI will sometimes be avoided, in practice, through naming humans as inventors.
Matt Hervey is head of artificial intelligence and partner at Gowling WLG. He can be contacted at: firstname.lastname@example.org
Image: Shutterstock.com / DamianArt