Analysis
From autonomous vehicles to automated healthcare, Artificial intelligence (AI) is a fast-evolving field driving major developments in technology and business and is set to revolutionise many industries. AI is optimizing business operations, driving down operational costs, and delivering enhanced customer experiences. AI has even made its way into our homes through personal devices like Alexa and Siri. The availability of large amounts of training data and the advances in affordable high computing power are driving the growth of AI. According to the international market research firm International Data Cooperation ‘IDC,’ over US$ 37.5 billion was spent on AI in 2019 alone.
Spending on AI systems is estimated to reach US$ 374 million in the MENA region in 2020 led by the banking and retail industries, with 20% of this attributed to the UAE alone.[1 p.5] Federal governments in the region are making significant investments towards digitising government transactions and building smart cities. As AI solutions advance, so too are the issues of patentability and inventorship of inventions made using AI.
So, what is AI?
In simple terms, AI is a general concept used to describe the use of computers and technology to simulate intelligent human behaviour. AI makes it possible for computer systems to learn from experience, adjust to new inputs and perform human-like tasks. Using AI technologies like machine learning, computer vision and natural language processing, AI models or systems can be trained to accomplish specific tasks by processing large amounts of data and recognising patterns in the data. AI has demonstrated that it can make intelligent decisions that have traditionally been accomplished by humans and can potentially do so even faster and more efficiently than we can.
Machine Learning (ML) is a subset of AI where the AI model requires structured or labelled data to produce consistent results.ML applications already steer many technologies that we use on a daily basis such as email spam filtering, predictive texting and machine translation. Deep Learning is an area within Machine Learning where data labelling is not required and a high volume of raw unstructured data can be fed to the AI model to produce results. Deep learning requires high computational power and according to a report published by WIPO[2 p.5], is largely responsible for the growth in AI-related patent applications filed since 2013.
In medicine, AI is already being utilised for online scheduling of appointments, online check-ins at medical facilities, digitisation of medical records and other administrative tasks. In addition, AI has been found to be successful in performing medical functions such as diagnosis and prognosis of disease, treatment and personalised medicine[3 p.5]. AI techniques such as artificial neural networks and fuzzy logic are being used for imaging analysis (radiographs, ultrasounds, CT scans and MRIs) and proven to be reliable in diagnosing skin cancer and prostate cancer.[4 p.5] The advantages of AI in boosting medical care are clear, increased efficiency and accuracy in diagnosis means decreased workload for clinicians and increased patient facetime and more personalised care. As the technology advances in medicine and other fields, questions over the patentability of AI-related inventions are rising.
Patenting AI
According to WIPO, over 340,000 patent families exist for AI-related inventions. The growth trend is exponential, with over half of these patents filed since 2013. The top three AI patent filers are IBM, Microsoft and Google. IBM is leading the way with a portfolio of over 8,000 AI-related patent applications and patents, largely focused on the Watson platform.[5 p.5] While AIrelated patents are on the rise, many applications have encountered rejections. The most common objections centre around patent subject matter eligibility, followed by issues of novelty, obviousness and inventorship.
Subject matter eligibility is one of the core criteria for obtaining a patent, in addition to novelty and inventiveness (or nonobviousness in the US). An invention must contain patent-eligible subject matter in order to obtain patent protection. The global position on excluded subject matter is somewhat harmonised as it relates to AI. Abstract ideas, business methods, mathematical models and computers programs are generally excluded from patentable subject matter. However, the approach to dealing with these exclusions varies between the US and Europe, with eligibility being more of an issue for AI patent applications in the US than Europe.
The US approach to AI inventions
Patentable subject matter is described in the US patent law under 35 U.S.C. 101[7 p.5] as “any new and useful process, machine, manufacture, or a composition of matter, or any new and useful improvement thereof.” Abstract ideas, laws of nature and natural phenomena are excluded subject matter. AI algorithms are regarded as ‘computational’ and ‘abstract in nature’ and therefore not patentable. Little guidance exists on how to determine whether the claims in an application are directed towards an ‘abstract’ idea but the mere use of machines to implement an abstract idea is not sufficient to pass the eligibility hurdle in the US.
AI patent applications are examined by the U.S. Patent & Trademark Office (USPTO) in the same way as computer-implemented inventions. For computer-implemented inventions, two Supreme Court cases, Mayo[8 p.5] and Alice[9 p.5], laid down the framework on subject matter eligibility. In Mayo, the Supreme Court found that generic computer functionalities that are “well understood, routine, conventional activities known to industry” cannot confer subject matter eligibility to an
invention. In Alice, the court set out a two-step test that must be met by any claimed invention:
- determine whether the claims are directed towards a patent-ineligible concept (abstract ideas, laws of nature or natural
phenomena); - determine whether the claim’s elements, considered both individually and as an ordered combination, transform the
nature of the claims into patent-eligible application.
These two Supreme Court cases present a hurdle that AI inventions must overcome to receive patent protection. Many AI patents are directed to the AI algorithms and machines used to generate those algorithms. Despite this, over 11,000 patent applications for AI-related inventions have been published in the US since 2016. Applications are more likely to get around the US exclusions if the claims are drafted towards solving a specific, new and non-obvious technical problem achieved by the AI algorithm.
The European approach to AI inventions
In Europe, the European Patent Organisation (EPO) has issued guidelines for patenting AI inventions.[10 p.5] Although little caselaw exists specifically on AI inventions, the EPO, like the US, is applying the same principles in its examination of AI inventions as computer-implemented inventions. The two-hurdle approach set out by the EPO requires inventions to:
- have technical character; and
- provide a technical contribution or a technical effect.[11 p.5]
Article 52(2) of the European Patent Convention[12 p.5] sets out the exclusions for patentability in Europe and includes mathematical methods, computer programs and business methods. The guidelines would suggest that the first hurdle, ‘the low hurdle’, can be overcome by simply drafting the claims to include ‘computer-implemented method’ or ‘processing hardware’ to demonstrate technical character. Terms such as ‘artificial neural network’ or ‘support vector machine’ are considered nontechnical per se.
The second hurdle, ‘the high hurdle’, requires inventions to solve a technical problem. Most AI inventions are known as ‘mixed-type inventions’ which involve some technical aspects and some non-technical aspects. Features that do not contribute to the technical character of an invention cannot support an inventive step. This second hurdle is known as the COMVIK approach.[13 p.5] So, if a given claim feature contributes only to the solution of a non-technical problem, for example classifying data records, without an indication of a technical use being made of the data then the invention does not solve a technical problem. The application must demonstrate that the claimed subject matter serves a technical purpose and, at a minimum, provides a technical effect that is more than simply achieving the solution more quickly or efficiently. The generic application of AI to solve a problem in a foreseeable way is not patentable.
The EPO has taken some additional steps to specify what are known as safe harbours for AI inventions. Claims that are directed towards specific technical systems or processes, for example an X-ray controlling an apparatus to produce a technical effect, are more likely to get around the exclusions. Additionally, processing of digital audio images or video enhancement or analysis are also considered to be outside the exclusions. Natural language processing techniques and systems for automated medical diagnosis have also been granted patents. There is also room for innovation in the improvement of AI systems. For example, improvements could arise in the way the training data is collected or mapped to the AI model, the way the AI model is trained or how the AI output is processed or interpreted.
Although the EPO provides more formal guidance on patenting AI inventions, both the EPO and US approaches for determining subject matter eligibility still involve a degree of legal uncertainty when it comes to the patentability of AI inventions. Provided the AI invention solves a specific technical problem in a new and non-obvious way, the likelihood of success for obtaining a patent will rely heavily on skilful patent drafting.
Other considerations
Aside from subject matter eligibility, uncertainties also exist in relation to patent inventorship, ownership and infringement liability. Generally patent systems only recognise individuals as inventors, not companies, or machines. The use of certain AI technology, particularly deep machine learning or self-evolving AI, raises questions as to who or what conceived the invention and should be named as the inventor. If the AI system creates something better than it was programmed to do, who owns this? And if this new solution infringes on someone’ s patent, who is liable?
Recently the issue of inventorship received attention with a USPTO decision regarding an AI system called ‘DABUS’.[14 p.5] The USPTO ruled that AI systems cannot be credited as an inventor in a patent.[15 p.5] Referring to the US patent act, the USPTO concluded that only ‘individuals’ and ‘natural persons’ may be named as an inventor in a patent application. This followed similar rulings from the UK IP office and the EPO. Unless the patent laws change in the future, AI is likely to remain an inventing tool rather than an inventor. The issues of ownership and liability are yet to be tested.
AI initiatives in the UAE
In the global race to embrace AI, the UAE has positioned itself as an early adopter in the public sector as well as the private. In April 2019, the UAE Cabinet approved a 10-year strategic plan – the National Strategy for Artificial Intelligence – intended to increase investments in AI adoption, boost government performance and establish the UAE as a global hub and test bed for AI technologies and legislation.[16 p.5] The strategy covers the development and application of AI in nine sectors including energy, health, renewable energy, space, transport and water. In the UAE today, AI technologies are actively deployed in the banking, health, energy, aviation and transportation sectors. According to the National Program for Artificial Intelligence, AI imaging analysis is being used by the Ministry of Health and Prevention to detect communicable diseases like Tuberculosis in new UAE residents. Emirates airline, Dubai’ s flagship carrier, has plans to implement AI solutions across the aviation experience from managing flight logistics to baggage handling services to enhance customer experience.[17 p.5]
The availability of data is essential to the application of AI solutions however effective data regulation is key in building trust in AI systems. The deployment of AI requires the development of policies and frameworks around the governance of AI, data privacy, and cybersecurity. Data integrity is also important for AI systems to be reliable. Because AI systems learn to make decisions based on historical data, they are prone to amplifying existing biases which can have potentially life-altering consequences. To manage bias and ethics as well as other issues around the use and standardisation of data, a specialised Data Committee has been set up by the UAE Artificial Intelligence and Blockchain Council. Issues of cybersecurity will also require increased governance as more AI-based transactions will mean more vulnerability to cyberattacks.
In addition to deploying AI, the UAE is also focusing on developing its own AI solutions. Abu Dhabi National Oil Company ‘ADNOC’ recently signed an agreement with Group 42, an Abu Dhabi-based AI company, to establish a joint venture to develop AI products for the energy sector.[18 p.5] The venture will leverage Group 42’s central processing and graphics processing units and its technical know-how and have access to ADNOC’ s vast data archives to develop new solutions. As part of its
commitment towards driving AI innovation in the oil and gas sector, ADNOC established two innovation centres in 2017, Thomama and Panorama which will explore solutions to drive down costs and increase revenues across the oil production value chain. In 2019, the UAE established the world’s first dedicated research-based artificial intelligence (AI) university, Mohamed bin Zayed University of Artificial Intelligence (MBZUAI). As well as building capabilities in AI, the university’s
programs will focus on the research and development of AI innovations.
As the UAE develops its own AI innovations, we expect to see a rise in patent applications for AI inventions in the UAE. Article 6 of the UAE patent law excludes ‘scientific and mathematical principles, discoveries and methods’ from patentability in addition to ‘guides, rules or methods followed to conduct business’.[19 p.5] Little formal guidance exists in the law in relation to the patenting of computer-implemented inventions, however algorithms and software code are considered non-patentable. Software developers currently rely on copyright to protect their source code. While copyright prevents others from copying the code or a substantial part of it, it does not protect ideas or the functional aspects of the software program. For example, copyright would only protect the source code written to implement an algorithm, not the algorithm or method that designates the strategy employed by an AI model. Copyright would not prevent someone from designing a new AI model with the same algorithm (and features) using a different source code to implement it. The functional features of an AI model and user
interface are not protected by copyright alone.
In its examination of AI inventions, we expect the UAE patent office’ s decisions to rely heavily on the opinions published by international examiners and follow the same approach used for computer-implemented inventions. This is likely to continue as we see patent laws evolve to deal with more complex issues of patentability and inventorship in other jurisdictions. For now, it is clear that IP protection strategies for AI inventions, like computer-implemented inventions, will require a combination of patents, copyright and trade secrets.
Written by Tamara El-Shibib.