Machines, learning – overcoming the IP hurdles raised by AI

AI is powering transformative changes across a range of industries but patent systems around the world are struggling to keep up

The term ‘artificial intelligence’ was first coined more than 50 years ago – over the ensuing decades, innovators have continued to develop computer intelligence in an effort to realise its potential. Yet it has only been in the past five to 10 years in particular that we have begun to comprehend the power of AI as a key driver behind the Fourth Industrial Revolution. Several key developments have accelerated AI’s growth in recent decades, including advances in computing power available to broad masses, accumulation and accessibility to large amounts of electronic data, and innovation in AI tools such as neural networks.

The AI community has transformed theoretical machine learning into useful problem solving in all walks of life – from AI-assisted pedestrian detection for autonomous driving to enhanced drug discovery. In fact, AI is affecting daily tasks in both visible and invisible ways, disrupting every major industry along the way.

The promise of this new technology is reflected in the exponential growth of investment and revenues tied to it. In 2019 global private AI investment exceeded $70 billion according to “The AI Index 2019 Annual Report” by Stanford University. Meanwhile, investments in AI start-ups have grown at an annual rate of almost 50% since 2010, totalling more than $40 billion in 2018. According to a report by Statista, the global AI software market revenue is estimated to reach $90 billion by 2025 – a growth of more than 12 times the revenues of 2018 (see Figure 1). Overall, PwC reports that the potential impact of AI on the global economy could be up to $15.7 trillion by 2030.

AI advancements exist at many levels – from core algorithmic machine learning techniques to eventual application areas (see Figure 2). As such, companies of all sizes – from nascent start-ups to the Fortune 100 – are actively participating in development activities and partnering with companies across industries to make these advancements possible.

R&D in AI – dubbed by some as the ‘new electricity’– is fuelling the creation of a new digital frontier, resulting in a tremendous amount of intellectual property. The most popular form of IP right in the sector is patents, evidenced by an exponential rise in the number of patents in AI-related technologies. Nearly 340,000 AI-related inventions were published between 1960 and early 2018, while the annual filing rate of AI-related inventions grew by a factor of 6.5 between 2011 and 2017 according to WIPO’s 2019 report on AI.

At the same time, the proliferation of patents in AI-related technologies and the acceleration of AI-powered real-world applications are complicating and fundamentally challenging current IP legal systems.

One core issue concerns inventorship. Historically, patents are awarded to human inventors, rather than the computers or machines used in the inventive process either to assist human inventors as a tool or to perform functions typically associated with humans.

AI is quickly surpassing its ability to acquire fundamental human-like capabilities (eg, speech and vision), and instead is moving towards mastering specialised tasks or cognitive thinking typically performed by humans only. If a creative computer run by powerful AI algorithms identifies a problem and creates a solution (eg, automatically detecting an anomaly for an autonomous driving car or detecting a biological trait that leads to a new cancer drug), is it eligible to be a named inventor?

Current patent systems tend to be constructed around bestowing inventorship status on humans and do not permit computers to be named as inventors. This article discusses the challenges with defining inventorship arising from AI-developed inventions under the existing legal regimes, the potential implications in key application areas such as AI-based autonomous driving and the pharmaceutical industry, and practical considerations for protecting AI-related inventions.

Figure 1. AI worldwide market revenue

AI worldwide market revenueSource: Statista 2020

Figure 2. Global AI industry

Figure 2. Global AI industry

Naming the inventor

Determining inventorship for various kinds of AI-related inventions under the current patent system is complex and challenging. One fundamental issue requiring resolution is whether inventors must be human.

Under the US patent statute, an inventor must contribute to the conception of an invention. In Townsend v Smith (36 F2d 292, 295, 4 USPQ 1269, 271 (CCPA 1930)), the Supreme Court defined ‘conception’ as “the complete performance of the mental part of the inventive act”, while Section 100(f) of the Patent Act defines an ‘inventor’ as “the individual… who invented or discovered the subject matter of the invention”. The term ‘individual’ even excludes legal entities such as corporations because, according to the court in New Idea Farm Equipment Corporation v Sperry Corporation and New Holland Inc (916 F.2d 1561 (Fed Cir 1990)), “people conceive, not companies”. By extension, under the US patent statute, inventors are presumably human. In other jurisdictions, an ‘inventor’ is either defined as an individual, human or person (eg, in China, Japan and South Korea) or undefined entirely (eg, in Europe).

To further understand the legal requirements for inventorship of an AI-related invention, it is helpful to consider the following questions, which the USPTO posed for public feedback in August 2019:

  • What are elements of an AI invention?
  • In what ways can a natural person contribute to the conception of an AI invention and be eligible to be a named inventor?
  • Does current patent law and regulation regarding inventorship need to be revised to take into account inventions where an entity or entities other than a natural person contributed to the conception of the invention?

The USPTO defines ‘AI inventions’ as “inventions that utilize AI, as well as inventions that are developed by AI”. For the former, the utility of AI algorithms in the inventive process is as an aid or a tool assisting human inventors (ie, without contributing to the conception of the invention).

For these inventions, the human inventors are expected to uniquely engage in high cognitive reasoning such as designing new hardware, software or data structures to be used by the AI algorithms (ie, the conception of the invention). This generally supports the idea that the human inventors are expected to be the named inventors.

Nonetheless, with non-human AI techniques demonstrating an increased ability for high cognitive reasoning, inventions that utilise AI are expected to involve an increasing participation of AI techniques in the inventive process. This may include deriving insightful results through self-learning or reinforcement learning.

In the latter group of inventions, a natural person arguably still identifies a problem or input parameters and develops a solution or platform while utilising AI techniques to refine the problem, find solutions or create insightful outputs. This human activity of utilising AI techniques may be considered a manner of making any resulting invention.

The 1952 US Patent Act abrogated any notion of a so-called ‘flash of genius’ doctrine, indicating that what matters is the advancement of science or useful acts achieved by the invention, not the inventor’s mental process: “Patentability shall not be negatived by the manner in which the invention was made” (see also Graham v John Deere Co, 383 US 15 (1966)). Therefore, the human inventors may still be expected to be the named inventors.

As AI advances towards mastering specialised tasks using deep learning (eg, tasks typically performed routinely by human experts), determining inventorship for the inventions developed by such AI becomes more challenging. The AI techniques in these inventions may be considered to supplement or even supplant “the leaps of human ingenuity” by generating insightful outputs and thus arguably contributing to the conception of the invention. In terms of US patent law, this queries whether the human has possession of the invention when AI has learned enough to create a solution for the problem at hand. Does this improvement go beyond the creation of the AI engine conceived by the human?

Although abolishing the flash of genius test indicates that what matters is the advancement of science or useful art achieved by the invention, there are no US statutory laws or judicial decisions on the boundaries of what constitutes the objective test of the advancement achieved by an invention. Further, court decisions have provided no clear standards to differentiate a mind (ie, mental process) that is qualified for inventorship from one that is not. Therefore, applying the individual-being-human assumption as recited in Section 100(f) of the Patent Act, the computers and machines used for the inventions developed by such AI would not be eligible to be named as inventors.

The question of whether inventors must be humans is yet to be answered concretely. Some have advocated extending inventorship to non-human inventors in response to the rapid development of AI. Indeed, Erica Fraser argued in “Computers as Inventors – Legal and Policy Implications of Artificial Intelligence on Patent Law” (Scripted, vol 13(3), December 2016), that “[t]he central justification for the patent system is to provide an incentive for innovative activity and the public disclosure of its results” and that recognising non-human inventors with inventorship appears to encourage innovation and accelerate their real-world applications. The proper balance, if any, between patent rights belonging to non-human inventors and public interests in AI-related inventions and who owns those with non-human inventors remains unsettled.

Europe weighs in

As inventions created by AI machines become more prevalent, the debate over patent rights belonging to non-human inventors is only growing – as highlighted by the recent EPO and UK Intellectual Property Office (UKIPO) refusal of various AI-invented applications. Following a non-public hearing on 25 November 2019, the EPO refused two EU patent applications in which an AI machine called “DABUS” was designated as the inventor.

In particular, the application was refused on the ground that it does “not meet the requirement of the European Patent Convention (EPC) that an inventor designated in the application has to be a human being, not a machine”. The EPO’s ruling refers to Article 81 and Rule 19 of the EPC, both of which concern the designation of the inventor, but neither of which specifically address the possibility of a non-human inventor.

The two EU patent applications filed in the United Kingdom via the Patent Cooperation Treaty were similarly refused by the UKIPO. The Hearing Office of the UKIPO stated that “the Office accepts that DABUS created the inventions set out in the two applications”, but since DABUS is a machine and not a natural person, it cannot be regarded as an inventor for the purposes of Sections 7 and 13 of the Patents Act 1977.

What is more, the novel issues raised by AI-related inventions are not limited to inventorship for patent protection. The UKIPO’s ruling further stated that DABUS has no rights to the invention: “There appears to be no law that allows for the transfer of ownership of the invention from the inventor to the owner in this case, as the inventor itself cannot hold property.”

Nevertheless, this issue of ownership has not been seen in the same way for creative arts created by AI machines or software – at least not in China. A Chinese court recently ruled that an AI-written article is protected by copyright and that the creator of the software that produced the article owns that copyright. The content of the article at issue was created by Tencent’s content automated software Tencent Robot Dreamwriter. The court found that the article qualified for copyright protection because it met the legal requirements (certain originality) to be classified as a written work and the software creator (ie, Tencent) owns the copyright for stories that the software produces.

It is worth noting that current US copyright law has a strict human authorship requirement; namely, it protects only the “fruits of intellectual labor” that “are founded in the creative powers of the mind”. Thus, it remains to be seen whether such a strict requirement could be challenged in the US courts.

The EPO and UKIPO rulings appear to signal that an invention is something that must be carried out by a human. However, a blanket ban of inventorship to AI machines is unlikely to answer the many IP questions posed by AI-related inventions. On the contrary, AI-related inventions have raised highly novel legal issues for patent offices around the world to address and have had a profound impact on business.

For example, is it enough for patent protection to list human beings as the inventors on a patent application regardless of the level of contribution from the AI machine? As IAM editor Joff Wild wrote after the decision on DABUS, it appears inevitable that “many other cases will have to be heard and decided before there is a comfortable level of predictability about what kinds of AI inventions can be protected and, crucially, enforced”. The copyright ruling from the Chinese court further demonstrates the highly complex issues and equally complex outcomes associated with the creativity and power of AI machines.

Behind the wheel

Autonomous driving is reshaping the entire transportation industry and AI is fuelling the development of fully autonomous vehicles by enabling the processing of, and making decisions based on, tremendous amounts of data gathered in real time by sensors and other vehicle components.

Companies are investing billions of dollars in the race for a fully autonomous vehicle and all the associated applications that come with it. Examples include GM’s Cruise Automation, in which Honda invested $750 million and Softbank $900 million, and VW Group’s $2.6 billion investment in Argo AI. In fact, a 2019 report on AI trends by CB Insights states that the autonomous vehicle market is projected to reach roughly $80 billion by 2025.

Given the level of R&D, it is no surprise that AI inventions related to autonomous driving have one of the highest growth rates among AI-related patent applications, according to WIPO’s 2019 Technology Trends on AI. Specifically, from 2007 to 2018, autonomous vehicle AI inventions experienced a 42% increase, with more than 5,500 filings in 2016 (see Figure 3).

According to one study, in 2019 the automotive industry was on pace to file more than 20,000 patent applications on critical components for autonomous vehicles (see Figure 4).

However, AI inventions related to autonomous driving present inventorship challenges under current patent systems.

For example, under the revised EPO Guidelines on AI and Machine Learning (G-III 3.3.1), examples of likely patentable AI inventions related to autonomous driving include AI-based navigation systems for the safer navigation of complex traffic situations and AI-assisted perception techniques for detecting pedestrians. For these inventions, human inventors arguably provide the real conception of invention (eg, defining parameters of complex traffic situations and how to detect pedestrians) AI may assist the human inventors in such an analysis, but the human governs the approach. Therefore, human inventors are expected to be granted inventorship.

Yet AI inventions go much further in the autonomous driving space. Many autonomous vehicle developers have focused their efforts on AI to provide or enable new or enhanced autonomous driving functionalities, such as using the AI self-learning process to build high-definition 3D maps from the images captured by LiDAR. The self-learning process means that AI itself could develop subject matter with limited or without human input.

Using DeepMind’s famous AlphaGo Zero as an example, the computer relies solely on reinforcement learning – without human data, guidance or domain knowledge beyond game rules – to develop unconventional strategies and creative moves to win games of Go. For autonomous driving inventions deploying such advanced AI techniques, would the computer powered by the AI algorithms be named as the inventor? Under the current patent systems, that seems unlikely.

Figure 3. AI patent filings related to autonomous driving

Figure 3. AI patent filings related to autonomous driving Source: WIPO

Figure 4. AI patent applications related to autonomous driving

Figure 4: AI patent applications related to autonomous drivingSource: IPlytics GmbH and IAM

One issue that may affect inventorship for AI-based autonomous driving inventions is the strong public interest in ensuring the safety of vehicles on the road. For example, Satinder Singh at the University of Michigan states that while AI is advanced in generating accurate 3D maps for autonomous vehicles, it is still “extremely limited in what it knows and in what it can do compared with humans”.

The need for humans to closely control how the autonomous vehicle will react to certain traffic or environmental conditions remains paramount. In other words, humans must decide the working parameters for vehicle control and reaction, while AI remains an important tool to assist them in deriving useful insights from input data (eg, gauging weather conditions, identifying road issues and sensing the behaviours of other drivers).

However, this raises questions as to whether the public interest aspect is best addressed by government regulation for transportation and how far AI-based inventions may go in learning novel approaches to autonomy – and who is named as an inventor for such discoveries.

A separate issue affecting inventorship in the autonomous vehicle space relates to who is running the AI engine at the time of any further improvement or act of invention. With the autonomous vehicle industry relying on collaborations and more complex supply chains with the integration of software and hardware from different sources, a new solution learned through AI may not result in a clearly identified inventor – human or otherwise. These issues must be addressed before the technology becomes more commonplace.

A healthy start

Healthcare is another prominent area for AI research and applications. According to a 2018 report by

CB Insights, healthcare AI start-ups have raised $4.4 billon across 576 deals since 2013, topping all other industries in AI deal activity (see Figure 5).

Figure 5. AI patents related to healthcare

Figure 5: AI patents related to healthcareSource: CB Insights

AI innovations are helping the healthcare industry across multiple fronts, including enhanced diagnosis, drug discovery and patient care, by improving the working methods of doctors and the accuracy and speed of diagnosis. According to the WIPO 2019 report on AI, between 2013 and 2016, patent filings in AI-related life and medical sciences grew by 12% with 4,112 filings in 2016, including medical informatics (18% growth) and public health (17% growth).

The healthcare industry can leverage AI to decrease costs while simultaneously improving efficiency and results. For example, AI can enable the use of data crunching to identify more cost-effective strategies in drug development (eg, by identifying patterns in large volumes of clinical data). This is highly desirable as, at present, nine out of 10 candidate therapies fail between phase I trials and regulatory approval, and the cost for developing each approved drug reaches about $2.6 billion, according to a May 2018 article in Nature.

One example of AI and machine learning being applied to drug discovery is Pfizer’s use of IBM Watson, which uses machine learning to power its search for immuno-oncology drugs. Moreover, a recent BenchSci Blog article identified 36 pharma companies using AI in drug discovery.

Machine learning can also “predict the physical and chemical properties of small molecules at quantum mechanics-level accuracy with much lower time-cost”, while AI can be used to “search for correlations between molecular representations and biological and toxicological activities” and “efficiently probe the pathways of synthesis of novel drug candidates” (Chan et al, “Advancing Drug Discovery via Artificial Intelligence” Trends in Pharmacological Sciences, 40(8):592-604 (2019)). As such, AI could completely transform drug discovery to drastically reduce the failure rate of drug candidates, decrease total development costs and help to identify novel drugs and treatment strategies.

Nevertheless, similar to the autonomous driving space, inventorship issues will likely arise with respect to AI inventions – in particular, with respect to drug discovery. Patents are particularly important to pharmaceutical companies as the regulatory review of a drug candidate typically extends well into the life of a patent covering the drug.

As drugs cannot be commercialised until they have received regulatory approval, the end of the patent term is often particularly important for a patent covering a commercialised drug product. What is more, pharmaceutical companies rely on patents to exclude competitors from marketing the same drug for the life of the patent in order to recoup the significant R&D costs involved.

This raises multiple questions, including whether inventorship of a drug developed using AI extends to the human who:

  • developed the AI;
  • identified the specific use of the AI resulting in the identification of the drug candidate; or
  • identified a data set combined with the AI resulting in the identification of the drug candidate.

Addressing the issue of inventorship at the time that a patent application is filed is critical.

Since US inventorship does not extend to companies, identifying the humans who contributed to the conception of the invention (eg, the drug candidate) is key, as a lack of proper inventorship can be a basis for challenging the patent. Thus, companies utilising AI in the healthcare space – and, in particular, in drug discovery – should fully explore the issue of inventorship when an AI-related patent application is filed.

Finally, while the obsolescence of specific AI tools may be an issue with respect to patent strategies in technologies such as autonomous vehicles, it is less of an issue with respect to drug discovery. Once a drug obtains regulatory approval, it is unlikely to be substantially modified, as this would lead to further regulatory review. Thus, a drug patent tends to be utilised for the full patent term and can reap significant value for the patent owner. This means that ensuring that the inventorship of a patent is correct is paramount.

The public has become accustomed to weekly news headlines about AI companies’ technical breakthroughs or massive funding hauls, and the frequency and level of new developments is only expected to increase. This rapid advancement in AI is driving the proliferation of patents in AI-related technologies.

At the same time, current IP systems are being critically challenged to promote the predictability and reliability of patenting AI inventions and to provide appropriate patent protection incentives. As such, effectively protecting the IP rights of AI inventions remains a tall order.

Action plan

As AI swiftly develops it is spawning numerous patents. Under the current patent systems around the world, AI is considered to be an aid or a tool to human inventors. Thus, inventorship is expected to be rewarded to the human inventors.

In view of the above, practical considerations include the following:

  • To secure inventorship to human inventors, it is helpful to specify alternative ways that human inventors can contribute to the conception of an AI invention (eg, defining the application(s) to be addressed by AI techniques, input parameters and data structures to be used by AI techniques on training data and expected results generated by utilising AI techniques). For the moment, be sure to list human beings as the inventors on your patent applications, at least for filings at the EPO and UKIPO.
  • Each AI invention must face the hurdles of Sections 101, 102 and 103 of the US Patent Act, regardless of whether AI was involved in its creation. Further, a practical concern for AI inventions is the relatively lengthy patent prosecution process from filing to issuance, during which some AI inventions become obsolete because of the rapid development cycle. Patent applicants may wish to sufficiently address inventorship issues to secure issuance reasonably quickly.
  • Document the analysis used to determine inventorship of an AI-related invention – in particular, for AI-related inventions in the healthcare space. Such documentation can be used to defend the inventorship determination if challenged at a later date (eg, in litigation).
  • Trade secret protection provides alternative IP protection where inventorship is not an issue. However, for trade secret protection to be actionable, it must be kept confidential, which can lead to cumbersome security measures (eg, implementing need-based protocol company-wide). Further, in a field subject to rapid technical development, keeping a valuable AI system secret can be challenging. Research suggests that for some systems, minimal targeted testing can reveal the AI’s underlying model. This could severely diminish the value of certain trade secrets.
  • When seeking copyright protection of creative arts produced by AI machines, it is important to be aware of the different requirements for human authorship in various jurisdictions.

The opinions expressed are those of the authors and do not necessarily reflect the view of the organisations with which they are affiliated. This article is for general information purposes and is not intended to be – and should not be – taken as legal advice.

Get unlimited access to all IAM content