The Technological and Policy Context
In recent years, a surge in generative AI (GenAI) techniques has sparked thought-provoking discussions within legal and policy circles regarding the interplay between AI technology and IP law. While often viewed as cutting-edge, GenAI is not an entirely novel technological phenomenon. For instance, generative adversarial networks (GANs), which triggered debates in copyright law in the highly publicized case of the painting Portrait of Edmond de Belamy, were introduced within the field of machine learning (ML) in 2014.
GenAI techniques generate output by leveraging deep learning architectures like generative pre-trained transformers and variational autoencoders, with common examples including diffusion models and large language models (LLMs). Compared to earlier generative models based on one particular type of data, current GenAI techniques employ multimodal approaches that transcend data types and can produce content involving different modalities, such as text, images, audio and video (for instance, a text-to-image system combines natural language processing and visual content generation). The emergence of advanced GenAI systems, particularly models like GPT-3, has prompted calls for a re-evaluation of existing IP frameworks.
These systems are often credited with the capability to autonomously generate output resembling works or inventions, raising questions about authorship, inventorship, the allocation of rights and protectability. Debates have centered on whether AI-generated content can or should qualify for IP protection and which actors within the AI value chain can or should be deemed lawful right holders. Furthermore, issues related to exceptions and limitations when IP-protected subject matter is used for training ML models, and the balance between protecting IP in existing creations and fostering future innovation, have been the focus of these discussions.
Currently, there are no pending legislative proposals to amend the EU or international IP framework with the aim of regulating the interactions between AI technology and IP law (except for a recent attempt to include a provision related to copyright law in the EU AI Act). Policy discussions are underway at the national level, and several judgments have been handed down by national courts addressing the peculiar questions posed by AI in IP law, with more cases yet to be decided. Several public consultations and studies have been commissioned or carried out by EU policymakers, national IP offices, and the World Intellectual Property Organization (WIPO).
The Institute’s Approach and Research Focus
The Institute’s early work on the interactions between AI and IP was undertaken within the research project “Regulation of the Data-Driven Economy”. From the outset, the focus has been on the capacity of ML techniques to generate content that might qualify for patent or copyright protection if it were created by humans without the use of ML. While the broader discourse on AI and IP has centered around AI-generated output, it has become apparent that proposals suggesting new forms of IP protection for output produced without sufficient human input lack credibility in the absence of innovation-based – let alone deontological – justification. Instead, the group has prioritized the input aspect, particularly the crucial role of access to critical inputs, such as input data, for the development of AI systems and applications. Enhanced access and use of data are prerequisites for both innovation within the AI field and innovation enabled through AI across various economic sectors. However, significant asymmetries persist in the supply and demand for training data, necessitating solutions that can effectively address these disparities.
Research Activity
Building on earlier work and exchange with IP scholars (as mentioned in the Institute’s Activity Report for 2018–2020), the position statement “Artificial Intelligence and Intellectual Property Law” of 9 April 2021 provides a systematic overview of AI and IP law issues arising throughout the AI innovation cycle. Focusing on substantive European IP law, the paper examines how existing categories of IP protection – copyright, patents, designs, databases and trade secrets – apply to input data, different components of the ML process, and outputs generated by ML models. The assessment maps out specific AI-related issues around the core questions of IP law, namely, the eligibility for protection under the respective IP regimes, the allocation of IP rights and the scope of protection. While the analysis mainly takes a de lege lata approach, it also identifies a research agenda that requires an in-depth investigation de lege ferenda, supported by interdisciplinary research. The position statement presents several propositions, among them: that introducing a new protection regime (such as a related right or sui generis form of protection) for AI-generated output would be unwarranted in the absence of a robust justification; that trade secrets protection might play a dubious role, potentially impeding access to data for AI system development; that the focus of the patent law debate should shift from inventorship to the concept of a skilled person and an inventive step; and that the system of copyright exceptions and limitations under EU copyright law should be re-evaluated, given the role of input data – often comprising IP-protected content – in developing ML applications.
Following this stocktaking exercise, key research questions have been identified at the intersection of AI technology and IP law that warrant in-depth exploration. Some of these questions continue to be explored in the dissertation and post-doc research projects supported by the Institute. For instance, the dissertation project “Legal Framework for AI-Based Work-Like Output” (Militsyna) seeks to develop an analytical approach that would make it possible to distinguish copyrightable from non-copyrightable output produced using AI applications. The dissertation project “Unlocking the Full Potential of AI – Towards Mandatory Data Access Rules for the Purpose of AI Development” (Chen) inquires into how the access-to-data regime should be designed to be more conducive to AI-enabled innovation. Allocation of value through the licensing of both IP-protected data and IP-unprotected AI-generated output is the focus of the dissertation “Shaping Europe’s Digital Future: Rethinking EU Copyright and Related Rights Remuneration Mechanisms for Outputs Generated by Artificial Intelligence Systems” (Dermawan, B II 2.XXX, p. XXX).
A notable development in patent law has emerged from the DABUS cases. Over the past few years, an international group of patent attorneys, operating within the framework of the “Artificial Inventor Project”, has filed patent applications worldwide for inventions purportedly created by the artificial neural network “Device for the Autonomous Bootstrapping of Unified Sentience” (DABUS). This project aims to demonstrate that the current patent system needs a profound update, in view of the AI’s alleged ability to invent autonomously. While the courts in most countries have established that only natural persons can be recognized as inventors, one court, namely, the Federal Court of Australia, ruled in the affirmative that DABUS can be recognized as an inventor in its own right (though this ruling was subsequently reversed on appeal). In its Position Statement on “Artificial Intelligence Systems as Inventors?” of 7 September 2021, the Institute criticized the decision for disregarding the lack of justification for attributing inventorship to an entity (a “technological artefact”) without legal capacity and failing to consider the legal consequences thereof. The statement emphasizes the need for a comprehensive analysis before such legal capacity can be acknowledged and highlights the broader relevance of these concerns to jurisdictions across the globe. The DABUS chronicle continues worldwide, so far with a clear tendency to uphold the principle that only a natural person can be acknowledged as an inventor. The article “The Paradox of the DABUS Judgment of the German Federal Patent Court” (Kim) further provides a detailed analysis of the decision of the German Federal Patent Court.
Recognizing the need to clarify factual and technical assumptions about the inventive capacity of generative AI techniques, Kim also collaborated with a group of ML researchers specializing in artificial neural networks and genetic algorithms, along with data scientists. The paper “Clarifying Assumptions About Artificial Intelligence Before Revolutionising Patent Law” provides a detailed analysis of instances frequently discussed in the literature on AI and patent law as examples of AI-generated inventions. It scrutinizes assumptions about ML systems inventing autonomously and identifies aspects within the application of ML techniques in technical problem-solving where human decision-making is decisive and directly influences the output. Overall, it contends that to address challenges in patent law the focus should be on defining the skilled person and assessing the inventive step requirement, rather than finding that there is no human inventor.
In the realm of copyright law, matters related to exceptions and limitations for text and data mining, particularly within the context of AI, have been the focus of Moscon’s work, as explored in “Data Access Rules, Copyright and Protection of Technological Protection Measures in the EU: A Wave of Propertisation of Information”. On the output side, Militsyna has proposed a test for assessing the sufficiency of human input in cases in which GenAI applications are employed (“Human Creative Contribution to AI-Based Output – One Just Can(’t) Get Enough”).
Furthermore, the edited collection titled Artificial Intelligence and Intellectual Property (Lee; Hilty; Liu (eds.), OUP 2021), based on the eponymous 2019 conference in Singapore, has been published, featuring contributions from Hilty, Hoffmann, Scheuerer, and Slowinski (see the Institute’s 2018–2020 Activity Report).
Outlook
Recent years have seen the challenge of predicting advancements in AI, including GenAI. Many technical aspects of the output production by GenAI, such as the propensity of artificial neural networks to “memorize” the input, require clarification before their implications for IP law can be scrutinized. While AI capabilities are often contrasted with those of humans, particularly in terms of the capacity to create or invent, it appears more pertinent to view GenAI applications as a complex interaction between humans and technologies. The challenge lies in defining just how much human involvement is required to warrant entitlement to authorship or inventorship.
While ongoing discussions about new forms of IP protection face challenges due to a lack of robust justification, a more pressing matter is addressing the disparities in the supply and demand for input data. It is also clear that AI innovation cuts across all legal domains and that, apart from IP law, liability and safety frameworks (in the EU, the draft AI Liability Directive for fault-based liability, the revised product liability framework and the new AI Act targeting AI safety challenges) are gaining particular prominence in shaping research and innovation activity in the field of AI. To examine the innovation implications of the interactions between IP and these frameworks would be very timely.
Publications
Drexl, Josef; Luc Desaunettes-Barbero; Jure Globocnik; Begoña González Otero; Reto M. Hilty; Jörg Hoffmann; Daria Kim; Shraddha Kulhari; Heiko Richter; Stefan Scheuerer; Peter R. Slowinski; Klaus Wiedemann, Artificial Intelligence and Intellectual Property Law – Position Statement of the Max Planck Institute for Innovation and Competition of 9 April 2021 on the Current Debate 2021 (Max Planck Institute for Innovation & Competition Research Paper, No. 21-10), 2021, 26 pages, External Link (DOI)
Drexl, Josef; Reto M. Hilty; Daria Kim; Peter R. Slowinski, Artificial Intelligence Systems as Inventors? A Position Statement of 7 September 2021 in View of the Evolving Case-Law Worldwide (Max Planck Institute for Innovation & Competition Research Paper, No. 21-20), 2021, 11 pages, External Link (DOI)
Kim, Daria, The Paradox of the DABUS Judgment of the German Federal Patent Court, GRUR International – Journal of European and International IP Law [GRUR Int] 71, 12 (2022), 1162–1166, External Link (DOI)
Kim, Daria; Maximilian Alber; Man Wai Kwok; Jelena Mitrović; Cristian Ramirez-Atencia; Jesús Alberto Rodríguez Pérez; Heiner Zille, Clarifying Assumptions About Artificial Intelligence Before Revolutionising Patent Law, GRUR International – Journal of European and International IP Law [GRUR Int] 71, 4 (2022), 295–321, External Link (DOI)
Moscon, Valentina, Data Access Rules, Copyright and Protection of Technological Protection Measures in the EU. A Wave of Propertisation of Information (Max Planck Institute for Innovation & Competition Research Paper, No. 23-14), 2023, External Link (DOI)
Militsyna, Kateryna, Human Creative Contribution to AI-Based Output – One Just Can(’t) Get Enough, GRUR International – Journal of European and International IP Law [GRUR Int] 72, 10 (2023), 939–949, External Link (DOI)