One of the most important questions in the field of AI is whether the traditional types of intellectual property (IP) protection are sufficient or even suitable to protect an AI and its byproducts.
This debate stems especially from the traditional view that IP protection has a human-centered nature. This humanistic view was echoed in Dentons’ AI survey: 58% believed that the user of the AI system should own the IP rights, while 20% believed the rights should go to the inventor of the AI system. Only 4% believed the AI system itself should hold the rights.
But does this view reflect the legal reality? To find the appropriate tools and strike the right balance, an option could be to mix and rearrange the underlying concepts of different IP rights as already known.
Protecting AI systems
Differences from ordinary computer programs and copyright protection
AI algorithms are designed in a way that allows them to infer the best approach from the data they are trained on; this differs from traditional computer programming where the programmer has prescribed the algorithm.
Broadly speaking, copyright only covers the elements expressing the creativity of the author. Copyright protection typically includes the source code, which constitutes an expression of the creativity of the author. However, the algorithm as such is typically not protected because it makes “autonomous” decisions and, thus, lacks sufficient authorship. Simply put, the underlying codes of an AI can be subject to copyright protection as a computer program. However, the copyright protection itself does not prevent someone from developing an AI system with the same algorithm but using a different source code. This may also create intellectual property protection challenges for peer-reviewed studies or regulatory applications, which generally require public availability and review of the underlying test data that contributed to the development of the product or study.
Since AI systems use data to create value, the selection or arrangement of data may also constitute an intellectual creation and be subject to IP protection. Some jurisdictions have a sui generis database right to protect the investment made in compiling a database. On the other hand, copyright protection is typically not extended to the data contained in a compilation itself, even if the compilations constitute copyrightable intellectual creations.
Certain inventions related to software programs are considered patentable in some jurisdictions, but not others. Although computer programs and mathematical methods are generally excluded from patentability, “computer-implemented inventions” providing technical effects are eligible for patent protection in the EU.
Since AI systems are usually not sufficiently protected by copyright or patent, one may consider protecting the AI algorithm as a trade secret. However, legal approaches to trade secrets vary by jurisdiction. The provisions on tortious acts, privacy, confidentiality or unfair competition are ordinarily applicable for trade secrets. This protection does not preclude the discovery by fair and honest means, e.g. independent creation or reverse engineering. It is also technically difficult and legally risky to prevent reverse engineering of software in AI, as contractual restrictions to achieve this may be considered null and void in some jurisdictions.
Due to its secret nature, trade secret protection contributes to the “black box” problem and provides a hurdle for data and technology sharing. Therefore, from both a policy maker’s and a right holder’s perspective, the best approach may be to combine trade secrets protection with copyright and patent protection, as is generally done with traditional software and related inventions.
Protecting AI creations
The outputs of AI technologies can be either AI-assisted or AI-generated. Since “AI-assisted” outputs can be achieved based on the creative input of a human being, it could be possible to deem the AI system as a tool—similar to an artist’s brush—in which case the current IP framework remains applicable.
“AI-generated” refers to the generation of an output by AI without human intervention. In this scenario, AI can change its behavior during the operation to respond to unanticipated information or events. However, the question remains as to how much human input is material.
Are AI-generated works copyright protectable?
As AI becomes increasingly sophisticated, it is now able to create outputs which are considered creative works. Recent examples include the Next Rembrandt’s painting, Robot Poet Deniz Yılmaz’s poetry, or Google’s Magenta—NSynth’s sounds. Broadly speaking, there is no form of legal protection for creative works made by AI under current law.
It comes as no surprise that the European Parliament stated in 2020 that, in apportioning IP rights, “the degree of human intervention” and “autonomy of AI” should be taken into account. Whereas “AI-assisted human creations” might meet the threshold of copyright protection depending on the individual case, “AI-generated creations” lack the required originality. To date, usually only the author’s own intellectual creations are regarded as eligible for protection. However, databases created with or by AI can be protectable by sui generis database rights to reward the investments made (also in developing the relevant AI systems) rather than the human creativity deployed.
Are AI-generated inventions patentable?
Similar issues and questions arise with respect to AI-generated inventions. In fact, the patent system is generally less centered on human creativity if compared to copyright and especially to the traditional European droit d’auteur doctrines. However, concepts like the moral rights of the inventor, inventive step and the person skilled in the art are also stretched by the peculiarities of autonomous computational inventions.
Who is the author or inventor of AI-generated subject matter?
In any case, the question of who the law would consider to be the person to be regarded as the author or inventor of AI-generated creations remains unsettled. Is it, for instance, the programmer of AI or the user of the AI program?
Adjusting IP protection in view of these developments must be a priority for lawmakers, in order to foster an environment conducive to creativity and innovation by rewarding creators—whoever this may be in the end.
Legislators all over the world have raised the issue of IP legislation regarding AI on several occasions.1 However, as at the publishing of this guide, no IP-specific AI legislation has been issued so far, even though many countries view the creation, protection and exploitation of IP in AI as a key measure toward improving competitiveness.
The Draft AI Act did not address IP-related questions regarding AI.
Many countries have started to discuss and imagine how to address IP in AI:
- The UK concluded a call for opinions on AI and IP in 2020. It covered patents, copyright and related rights, designs, trademarks and trade secrets as well as issues that cut across the IP rights.
- In Italy, the strategic guidelines on industrial property for 2021–2023, issued on June 23, 2021, expressly stated that “All opportunities shall be considered to deeply analyze the relationship between new technologies that are emerging, such as “artificial intelligence” (AI), and industrial property, in order to understand how to improve the effectiveness of protection systems and what forms of protection to provide for products created with the help of or directly from AI technologies”.2
- Japan is discussing how to improve the patent system to make it more appropriate for AI and technologies. One of the topics there is enforcement based on AI-related technologies.
- Ireland published its National AI Strategy in 2021, where it mentions that work is ongoing at the national level to examine the legal gaps for AI, including the legal regime on IP.
- China’s “New Generation Artificial Intelligence Development Plan” from 2017 proposes establishing AI technology standards and strengthening the protection of IP rights in the field of AI, including a proposal to set up an AI patent pool to promote utilization and diffusion of new AI.
- The United States’ National AI Initiative Act of 2020 provides for a coordinated federal program to accelerate AI research and applications for purposes of the economy and national security and will examine the issue of IP protection in the context of driving innovation.
What do you need to be aware of?
Due to the complex nature of AI that distinguishes it from conventional software, the types of IP protection envisaged for software are not sufficient to protect AI systems. Regarding the creations realized through AI, the human-centered nature of traditional IP protection types leaves the question of who will own AI-generated or AI-assisted creations unsettled.
According to Dentons’ AI survey, 86% of respondents believe that legislation is needed to clarify IP protection in the context of AI—45% of respondents indicated this was an urgent need.
Within the scope of national or international legislation, the legal aspects of AI are still at the stage of theoretical discussion. The legal rules and principles that will form the basis of the regulations have not yet emerged. In the event of a dispute, the courts may reach a different conclusion from the terms and conditions determined in the contractual relationship; and challenges and inconsistencies are likely to arise in terms of AI systems and/or their creations.
Companies should carry out an evaluation for each jurisdiction on how to provide for rights on AI as well as on creations realized through or by AI. They should then set up a plan to effectively protect the IP (and data) generated by their AI systems, mixing different IP-rights strategies with contractual protections.
This article is a chapter from Dentons’ Artificial Intelligence Guide 2022. Click here to access other chapters or download the full guide.
 E.g. World Intellectual Property Organization, “WIPO Conversation on Intellectual Property (IP) and Artificial Intelligence (AI)” WIPO/IP/AI/2/GE/20/1 REV (21 May 2020); European Parliament, “Draft Report on Intellectual Property Rights for the Development of Artificial Intelligence Technologies” 2020/2015(INI) (24 April 2020); European Parliament, “A Comprehensive European Industrial Policy on Artificial Intelligence and Robotics” 2018/2088(INI) (12 February 2019), paras 136-137; European Commission, “Artificial Intelligence for Europe” COM(2018) 237 (25 April 2018), p. 14.
 “Linee di intervento strategiche sulla proprietà industriale per il triennio 2021-2023”, available at: https://uibm.mise.gov.it/images/LINEE_DI_INTERVENTO_approvate.pdf