Artificial Intelligence (AI) is the ability of a computer or computer-controlled products to perform tasks that only intelligent beings are able to do. The most common use for this term is to categorize projects, systems, or products that are able to develop intellectual processes normally performed by humans, such as reasoning, analyzing, generalizing, learning from previous experiences, etc. Business analysts expect that the AI industry invade business processes around the world, estimating that the AI market will grow at an annual rate of 20%. In the past four years, there has been a 270% increase in organizations that have recently implemented AI products. Additionally, analysts estimate that 80% of emerging technologies will have an AI component.
The recent disruption of this sector has also reached the intellectual property sector. According to the World Intellectual Property Organization (WIPO), the AI technique showed an average growth rate of 28% between 2013 and 2016. From 1956, which was the year that AI was invented, to 2017, more than 1.6 million scientific articles related to AI have been published. In this same period, approximately 340,000 patent applications of AI-related inventions have been presented. In 2011, 12,473 AI patent applications were filed; at the end of 2017, the WIPO registered 55,660 AI patent applications, a patents’ increase of approximately 300% in this period.
This new and increasing technology trend has posed several questions to the actual intellectual property law. One of the ongoing debates in the intellectual property sector is the ambiguity some AI inventions have in terms of its inventorship. In the majority of the cases, AI is a technology that either helps the inventor create the product or forms part of it. In this sense, AI inventions are not necessarily different from other inventions assisted by computers, such as customer relationship management software’s. However, it seems obvious that given the nature of AI, this technology is able to create inventions independently. According to the WIPO, there have already been several patent applications where the applicant has declared the AI technology per se as the inventor.
This AI topic generates several questions that need to be answered by the intellectual property authorities. For instance, should the intellectual property law allow that the AI technology is declared as the inventor of a patent application? Or should it mandate that only human beings could be inventors to avoid ambiguity in the invention’s ownership? If the IP laws decide that the latter is the most appropriate regulation, should the law regulate how human inventions will be determined? Or should the inventorship determination be left to private agreements in the corporate sector? Each of these questions poses several difficulties for the actual intellectual property law application.
Another problem that the WIPO highlights in its documents is the patentability guidelines that may or may not be modified to regulate AI-based inventions. Computer-based inventions have been subject to many discussions all around the world due to its intellectual property regulation complexity. AI-related inventions have not been excluded from these debates. In particular, the WIPO has posed these unsolved questions in its most recent document that expands on AI intellectual property regulations: Should inventions that are generated autonomously by AI should be excluded from patent eligibility? Should AI inventions be regulated by existing regulations of computer-assisted inventions? and given the recent invasion of AI inventions, should the intellectual property authorities add specific guidelines for the regulation of AI inventions? Regarding these questions, the WIPO has implied in its documents that there is no right answer for these issues. For example, if experts consider that we need to add new guidelines, which parts of patent examination guidelines will the modifications cover and why?
The third issue that WIPO discusses is related to the one with nanotechnology, where the non-obviousness evaluation of the invention is ambiguous due to the nature of the technology. We know and have previously discussed that non-obviousness evaluations are performed by people skilled in the field of the “art” to which the invention belongs. In the AI case, we need to ask in which field does AI fall, and whether intellectual property firms will need to hire a skilled person in AI or a skilled person in the field of the application of the AI technology (e.g., if the application is AI data optimization, will we need to consult a scientist expert in data or a scientist expert in AI computers?).
Moreover, the issue of non-obviousness generates an incredibly complex degree of difficulty for its evaluation when we consider AI inventions autonomously done by AI technologies. The questions “which is the prior art we need to evaluate?” and “which is the expert we need to call?” become even more difficult to answer. To solve this issue, we need to first reflect: who is the expert on the field of a non-human created invention? This reflection leads to an ongoing debate of whether the expert should be a human being or a trained algorithm with data. This nuance can cause major changes to the actual way we evaluate non-obviousness in new AI inventions.
Another major conflict that comes with this conflict of replacing humans skilled in the fields of AI inventions with machines and algorithms trained with data is the implications this change would have for the intellectual property application in previous inventions. To think about this issue, we need to ask ourselves: Would existing patents on the field need to be revised using non-obviousness evaluations of these machines and algorithms? And if machines and algorithms are used to evaluate this intellectual property aspect, would they be more permissible or more rigorous for determining whether the invention is non-obvious or not?
AI inventions could also conflict with the disclosure aspect of patents. According to the WIPO, one of the main goals of the patent system is to disclose technology so that the public domain is enriched constantly to achieve that a systematic record of human inventions is accessible and useful to everyone. The disclosure goal enables scientists and entrepreneurs to continue innovating using existing technologies, while respecting their uniqueness. Regarding AI inventions, disclosure of inventions that were created by AI systems could be complicated.
One of the major components of AI technologies is machine learning, where an algorithm is trained with previously generated data. With this data, the algorithm is capable of learning skills and parameters that will be applied to new data, so the trained algorithm could perform its functions (e.g., classifying things, reasoning through a problem, etc.). The performance of the algorithm will change depending on the training data you input. Hence, the algorithm constantly changes. The issue that this changing status of the algorithm poses to the disclosure condition is the following: Will the disclosure of the initial algorithm be sufficient for the full disclosure of the invention? Or will the WIPO need to establish a mechanism so that the inventor provides constant updates of the algorithm considering that it changes over time?
Given the importance of this issue, in 2019, the WIPO even discussed if the creation of an algorithms’ deposit, analogous to a microorganisms’ deposit, would be appropriate for achieving full disclosure of these new technologies. A decade ago, proposing the creation of an algorithms’ deposit would have been seen as utopic. Now, considering the recent irruption in the field and the amount of inventions that have been generated, the algorithms’ deposit seems realistic and plausible.
The data used for training the algorithm may also pose several issues to the disclosure condition. The training data shapes the algorithm and its performance since the algorithm will learn to perform tasks using this type of data. Hence, the WIPO encourages discussions of the following questions: How should training data be treated for disclosure? Should it be fully disclosed? Or should it only be described in the patent application? If the latter is most appropriate, would not disclosing the training affect the total disclosure condition of the invention?
Advances in computational force are accelerating the AI revolution affecting multinational and local organizations all over the world. Eventually, almost every organization could profit and benefit from the utilization of AI. The effect of AI is already found in applications that individuals use daily, such as transport, health, finance, law and other areas. Similarly, as with each new innovation, AI offers favorable circumstances to early adopters, but it also poses numerous difficulties. As intellectual property (IP) collaborators, we have the responsibility of anticipating difficulties that may arise with AI-related patent applications.
There are many more issues to be discussed regarding patent aspects and eligibility of AI inventions. The purpose of this article is not to solve these questions and topics that the WIPO has communicated previously. Conversely, the article’s purpose is to inform intellectual property agents of these ongoing debates. As the WIPO has stated, there is no wrong answer to these issues, and the solutions are far from being obvious. Therefore, by discussing these issues, we will deepen and challenge our actual knowledge of the intellectual property law, while we prepare ourselves for the changes AI inventions could cause to the current IP laws.